COOPERATIVE SYSTEMS DESIGN
Frontiers in Artificial Intelligence and Applications Series Editors: J. Breuker, R. Lopez de Mantaras, M. Mohammadian, S. Ohsuga and W. Swartout
Volume 74 Previously published in this series: Vol. 73, In production Vol. 72, A. Namatame et al. (Eds.), Agent-Based Approaches in Economic and Social Complex Systems Vol. 71, J.M. Abe and J.I. da Silva Filho (Eds.), Logic, Artificial Intelligence and Robotics Vol. 70, B. Verheij et al. (Eds.), Legal Knowledge and Information Systems Vol. 69. N. Baba et al. (Eds.). Knowledge-Based Intelligent Information Engineering Systems & Allied Technologies Vol. 68, J.D. Moore et al. (Eds.). Artificial Intelligence in Education Vol. 67. H. Jaakkola et al. (Eds.), Information Modelling and Knowledge Bases XII Vol. 66. H.H. Lund et al. (Eds.), Seventh Scandinavian Conference on Artificial Intelligence Vol. 65, In production Vol. 64. J. Breuker et al. (Eds.), Legal Knowledge and Information Systems Vol. 63, I. Gent et al. (Eds.), SAT2000 Vol. 62, T. Hruska and M. Hashimoto (Eds.), Knowledge-Based Software Engineering Vol. 61, E. Kawaguchi et al. (Eds.), Information Modelling and Knowledge Bases XI Vol. 60, P. Hoffman and D. Lemke (Eds.), Teaching and Learning in a Network World Vol. 59. M. Mohammadian (Ed.), Advances in Intelligent Systems. Theory and Applications Vol. 58. R. Dieng et al. (Eds.). Designing Cooperative Systems Vol. 57, M. Mohammadian (Ed.), New Frontiers in Computational Intelligence and its Applications Vol. 56, M.I. Torres and A. Sanfeliu (Eds.), Pattern Recognition and Applications Vol. 55. G. Cumming et al. (Eds.), Advanced Research in Computers and Communications in Education Vol. 54. W. Horn (Ed.), ECAI 2000 Vol. 53. E. Motta, Reusable Components for Knowledge Modelling Vol. 52. In production Vol. 51, H. Jaakkola et al. (Eds.), Information Modelling and Knowledge Bases X Vol. 50, S.P. Lajoie and M. Vivet (Eds.), Artificial Intelligence in Education Vol. 49. P. McNamara and H. Prakken (Eds.), Norms, Logics and Information Systems Vol. 48. P. Navrat and H. Ueno (Eds.), Knowledge-Based Software Engineering Vol. 47. M.T. Escrig and F. Toledo, Qualitative Spatial Reasoning: Theory and Practice Vol. 46. N. Guarino (Ed.), Formal Ontology in Information Systems Vol. 45. P.-J. Charrel et al. (Eds.), Information Modelling and Knowledge Bases IX Vol. 44. K. de Koning, Model-Based Reasoning about Learner Behaviour Vol. 43. M. Gams et al. (Eds.), Mind Versus Computer Vol. 41, F.C. Morabito (Ed.), Advances in Intelligent Systems Vol. 40. G. Grahne (Ed.), Sixth Scandinavian Conference on Artificial Intelligence Vol. 39. B. du Boulay and R. Mizoguchi (Eds.), Artificial Intelligence in Education Vol. 38. H. Kangassalo et al. (Eds.), Information Modelling and Knowledge Bases VIII
ISSN: 0922-6389
Cooperative Systems Design A Challenge of the Mobility Age Edited by
Mireille Blay-Fornarino Nice University, Nice, France
Anne Marie Pinna-Dery Nice University, Nice, France
Kjeld Schmidt IT-University of Copenhagen, Copenhagen, Denmark
Pascale Zarate Institut National Polytechnique de Toulouse, IRIT, Toulouse, France
IOS Press
Ohmsha
Amsterdam • Berlin • Oxford • Tokyo • Washington, DC
© 2002, The authors mentioned in the Table of Contents All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior written permission from the publisher. ISBN 1 58603 244 5 (IOS Press) ISBN 4 274 90503 9 C3055 (Ohmsha)
Publisher IOS Press Nieuwe Hemweg 6B 1013 BG Amsterdam The Netherlands fax: +31 20 620 3419 e-mail:
[email protected]
Distributor in the UK and Ireland IOS Press/Lavis Marketing 73 Lime Walk Headington Oxford OX3 7AD England fax:+44 1865 75 0079
Distributor in the USA and Canada IOS Press, Inc. 5795-G Burke Centre Parkway Burke,VA 22015 USA fax: +l 703 323 3668 e-mail:
[email protected]
Distributor in Germany, Austria and Switzerland IOS Press/LSL.de Gerichtsweg 28 D-04103 Leipzig Germany fax: +49 341 995 4255
Distributor in Japan Ohmsha, Ltd. 3-1 Kanda Nishiki-cho Chiyoda-ku, Tokyo 101-8460 Japan fax: +81 3 3233 2426
LEGAL NOTICE The publisher is not responsible for the use which might be made of the following information. PRINTED IN THE NETHERLANDS
Preface The main goal of the COOP conferences is to contribute to the solution of problems related to the design of cooperative systems, and to the integration of these systems in organisational settings. The conferences are aimed at: • the advancement of the comprehension of human-human and human-machine cooperative work processes; • the development of models of co-operation and cooperative work from different perspectives; • the development of appropriate design methodologies and of new functions for cooperative systems with a view to supporting cooperative work. The main assumption behind the COOP conferences is that cooperative system design requires a deep understanding of cooperative work in groups and organisations, involving both artifacts and social practices. The previous COOP conferences (1995, 1996, 1998, 2000) attracted theoretical as well as empirical contributions on such topics as analysis of complex work situations, conceptual frameworks for understanding cooperative work, guidelines for modelling collective activities, methodological aspects of cooperative systems design and innovative and technological solutions. The COOP 2002 conference is mainly devoted to the following issues: • The gap between 'virtual' and 'material' artifacts in human collaboration: how to bridge it, conceptually as well as technologically. • Collaboration among mobile actors: the coordinative challenges of mobile interaction, coordination practices of mobile actors, novel technologies to support mobile interaction. • The WWW as a platform for cooperative systems: how to overcome the limitations of current web technologies. • Changing work practices and organisations in the wake of cooperative systems: realworld exploitation and impact of cooperation technologies (web-based communities, organisational uses of groupware, intranet etc.) The design of cooperative systems offers several challenges. Through contributions presented in this book, we could see that the WWW and mobility issues imply a multidimensional context for cooperative system design and that Knowledge Management is an important issue in such a context. Many traditional approaches break down under constraints of cooperative system design. Several contributions in this book offer solutions presenting new tools and techniques. Some contributions try to answer the question: which kind of artifacts for cooperation? Finally, very interesting case studies present a valuable contribution to this book.
Acknowledgements Many persons and institutions have made the publication of this volume possible. We would like to thank them for their valuable effort. We would especially like to thank the members of the conference committee for their collaboration; the members of the program committee and the additional reviewers for helping the authors in improving the quality of the papers. We would also like to express our gratitude to the organisation committee and to the sponsors of the COOP'2002 conference. Last but not least, we would like to thank Rose Dieng, Alain Giboin and Monique Simonetti of the organising committee for their devotion to the COOP conferences. Toulouse, 1 February 2002 The editors Mireille Blay-Fornarino, Anne Marie Pinna-Dery, Kjeld Schmidt, and Pascale Zarate
Conference Committee Conference co-chairs Mireille Blay-Fornarino, France Anne-Marie Pinna-Dery, France Program co-chairs Kjeld Schmidt, Denmark Pascale Zarate, France Local organisation Rose Dieng, France Mireille Blay-Fornarino, France Alain Giboin, France Corinne Jullien, France Anne-Marie Pinna-Dery, France Monique Simonetti, France Proceedings co-chairs Mireille Blay-Fornarino, France Anne-Marie Pinna-Dery, France Workshops chair Regine Teulier-Bourgine, France Posters chair Manuel Zacklad, France Doctoral colloquium co-chairs Camille Rosenthal-Sabroux, France Peter H. Carstensen, Denmark Fabien Gandon, France Sponsors liaison Rose Dieng, France Michel Riveill, France
Program Committee Co-chairs Kjeld Schmidt Pascale Zarate Members Mark S. Ackerman, USA Sebastiano Bagnara, Italy Liam Bannon, Ireland Michel Beaudouin-Lafon, France Richard Bentley, UK Alex Borgida, USA Susan E. Brennan, USA Peter Carstensen, Denmark Francoise Darses, France Francoise Decortis, Belgium Chrysantos Dellarocas, USA Sylvie Despres, France Giorgio De Michelis, Italy Rose Dieng, France Paul Dourish, USA Ken Eason, UK Skip Ellis, USA Pierre Falzon, France Gerhard Fischer, USA Geraldine Fitzpatrick. UK Alain Giboin, France Marie Pierre Gleizes, France Saul Greenberg, USA Jonathan Grudin. USA Cristine Halverson. USA
Christian C. Heath. UK Jean-Michel Hoc, France Erik Hollnagel, Norway Laurent Karsenty, France John King, USA Kari Kuutti, Finland John C. McCarthy, Ireland Gloria Mark, USA Patrick Millot, France Bernard Pavard, France Wolfgang Prinz, Germany Dave Randall, UK Thomas Rodden, UK Yvonne Rogers, UK Camille Rosenthal-Sabroux, France David Sadek, France Carla Simone, Italy Jean-Luc Soubie, France Lucy Suchman, UK Regine Teulier-Bourgine, France Jonathan Trevor, USA Yvonne Waern, Sweden David Woods, USA Volker Wulf, Germany Manuel Zacklad, France
Additional reviewers Tony Marchand, France Simon Kaplan, Australia Eswaran Subrahmania. USA
Sponsors AFIA, club CRIN ITI, Region Provence-Alpes-Cote d'Azur, GDR-I3, I3S, INRIA
Contents Preface Conference Committee Program Committee Additional Reviewers and Sponsors
v vii viii viii
Invited Speaker Mobility and Cooperative Systems: Failures, Successes and Possibilities, Richard Harper
3
Cooperation for Multi-Dimension Context ThreeDness: Representing Awareness in Cooperative Applications, Fabrizio Nunnari and Carla Simone
1
Reconciling Different Perspectives: An Experiment on Technology Support for Articulation, Gloria Mark, Victor Gonzalez, Marcello Sarini and Carla Simone
23
Growing Networks: Detours, Stunts and Spillovers, Margunn Aanestad and Ole Hanseth
38
Tools and Techniques for Cooperative Systems Design Modelling Cooperative Work: Chances and Risks of Structuring, Thomas Herrmann, Marcel Hoffmann, Gabriele Kunau and Kai-Uwe Loser
53
Understanding the Benefits of Graspable Interfaces for Cooperative Use, Eva Hornecker
71
Defining Task Interdependencies and Coordination Mechanisms for Collaborative Systems, Alberto B. Raposo and Hugo Fuks
88
Cooperative Engineering Engineering CSCW, Tony Lambie and John Long
107
Let's Work Together: Supporting Two-Party Collaborations with New Forms of Shared Interactive Representations, Mike Scaife, John Halloran and Yvonne Rogers
123
Studies of Computer Supported Collaborative Writing. Implications for System Design. Teresa Cerratto and Henrry Rodriguez
139
Cooperative Case Studies A Common Work Space to Support the Cooperation in the Cockpit of a Two-Seater Fighter Aircraft, Marie-Pierre Pacaux-Lemoine and Anthony Loiselet
\ 57
AMANDA Project: Delegation of Tasks in the Air-Traffic Control Domain. Serge Debernard, Stephane Cathelain, Igor Crevits and Thierry Poulain
173
Recursive Articulation Work in Ariadne: The Alignment of Meanings. Marcello Sarini and Carla Simone
191
Cooperative Knowledge Management and Context Cooperative Organizational Memories for IT-Based Process Knowledge Management, August-Wilhelm Scheer, Frank Habermann, Oliver Thomas and Christian Seel
209
Towards a "Knowledge-Based Marketplace" Model (KBM) for Cooperation between Agents, Jean-Pierre Cahier and Manuel Zacklad
226
Putting CSCW in a Business Context - Implications for Research and Systems Design. Per-Arne Persson
239
Which Kind of Artifacts for Cooperation? Coordinative Artifacts in Architectural Practice. Kjeld Schmidt and Ina Wagner
257
Co-Constructing Collaborative Computer-Based Artifacts and Mentefacts. Peter Mambrey and Bettina Torpel
275
Panel Abstracts What's New in Knowledge Management? Liam J. Bannon
289
Understanding the Roles of Artifacts in Cooperative Work. Kjeld Schmidt
292
Author Index
293
Invited Speaker
This page intentionally left blank
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press, 2002
Mobility and Cooperative Systems: failures, successes and possibilities Richard Harper, Digital World Research Centre (DWRC), University of Surrey, England
Mobile networks have created what some have called the second stage in the digital revolution, the first being driven by the emergence of the world wide web. But just what has been the impact of this so-called revolution? Have mobile technologies facilitated and shaped radical transformation in organisational space? Have mobile devices afforded new opportunities for remote and distributed work? And what will be enabled by so-called 3rd generation networks? In this presentation, materials from a number of ongoing projects at the DWRC will be reported that indicate that the mobile age is not what it might seem. Starting with some comparisons between the expected and actual impact of fixed line telephony and the expected and actual impact of mobile devices, the presentation will then explore why it is that mobile communications devices are, at the current time at least, displacing data supporting devices like PDAs and laptops in the world of mobile professionals. It will then examine how the use of 'anytime anywhere' technologies is leading mobile professionals to create new forms of 'communications management' processes. These combine social and technological tools. The presentation will then consider what implications the emergence of these tools will have for 3rd generation networks, devices and services, before concluding with reflections on the relationship between future mobile communications technologies and other informational artefacts in the office and organisational spaces of the future.
This page intentionally left blank
Cooperation for Multi-Dimension Context
This page intentionally left blank
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press, 2002
ThreeDness: representing awareness in cooperative applications Fabrizio NUNNARI Department of Informatics University of Torino, Corso Svizzera 185, 10148 Torino, Italy
Carla SIMONE Department of Informatics Systems and Communication University of Milano-Bicocca, via Bicocca degli Arcimboldi 8, 20126 Milano, Italy Abstract The paper presents an approach and an architecture supporting the representation of awareness information in cooperative applications. The approach is based on the choice to make a clear distinction between the representation of actors current focus and of the various degrees of awareness information. The architecture is influenced by the basic choice to support flexibility and adaptability of the representation according to users needs. The paper illustrates the approach through a prototype that serves as a demonstrator and discusses the improvements that have been derived from this experience.
1 Motivations and background The recent evolution of cooperative applications shows a great improvement in their functionality, interoperability and integration with WEB technologies. Unfortunately, the design of their interfaces does not show, in general, the same degree of innovation. Most of the application interfaces are still based on a paradigm based on windows, buttons, etc., which is often adopted in WEB interfaces too. Of course, these considerations can be done for single-user application interfaces but the limits of the above paradigm are especially evident in the case of cooperative applications, at least for two reasons. First, the distributed nature of cooperation increases the complexity of the type of information that has to be represented. In fact, users are not only involved in several activities at the same time (like in single-user applications) but they are also part of several cooperative environments in order to perform them. This means that the user interface has to take into account the co-existence of these environments not only from the point of view of their specialized functionality and data exchange but also from the point of view of the co-existence of several groups of users cooperating within or across these environments. This brings to the second source of complexity, namely the representation of the information that is fundamental for an effective articulation of those cooperative activities. It is now recognized that effectiveness of cooperation is obtained by a mix of coordination supports (like protocols and coordinative artifacts) and awareness of the state of the cooperative work arrangement and field of work. Both coordinative supports and awareness information have to be represented in the interface. In [1, 13] we stressed that this representation has to be integrated as the related supportive technologies should be. In [2]
8
F. Nunnari and C. Simone / ThreeDness
we proposed conceptual tools towards this integration, with a special emphasis on protocols. However, these tools are useful to elaborate coordination and awareness information only, and leave completely out any aspect of their representation. The latter are instead the focus of the present paper. The innovation of the cooperative application interfaces is mainly based on the shift from desk-top like metaphors to metaphors describing aspects of the broader space in which cooperation happens. These aspects concern the presence of the involved actors, their location, their availability, their current activities, and so on. Hence, it is natural to look at technologies able to represent the space of cooperation and take into account the above mentioned aspects. For example, WEB-based interfaces can be bent to transform the typical navigation into a journey in an emulated three-dimensional world. Typically, the discrete clicking on active cues (like icons and buttons) is replaced by the continuous displacement of the mouse on the screen to get a more detailed visualization of some parts of the space; or the combination of animations and images represent the change of perspective during the journey. The obvious alternative is to use 3D technologies to simulate the space trough a virtual representation. Since many years there is a lot of researches in the field of VR (Virtual Reality) whose main goal is to reproduce as much as possible realistic environments through computer graphics. With the recent phenomenon of Internet technology, many approaches have been attempted to combine the VR technologies with CSCW principles bringing to what has been defined CVE (Collaborative virtual environment). The main idea is to create a shared virtual world in which several users, represented through avatars, can meet and interact with each other via multi-modal communication [16]. This type of technology has been recently applied, with suggestive results, to interpret artistic performances as the result of a collaboration among real and virtual characters, members of the audience and a production crew [3]. In general, CVE applications are based on scenarios characterized by a tight collaboration among the involved actors in the virtual world. A crucial issue in this respect is the relation between this collaboration and the concomitant collaboration in the real world. In other words, the general question is: how can people be simultaneously immersed in the two worlds? How can the latter be coherent? Another field of application of special types of CVEs is collaborative engineering: here, for example, the possibility to reproduce detailed features of products allows designers to perform distributed simulations in various phases of the design process, with the obvious advantage to collaboratively anticipate and solve emerging problems. It not immediately evident how the outcomes of the above experiences can highlight the design of interfaces for less specialized CSCW applications. We believe that each situation requires ad hoc considerations, and that it is worthwhile to proceed in different directions before trying any generalization. The scenario we are taking into account is characterized by the presence of distributed autonomous actors, cooperating in various kinds of activities. Here, distributedness and autonomy imply the articulation of local, autonomous choices to focus on different, concurrently executable activities. In this scenario, cooperative actors do not primarily need full immersion in, nor exacerbated realism of, the representation of the common working space. As for the first aspect the interface should give cooperative actors the possibility to distinguish their currently focused working space from the space generating awareness of the rest of the collaborative environment. As for the second aspect, the representation of the latter has to reproduce the relevant information in a way that it is easy to grasp and understand. Hence, altogether, our approach is closer to the one proposed in Nessie [6] and Tower [7] where the focused working space is kept separated from the awareness space. However, our
F. Nunnari and C. Simone / ThreeDness
9
approach has some differences. First, we do not aim, at this stage, to 'facilitate encounters' although this feature could be part of our goals in the future. Second, the interface has to give cooperative actors a view of the awareness space from their individual perspective. Finally, the collaborative environment is represented on the basis of spatial metaphors that flexibly combine features of the physical setting with features characterizing the logical structure of the collaborative environment. Accordingly, we want to use the capabilities of the technologies constructing virtual environments to create Virtually Augmented Common Working Spaces, revisiting the ideas underlying the approach called Augmented Reality, whose aim is to populate and enrich a real world with computer generated graphics. An additional, long-term goal is to realize those spaces so that users can specify them according to their preferences in full integration with the technology they use to support cooperation. The paper is organized as follows: the next section illustrates the above goals more in detail. Then the approach is exemplified in Section 3 through the description of an implemented demonstrator. Section 4 illustrates how the new architecture goes in the direction of the abovementioned goals. The concluding section identifies the next steps of this research effort. 2 Basic principles and key concepts The goals mentioned in the previous section require a more precise argumentation where to ground the solution we want to present. To this aim we have to better specify the scenario we want to consider. Then we discuss the theme of the integration of the new functionality with the possibly pre-exiting technology. 2.1
More about the considered Common Working Space
We described our main goal as the construction of a Virtually Augmented Common Working Space. This locution recalls the notion of Common Information Spaces widely discussed in [4]. The two notions, although similar in some respect, show a different emphasis: the latter emphasizes information while the former emphasizes work. We believe that the emphasis on information is too narrow to capture all the aspects we are interested in. In fact, the reference metaphor and almost all the field studies used to illustrate the concept refer to an enlarged view of a repository of information. Instead, we aim at a space populated by actors, activities and all kind of resources, all considered on the equal foot. Hence, this space is dishomogeneous and distributed. Moreover, its structure is dynamic since the above entities can be strongly or loosely connected depending on the evolution of the state of the cooperation happening in the space. For us the term work is more adequate to represent the above richness. The co-related complexity makes this space sometimes difficult to be perceived by the cooperating actors. In fact, it is not identified through a single or well recognizable physical or logical space. The distributed activities make parts of the space opaque for some observer: the pertinent information is not reachable and cannot be used for coordination purposes. But we know that the awareness of what is going on, in a close or peripheral way, is the glue that makes the space exist, take a meaning for the involved actors and play a role in cooperation. In this scenario the use of a technology supporting the perception of the common working space is especially motivated since it provides coordination capabilities that are otherwise impossible [5].
10
F. Nunnari and C. Simone / ThreeDness
Figure 1: Representation of the focused working space and of the awareness information.
On the other hand, augmenting the capability to perceive the above complexity can overwhelm the actors and make the related information practically useless. Hence, it should be possible for each user to access exactly what is pertinent to her current interest. With the concept of focus we denote someone's attention for some specific aspects of the space. For example, although people are normally involved in several tasks at the same time, however nobody is executing more than one task in any given moment. Hence, people switch, more or less frequently, from one task to another and in so doing they change the point of view from which they acquire awareness from what is going on in the common working space. We want to use the notion of focus for two purposes: first, to create a place where actors can concentrate on their current duty or interest and find the related resources; second, as the perspective (the source of the geometrical axes) from which the broader cooperation space is looked at (represented). This choice has in turn two implications. On the one hand, the information the actor is focusing on and the awareness information have to be easily distinguishable. The way in which we want to obtain this result is to use different supports: typically, two screens devoted to contain the representation of the focused working space and of the awareness information, respectively (Figure 1). The same idea is adopted in [6, 7] and its effectiveness has been discussed in [ 12] on the basis of a field investigation. On the other hand, awareness information has to be represented by taking into account different degrees of 'proximity' with the current focus: from very close to fully peripheral. This helps actors to grasp in an immediate way an important feature of awareness information. To this aim we conceive the entities populating the common working space as linked by different types of relations: from simple physical proximity up to pure logical relations. These relations define the texture of the space and allow the definition of what we call logical distance between entities. By taking the perspective of the observing actors together with their current focus and by combining the mutual distances between entities (a sort of transitive closure on the graph defined by the texture) we can compute the distance between the observer and each observed entity. While physical distance can be univocally obtained by a simplification or abstraction of the real work setting, logical distance is more difficult to define since the relations constructing the texture of the logical space are, in general, not ready at hand. This is where the integration of the technology supporting cooperation with the one promoting awareness plays a fundamental role. As illustrated in [13] the needed relations can be derived from the logical
F. Nunnari and C. Simone / ThreeDness
Figure 2: Information flow.
structure of the applications supporting cooperation, possibly augmented by ad hoc relations for sake of awareness. The same paper discusses how this can be done by different types of actors (playing the role of application designers and/or end-users) through the functionality provided by a software module, called AW-Manager. Obviously, this type of solution is not mandatory: the lack of a similar integrated functionality requires other strategies to determine the relations that can be meaningful for sake of awareness promotion. The proposed approach has the advantage to capture, 'at the source', the information related to the cooperative applications concerning articulation work as well as the field of work [14]. And more importantly, it allows for the dynamical updating of this information according to changes affecting the applications in the course of their execution. 2.2
The information flow
For the above reasons we consider the scenario illustrated in Figure 2. Users are the source and the destination of the awareness information. They play as source by using the applications supporting coordinated activities. The related facts are passed to the component in charge to transform them into awareness information. The latter is interpreted by the component managing its representation and finally presented on the user interfaces. The information flowing through the different components has to be codified so that the next component can elaborate it. To this purpose, the central component, in our case the AW-Manager, plays a pivotal role. The facts it is able to interpret mainly concern: the creation/deletion of entities and of the involved relations (i.e., the evolution of the space), and the change of state of the entities and of their position in the logical space (events like an actor assuming a role or activating a task or accessing a resource change the proximity of the related entities). These facts are transformed by the AW-Manager into awareness information on the basis of the aware ness model it incorporates. The latter (called MoMA) belongs to the class of the Space-based Awareness models [6, 7, 8, 9, 10, 11] and is based on a reaction-diffusion metaphor. The AWManager elaborates the incoming facts in terms of modifications of the awareness (logical) space or of the proxies of the application entities populating the space itself. In addition, it generates the related awareness information on the basis of scripts of primitives that define the implications of the above modifications. For example, a script can establish that when an
12
F. Nunnari and C. Simone / ThreeDness
actor activates a specific task a notification has to be generated. The target entities are determined on the basis of the texture of the space, their logical distance from the source entity, the strength of the message to be notified, the propagation law associated to the notification (for example, the original strength can decrease with the distance), and finally the threshold of sensitivity of the candidate receiving entities. Hence, the awareness information to be represented concerns the changes of states and position of the entities, and the type and strength of the notification reaching the entities that are proxy for an actor. The 'values' of the notification to other types of entities are possibly used to define and elaborate the propagation of awareness information but are not involved in any representation at the user interfaces. The last component, that we call ThreeDness, is in charge of the graphical representation of the awareness information received from the AW-Manager. This component is illustrated in the following sections for what concerns its functionality and internal architecture. 3 The current prototype In order to check the feasibility of the basic ideas illustrated in the previous section we developed a prototype to be used as a demonstrator. It does not contain all the features fulfilling our short and long term goals. However it serves as a test-bed for the preliminary versions of the functional and the architectural solutions. In the next section we will discuss the improvements we are working on, also in relation to this experience. In order to construct the demonstrator we have to identify a reference scenario: it has to be general enough to describe a wide range of working environments, and concrete enough to create a natural mapping between logical entities and graphical visualization. Moreover, in order to avoid, at this stage, serious integration problems with the AW-Manager (whose implementation is still under improvement) the demonstrator is built as a stand-alone component which receives awareness information facts concerning potential cooperative applications from an interface devoted to simulate this interaction. 3.1
The reference scenario
The scenario considers a work arrangement in which actors cooperatively accomplish tasks. Tasks are steps of a distributed project and require resources. This simple scenario allows us to take into account both focus (on the cooperation around a single task) and various degrees of peripheral awareness concerning other tasks as well as the related actors and resources. The scenario is described in terms of entities and relations among them as follows. 3.1.1
The texture of the logical space
The reference logical space is populated by four types of entities: Task, as the unit of the cooperative work; Role, defining the Actor's responsibility for Tasks and the Resources used in the accomplishment of Tasks. There are two types of resources: informative resources and tools. Tools are all kind of physical resources, as computers, printers or desks. Informative resources are of a more abstract type, as a working document. The above entities are related with each other in the following way: Actors <-> Roles An actor can play different roles and vice versa. Actors <-> Resources An Actor can use more resources at the same time, but a resource can be used at most by one actor.
F. Nunnari and C. Simone / ThreeDness
13
Figure 3: AND/OR graph of a task-split project.
Roles <-> Tasks A role is responsible for a task. A task can have just one responsible role which can be played by several actors. Hence, the relation between Actors and Tasks is mediated through the notion of Role. Roles <-> Resources A role is responsible for a resource. A resource can have just one responsible role which can be played by several actors. Tasks <-> Resources The resource is necessary for the accomplishment of a task. A task may need several resources, and a resource can be used in the accomplishment of several tasks. Task <•> Task A project is described as a partial order among tasks to express priorities and constraints in their accomplishment. This relation is represented by an AND/OR graph. Figure 3 shows an example. T0, Tl, ..., T5 are tasks. R0, Rl, ..., R5 are the roles assigned to their accomplishment. To develop the project one can choose to start with task T5 or to follow the alternative branch (OR). In this case, task TO has to be finished before tasks Tl and T2 can start: the latter can be developed in parallel (AND). Before task T4 is started, task Tl and T3 have to be concluded. The same holds between task T3 in relation to task T2. 3.2
The resulting Virtually Augmented Common Working Space
In this section we describe how ThreeDness visualises the above scenario. First, we illustrate the representation of the background scene. Then we describe how is it personalized and augmented for each user. 3.2.1
The visualization of the background scene
Figure 4 shows a snapshot of the three dimensional scene visualizing the background scene. The visualization is based on a simplification of the physical space. The leading idea is to see the latter as composed of rooms, each containing the work places of the Actors living in it. A work place is composed of an avatar, a desk, a table and an archive. The avatar represents the Actor. The desk is its local working space, that is, the place the Actor focuses on while working on a specific task. The table represents the place where the Actor piles suspended tasks, that is, tasks in which it is involved but not currently working
14
F. Nunnari and C, Simone / ThreeDness
Figure 4: The background scene.
on. The archive represents the place where an Actor locates its information resources: they will be part of the personalized working place discussed in the next section. Tools can be located in common spaces, like corridors, or be part of a working place. Figure 4 shows an example of both cases. The same figure contains empty rooms. They represent a portion of the space that does not play any role in the considered cooperation arrangement. However, their presence improves the representation of the physical setting and can help users to move more easily in it. There is a one-to-one correspondence between ThreeDness users and avatars in the scenario. Hence every ThreeDness user sees both the representation of himself and of all the other users. The first time a user activates the ThreeDness application, a window pops-up presenting the background scene, built of rooms and work places. The latter are used as basis for additional graphical representation of the awareness information we want to promote. At this point, each user can set some parameters through the panel (shown in Figure 4 ) in order to define which type of information will be added to the background. This selection can be changed at any moment, through the same panel. From now on, each user interface will be different: partly because of the above choice, partly because the background will be shown from the logical and physical perspective of each user, as illustrated in the next sections. 3.2.2
The focus and its closer awareness information
As discussed in Section 2, we propose to use two supports (typically, two screens) in order to represent the current focus and the awareness information about the surrounding Common Working Space, respectively. In this view, the first screen contains the interface of the application used to perform the focused task. Possibly, this interface contains cues representing awareness information closely related to the application itself: for example, in the case of a co-authoring system telepointers or cues representing the current state of portion of the text, and the like. The second screen contains the background scene illustrated above, enriched with additional information. In our reference scenario a key information concerns tasks: hence we describe how tasks are represented and the resources they use (Figure 5). The Tasks an Actor is working on are represented as little coloured boxes on the desk or on the table of that Actor. Boxes are labelled with the name of the Task they represent.
F. Nunnari and C. Simone / ThreeDness
15
Figure 5: The representation of Tasks.
Both their position and colour convey a meaning. Boxes on the table represent Tasks that are already done (grey) or that will be activated in the future (blue), according to the partial order relation described in Section 3.1. Their mutual position in the pile represents the temporal order by which they have been or will be executed. Boxes over the desk represent tasks that are currently under consideration (red). The particular Task whose specialized working environment is the content of the first screen, is still represented with a box on the desk, but with a green colour. The representation of the Resources requires a specific comment. Recall that a Tool is a physical object while an Informative Resource may be not. Tools are represented by three dimensional objects, possibly abstracting from details, that everyone can recognize as such: e.g. a printer. Informative Resources are more difficult to represent: in fact, they can be a piece of paper, or a little book, or a text file in some computer. Beside the content it could be useful to represent the media used to keep it. Hence, such a representation is more subjective that in case of Tools. In the construction of the demonstrator we adopt a simplified solution and represent Informative Resources as folders in the archive. Again, different colours describe their state (e.g., in use or idle) and a label describes its contents. We will come back to this solution in the concluding section. The representation takes into account the two possible relations between an Actor and a Resource. If the Actor is responsible for a Tool, the latter is placed close to its working place. If the responsibility concerns Informative Resources, the latter are located in the lower part of the archive of the responsible Actor's working place. Information Resources in use of some Actor are located in the upper part of its archive, while Tools in use are logically connected to the Task where they are used: this link is visualized through a continuous line if the related option has been selected in the above mentioned panel. The representation of the focused space is coherent with the choice to incorporate the Actor (avatar) representing the user in the background scene together with the other Actors. We believe that this double presence naturally provides a continuity between the contents of the two screens. Moreover, the abstract representation of the information concerning the focus (the green task) and the user's avatar define the point of view from which the space is perceived, and therefore help the user to keep the sense of orientation in the more peripheral
16
F. Nunnari and C. Sirnone / ThreeDness
Figure 6: The Shared Tasks representation is different for different users, awareness space: a sort of spot telling a "you are here" information.
3.2.3
Representing peripheral awareness information
All what (physically and logically) surrounds the focused Tasks can be source of awareness information. However, the latter can be of different relevance in relation to the (physically and logically) distance between the observed entity and the observer. Hence, awareness information has to be represented according to different degrees of relevance. Here below we discuss two possibilities that we considered in implementing the demonstrator. An example of awareness information that is close to the Actor and its focused Task concerns the other Actors involved in it. Of course, this information can be part of the interface concerning the focus but the same information in the representation of the Common Working Space can provide additional information. Typically, this happens when some Actor is not working on this Task as the observer could expect. The representation of the Common Working Space can tell the observer something about why this is the case. The graphical visualization of the Tasks allows the scene to contain several boxes representing the same Task, each over a different desk (or table) because each Task can involve different Actors. To make this fact evident, boxes representing the same Task are connected with a little coloured tube. Figure 6 shows an example of the visualization of the information about shared Tasks: they are connected by tubes suspended in the air, crossing walls and furniture, to highlight the fact that the two boxes at their ends stand proxy for exactly the same Task. The colour and position of the task instances together with the current focus of each visible Actor give the additional information we mentioned before: about which Task each Actor currently considers of higher priority (its focus) and the priority of the Tasks shared with the observer (their mutual position in the piles of the working place of each visible Actor). This information cannot belong to or be generated by the application executing the focused Task of the observer. The above visualization can be done for all Tasks in the working space of the observer by pointing them with the mouse. The user can choose if the visualization has to be permanent or transient (i.e., disappearing with the displacement of the mouse). Notice that, since the drawing algorithm is strongly depending on the identity of the observer, every user interface shows a different visualization of the (potentially) same awareness information. In the above situation, the value of the distance between the connected Tasks is one, that is they are immediately connected by the (derived) relation: being an instance of the same Task. In the more general case, the logical distance defines several degrees of proximity. We choose to represent distance in terms of the dimension of the involved object. The farer is
F. Nunnari and C. Simone / ThreeDness
17
Figure 7: The Logical Distance representation is modifying the size of the rooms in the scenario.
an object, the smaller it is represented. This is coherent with the physical representation of the background scene. However, acting at the level of single object would be confusing for the observer: in fact, each room can contain different objects with different dimensions. We tried this solution and realized immediately that it was unacceptable. Hence, we decided to apply the change of size to the rooms of the scenario. For every room a resizing algorithm performs an evaluation of the logical distance between the user of the application and the content of the room. The algorithm bases the resizing on the logical distance of the elements contained in the room. A room close enough is kept in its original size, while the (logically) farer is a room, the more it is shrunk. Hence, if the content of a room is logically close to the observer, the room is big enough so that its content is clearly visible. On the contrary, if a room is logically far, it is shrunk to make its content hidden. Figure 7 shows the difference between an interface where the logical distance is visualized and another one where it does not. The picture on the left is the normal scenario, the one on the right is showing the same scenario modified by the resizing algorithm. The visualization is changing from an observer to another one. First, every user can choose through the mentioned panel whether to activate the logical distance representation or not. Secondly, the computation of the logical distance of the rooms is different for each observer because it is dependent on each user current focus. So the topology of the background scene is the same, but the size of the rooms can be different. Notice that if a shrunk room contains a task shared with the observer, the previously described tubes enter the room without any additional precision. Again, by pointing the room it can be zoomed and the more detailed information will appear. 4 The ThreeDness architecture We describe now the internal architecture of ThreeDness in the general context described in Figure 2. As for the technologies used for its implementation, the code is written in Java, the network communications are RMI based and the 3D graphics engine is based on the standard Java3D extension. In this section we first describe the client/server architecture of ThreeDness. Then, we give a deeper look at the Plugin system, a general purpose software architecture that has been incorporated in ThreeDness.
18
F Nunnari and C. Simone / ThreeDness
Figure 8: Client/Server Architecture of ThreeDness.
4.1
The Server and the Clients in ThreeDness
ThreeDness is composed of two distinct modules: the server and the client, as shown in Figure 8. The bottom part shows the server side architecture, while the top part shows the client side. The dotted line in between represents the network communication between the two sides. Following the standard client/server paradigm, there is a single machine running the server application, while several clients establish a network connection with the server. Server side (bottom): During its execution the server communicates with the AW-Manager, the application in charge of collecting facts about users activities and transforms them into Awareness Information. A deeper discussion about the AW-Manager and its underlying architecture can be found in [13]. The ThreeDness server is mainly composed of three parts: A PluginServer, an Awareness Information to 3D-Awareness Information conversion module and a Plugins repository. The conversion module contains a set of algorithms to transform the Awareness Information, coming from the AW-Manager, into a data structures suitable to deal with its graphical representation. In fact, the AW-Manager is a source of data fully unaware of any graphical representation. We call the converted information 3D-Awareness Information. The PluginServer functionalities and the Plugin system managing them will be discussed in Section 4.2. The PluginServer contains a Plugin dedicated to collect 3D-Awareness Information from the conversion module and send the related updated data to all the connected clients. Finally, the Plugins repository contains the Plugins devoted to the representation of the
F. Nunnari and C. Simone / ThreeDness
19
3D-Awareness information: they are loaded by the PluginServer and sent to the clients when they connect. Moreover, when they are updated for some reason, the PluginServer reloads them and all the connected clients are forced to download a copy of the new Plugins. Client side (top): The ThreeDness client is composed of three parts: The PluginContainer, the 3D graphic engine and the graphical user interface. As before, the details of the PluginClient will be discussed in Section 4.2. When the ThreeDness client establishes a connection with the server, the PluginClient downloads from the server all the needed Plugins. They constitute the main part of the code running on the client. One of these Plugins, named "Local copy of awareness informations" is dedicated to manage the copy of the 3D-Awareness Information received from the server. The latter is elaborated by the other Plugins that are responsible to convert them into 3D Directives for the 3D Graphic Engine that generates the graphics for the graphical user interface. As anticipated in Section 3.2.1, users can set some parameters to customize their 3D interface. These parameters affect the above mentioned Plugins. Notice that customization is local to every client, that is, every client receives the same 3D-Awareness Information but visualizes it differently depending on the users preferences. 4,2
The Plugins System
The Plugin system is a peculiar characteristic of ThreeDness. First we give the basic motivations and describe how it fits into the ThreeDness architecture. Then we give some technical detail. 4.2.1
Plugins usage in ThreeDness
The idea of having remotely downloadable Plugins came out from two needs: the customization of the visual representation and the efficient management of clients updates. Since the combination of these two needs, namely customization and efficiency, is typical of many applications, we developed a general purpose Plugins system that can be used irrespective of the type of functionality provided by the managed Plugins. Of course, there are ready-at-hand solutions (like, e.g., Java beans [17]) but generally they show two problems: they are too complex to manage and more importantly they cannot be used as general RMI objects. In order to fulfill the need of customization in our framework, it is necessary to provide every client with a copy of the algorithm translating 3D-Awareness Information into 3D Directives for the 3D Graphic Engine. The choice to let the server compute the 3D scenario and then distribute it to the clients has two disadvantages: although simpler, this architecture makes customization more difficult and over-charges the server with heavy and complex 3D calculations. The need of customization implies also a modular architecture that facilitates the modification of partial functionalities and their integration in the overall application. All the above considerations motivate the choice to organize all the exchanges of code and data structure between the server and the clients in terms of Plugins. In this way, we obtain at the same time: modularity, mobility and an efficient management of the 3D computations. In fact the ThreeDness server does not need to deal with 3D graphics and to have any 3D library installed. This brings another advantage. With most of the code downloaded dynamically at boot time, the need of reinstalling a client on every machine, when some code change occurs, is significantly reduced. By the way, this makes easier the
20
F. Nunnari and C. Simone / ThreeDness
Figure 9: The hierarchy of the PluginContainers class.
incremental development of ThreeDness, where changes are quite frequent. Moreover, all clients do not need to share the same 3D libraries. One may argue that a complete upload of all Plugins may be heavy from the point of view of the network load. Currently the application has been developed for local area networks, where transfer speed is fast enough and does not create any problem. Anyway, we believe that, if band-with becomes a problem in some situation, a cache mechanism might be implemented to reduce the data transfer by preventing the client from the upload ofPlugins that are not changed in the server repository. 4.2.2
Components and features
The purpose of the Plugin system is to give support for a flexible and easy exchange of Plugins. Figure 9 shows the three most significant classes of the system. The PluginContainer at the top of Figure 9 is a class providing the basic functionalities that are common to both the PluginClient and the PluginServer subclasses (bottom of the figure). These functionalities handle the collection of Plugins in order to: • Insert and remove Plugins to and from the container. This is the basic feature that allows the designer to decide which Plugin are available in ThreeDness. • Initialize and destroy a Plugin once it's inserted in the container. Every Plugin has the responsibility to initialize itself after being inserted in the container and to de-initialize itself before being removed. • Realize Plugins inter-communication. With this mechanism every Plugin has the list of all the other Plugins inserted in the container and can interact with them. The two sub-classes derived from this class add networking functionalities to the PluginContainer. Normally, when instantiated, a PluginContainer is empty. The PluginClient is a special case of a PluginContainer with the additional capabilities to dynamically download
F. Nunnari and C. Simone / ThreeDness
21
Plugins from the specified server. The PluginServer is a specialized PluginContainer with the additional capability to upload Plugins to a client requesting them. Note that the Plugins contained in the server are different from the Plugins that will be sent to the client. In fact the server is containing two sets of Plugins: one set for itself, as extension of a stand-alone container, another one of Plugins that will be uploaded to the clients. In addition, the Plugin system has the possibility to force all clients, at run-time, to de-initialize and remove the contained Plugins and to reload a new set from the server. 5
Conclusions
The paper presented an approach and an architecture supporting the representation of awareness information in cooperative applications. The approach is based on the choice to make a clear distinction between the representation of the focus and of the various degrees of awareness information. The architecture is based on the choice to support flexibility and adaptability of the representation according to users needs. The presented demonstrator served as a test-bed for the initial solutions as well as a means to collect initial impressions from people with different interests: potential users and interaction designers. Irrespective of the evident limits of the demonstrator, ThreeDness was perceived as an original approach to the problem. By the way, it served as a vehicle bringing the themes concerning the promotion of awareness information to professionals involved in single-user interfaces design: they are almost unaware of the complexity inherent to cooperative applications. They recognized it as a quite challenging field of research where consolidated techniques are of little help. Moreover, they pointed to some application areas (typically, remote process control) where similar ideas could be profitably applied. Perhaps, an interesting outcome was that they recognized that also the design of single-user applications could profit from our approach. The new version of ThreeDness is based on the architecture presented in Section 4. The flexibility it provides is currently at the infrastructural level as a fundamental step toward flexibility at the application level, where we aim at the highest degree of customization. This means, customization is not only based on tuning parameters but more substantially on the choice of the appropriate metaphor to represent the background scene as well as the personalized awareness information. The choice to let the representation be constructed at the client side, under user control, realizes this approach without creating interpretation problems. In fact, the basic information is the same at each site, the only difference being how it is filtered and represented. When information goes from a site to another, it is represented here according to the local customization. Moreover, the virtual space is used for sake of awareness promotion and not necessarily as an arena where to perform cooperative activities. Finally, the distinction between focus and awareness space seems to open the possibility to apply the proposed approach in the case in which the focus interface is incorporated in objects distributed in the physical space. In this case, the awareness information represented on the 'second' screen can help to reconstruct an even more distributed space whose perception depends on both the logical and physical position of the current focus of the observer. References [1] K. Schmidt, C. Simone: "Mind the gap! Towards a unified view of CSCW". In COOP2000. The Fourth International Conference on the Design of Cooperative Systems, Sophia Antipolis, France, 23-26 May 2000, 2000, pp. 205–221.
22
F. Nunnari and C. Simone / ThreeDness
[2] S. Donatelli, M. Sarini, C. Simone: "Toward a Contextual Information Service supporting adaptability and awareness promotion in CSCW systems". 2000. Designing Cooperative Systems. R. Dieng et al. IOS Press, pp. 83–98. [3] A. Drozd, J. Bowers, S. Benford, C. Greenhalgh, M. Fraser: "Collaboratively improvising magic: An approach to managing participation in an on-line drama". 2001. Proceedings of the 7th European Conference on Computer-Supported Cooperative Work. pp. 159-178. Kluwer Academic Publisher. [4] L. Bannon, S. Bodker: "Constructing Common Information Spaces". 1997. Proceedings of the Sixth European Conference on Computer-Supported Cooperative Work, 12–16 September 1999, Copenhagen, Denmark. Kluwer Academic Publisher, pp. 81–96 [5] M. Buscher, P. Mogensen, D. Shapiro: "Spaces of Practice". 2001. Proceedings of the 7th European Conference on Computer-Supported Cooperative Work. pp. 139–158. Kluwer Academic Publisher. [6] W. Prinz: "NESSIE: An Awareness Environment for Cooperative Settings". 1999. Proceedings of the Sixth European Conference on Computer-Supported Cooperative Work, 12-16 September 1999, Copenhagen, Denmark. Kluwer Academic Publisher. [7] V.A.: "Tower: Cooperation Awareness in a Populated Information Landscape". 2001. Seventh European Conference on Computer-Supported Cooperative Work, Conference Supplement-Demonstrators, pp. 99101. [8] C. Greenhalgh, S. Benford: "MASSIVE: a collaborative VE for Tele-conferencing". 1995. ACM TOCHI, vol. 2, no. 3, pp. 239-261. "http://www.crg.cs.nott.ac.uk/research/systems/MASSIVE". [9] S. Benford, C. Greenhalgh: "Introducing third party objects into the Spatial Model of Awareness". 1997. In 5th European Conference on CSCW, Lancaster - UK, ed. W. Prinz, T. Rodden, and K. Schmidt. Kluwer Academic Publishers, pp. 189-204. [10] T. Rodden: "Populating the Application: A Model of Awareness for Cooperative Applications". 1996. In CSCW '96. Proceedings of the Conference on Computer-Supported Cooperative Work, Boston, MA, November 16–20, 1994. New York: ACM Press, pp. 87-96. [11] O. Sandor, C. Bodgan, J. Bowers. "Aether: an awareness engine for CSCW". 1997. In 5th European Conference on CSCW, Lancaster, UK, ed. W. Prinz, T. Rodden, and K. Schmidt. Kluwer Academi Publishers, pp. 221-236. [12] J. Grudin: "Partitioning Digital Worlds: Focal and peripheral awareness in multiple monitor use". 2001. Proceedings of the CHI 2001 Conference. March, 2001. [13] C. Simone, S. Bandini: "Integrating awareness in cooperative applications through the reaction-diffusion metaphor". Computer Supported Cooperative Work: The Journal of Collaborative Computing - Special Issue on Awareness in CSCW (to appear). [14] K. Schmidt, C. Simone: "Coordination Mechanisms: Towards a Conceptual Foundation of CSCW System Design". 1996. Computer Supported Cooperative Work: The Journal of Collaborative Computing, Kluwer Academic Publishers, pp. 155–200. [15] M. Divitini, C. Simone: "Supporting different dimensions of adaptability in workflow modeling". 2000. Computer Supported Cooperative Work: The Journal of Collaborative Computing -Special Issue on ' Adaptive Worflow Systems', vol. 9, no. 3–4, pp. 365-397. [16] S. Benford, C. Greenhalg, T. Rodden, J. Pycock: "Collaborative Virtual Environments". 2001. Communications of the ACM, vol. 4, no. 7, pp. 79-85. [17] http://java.sun.com/products/javabeans/
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press, 2002
23
Reconciling Different Perspectives: An Experiment on Technology Support for Articulation Gloria Mark1, Victor Gonzalez1, Marcello Sarini2 and Carla Simone3 'Department of Computer Science, University California, Irvine, Irvine, CA 92697 e-mail: {gmark, vmgyg}@ics.uci.edu 2 Dept. of Computer Science, University of Torino, Corso Svizzera 185, 1-10148 Torino, Italy, e-mail:
[email protected] Department of Informatics, System and Communication (DISCO), University of Milano-Bicocca, Via Bicocca degli Arcimboldi 8, 20126 Milano, Italy, e-mail: carlo, simone @ disco, unimib. it Abstract A challenge in developing collaborative technologies has been to support communication of people who are distributed, and who have different perspectives. We conducted an experiment to test the Reconciler, a system designed to aid communicating partners in developing and using shared meanings of terms. We compared the use of technical terms in text-based online communication, with and without the system. We found that groups who did use the system used fewer clarifications and explanations, and had a higher proportion of task-related messages. We interpret this to mean that the system aided the memory of shared meanings, which was related to more of a task focus. We argue that technology support for articulation can provide benefits in distributed settings. Keywords CSCW, Computer-mediated communication system, experimental study, system evaluation, human cooperation, articulation work, semantic interoperability.
1 Introduction For some time, organizations have been moving toward designing activities that are characterized by a combination of highly specialized skills and capabilities. A high degree of communication and coordination is required to achieve the appropriate mix of competencies to carry out this kind of work. Consequently, organizations are increasingly evolving toward a structure of autonomous, specialized components forced to cooperate to accomplish complex activities. This trend triggers, and is triggered by, the evolution of technology that supports communication and cooperation in diverse work settings. Despite the richness and powerfulness of current technologies (in terms of functionality, connectivity, integration and so on), there is not adequate support for the articulation of such kinds of distributed activities. In this paper, articulation refers to the communication effort that is needed to mesh different perspectives to achieve coordinated activity. Specifically, we refer to the tension between the need of individual perspectives to effectively work in local contexts (local articulation) and the need to develop shared
24
C. Mark el al. / Reconciling Different Perspectives
meanings to interpret boundary objects and actions which characterize intergroup cooperation (global articulation). Although this tension has been documented by many field studies [e.g. 3, 6, 15, 17] and these results have been reflected upon [1, 2], no solutions have been proposed either for technological support, or as conceptual foundations for developing technological support [14]. In a previous paper [13] we illustrated how this problem takes a specific connotation if considered from a CSCW perspective, although it has been considered in other ways by disciplines dealing with semantic interoperability (e.g., Information Systems and Knowledge Engineering). In fact, this point of view prevents the adoption of solutions that work well at the technological level, but are impractical or unacceptable by human cooperating actors. This set of motivations led us to develop specific functionality aimed at aiding actors in reconciling differences in their perspectives. This functionality is incorporated in a software module, called the Reconciler, that can be ideally integrated in different technologies supporting cooperation, via the design of suitable information interfaces (API). In [13], the Reconciler was presented in relation to the technological framework supporting the design of coordination mechanisms [12] with a specific emphasis on their adaptability by users in relation to their current needs [5]. The goal of supporting the reconciliation of different perspectives in cooperation can be achieved in two ways: either incorporating in the support a predefined model of reconciliation or to propose a light support and let users define their own way to handle the reconciliation process. In the first case, models based on argumentation and explanation could be adopted as well as based on the structure of the dialogue (e.g., 4,7,10,16,18). However, our approach was to implement a light support, to experiment its usage and to base a possible set of richer functionalities on the result of the experimentation. In the practices of people we can recognize the validity of features proposed in other approaches and incorporate them on the basis of empirical evidence. In fact, one of the primary aspects we want to test is the overall acceptance of this kind of technology in a scenario that is not so usual in CSCW. Although the conception of the Reconciler was based on empirical work [11], its functionality was never tested by users. Similar motivations characterized another proposal which, to our knowledge, is the only one trying to help people manage the tension between different perspectives. Macadam [6] is a prototype allowing users to construct personalized local views of a global classification scheme organizing information in an engineering environment. It provides users with means to access information, making those views transparent to them. For contingent reasons, Macadam too, was never implemented in that real setting. The lack of empirical studies led us to plan an experiment in order to understand how a tool like the Reconciler can support articulation. To make the experiment more feasible and meaningful, we chose a scenario that was not based on groups having a significant tradition of local (intragroup) cooperation using mechanisms like workflow systems or shared workspaces. We chose instead a scenario that concentrates on the communication and global articulation needed for the cooperation of people with different skills and perspectives, that is, with different professional languages who are working in a distributed setting. This cooperation would lack the natural mediation of a shared local framework. Here the main problem is to understand and use the technical terms typical of the profession of one of the two interlocutors. The paper is organized as follows: the next section provides a description of the interfaces provided by the Reconciler and their functionalities in supporting the collaboration among people. Section 3 describes the experimental setting. Section 4 and 5 provide the results of the experiment and the related discussion and implications for the design of a technology supporting articulation, respectively. Conclusions identify main issues and open problems.
G. Mark et al. / Reconciling Different Perspectives
25
2 The Reconciler interface and functionality The support provided by the Reconciler to two communicating professionals can be divided into two parts: the first one is devoted to the definition of terms of one profession with their 'meaning' as described from the point of view of the other interlocutor; the second one supports the use of this 'reconciled' terminology in the messages exchanged between professionals during their cooperation. Accordingly, the environment for the Reconciler used in the experiment is structured as illustrated in Figure 1. Definition Interface
Definition Interface
Reconciler
Communication Interface
Communication Interface
2.1 The definition interface The Reconciler definition interface is a component (written in Java) which is shared between the two interlocutors by using an application sharing software. This interface allows the interlocutors to establish 'correspondences' between the different terms they are likely to use in their cooperative activity, so that they can better communicate. We refer to these as definitions. To give an example, the two interlocutors are A, a Chief Technology Officer, and B, a Product Manager. Without loss of generality, we illustrate the case where a given list of computer science terms used by A is considered. With the help of A, user B identifies descriptions which she, as a non-technical person, can understand and use. The interface is designed to facilitate the introduction of definitions for the technical concepts (Figure 2): it is possible to define or delete concepts and associate new definitions with them, or to change them. Moreover, the user can reuse existing definitions using the Define As button. This is useful, since for non-technical persons, different technical terms can be naturally described in the same way. If, afterward, the user wants to distinguish between them, it is also possible to associate further information to better specify the concepts by adding attributes. For instance, the concepts HTML and WML could both be defined by B as WEB PROGRAMMING LANGUAGE (Figure 2). Since WML is related to a more specific use of a PDA/mobile phone, while HTML is more related to a PC, this additional information can be expressed as a specific value of the usage attribute. An automatic selection is provided, in order to help users recognize the pre-defined correspondences. That is, when a concept or a definition is selected, the corresponding definition or concept(s) is highlighted.
26
C. Mark et al. / Reconciling Different Perspectives
Figure 2: The definition interface
At the end of the definition phase the collected information is saved in a persistent format and is used by the Reconciler in the communication phase. 2.2 The communication interface Once negotiated shared meanings are found, the interlocutors can incorporate the shared meanings into their communication using the communication interface of the Reconciler. The communication interface is a component (written in Java) that the professionals use as two instances of the same application customized to their specific role. This means that the two interfaces are similar (see Figure 3 and Figure 4) but each one shows in the Concepts list panel the list of concepts defined in the previous step according to the specific user's point of view: A will find all the technical concepts while B will find the corresponding negotiated definitions. In addition to the concept list, the interface is structured as a chat tool where the Message History panel keeps track of the messages part of the conversation between A and B. The Outgoing Message panel contains the message that is currently sent. The message is a combination of free text (as for standard chat messages) and concepts chosen from the concept list. The latter are automatically inserted in the message text in square brackets so that the users can immediately recognize that these terms are not simply words, but something they have already discussed before and that the interface will handle in a special way. Once the message is defined, the Send Message button transfers the message to the main component, the Reconciler, in order to be elaborated. The Reconciler algorithm parses the
G. Mark et al. / Reconciling Different Perspectives
27
message to recognize the concepts and formulates them according to the definitions given in the previous step. Notice that in the simplest case the algorithm operates a one-to-one substitution, as in the case of SOFTWARE TO RECOGNIZE HANDWRITING (Figure 2), which is univocally associated to the technical term OCR. A more complicated case arises when two or more concepts share the same definition (this is the case of HTML and WML, both defined as WEB PROGRAMMING LANGUAGE). Here if user B writes WEB PROGRAMMING LANGUAGE there is no unique correspondence for user A to understand which of the two concepts B is referring to. To overcome this problem, the definition interface, as explained above, allows more specific definitions using attributes. The two instances of the Communication interface exchange messages through Java RMI. This allows them to communicate through remote objects which export methods making them visible to the other applications through the network. In the presented simplified scenario, the remote objects simply export a "print" method invoked by the sending instance to write the translated message in the Message History panel of the receiving instance. After pilot tests, some features have been added to the Communication interface according to users' suggestions: • Reminding function: sometimes the users could not remember the definitions given in the previous step. This led us to add new features to the interface trying to help users in remembering and to support their mutual learning process: the first one is a text box which shows the definition of a concept when it is selected from the list (for instance, in Figure 3 the selection of the HTML concept shows the corresponding definition: WEB PROGRAMMING LANGUAGE). The second feature presents to the user the translated concept in the messages to be sent, together with the corresponding definition (see in Figure 4 the last message of the Message History). •
Automatic word completion: In the pilot experiments, the users felt disrupted in their online chat communication. In fact, in order to use the negotiated meanings in their messages, they had to select them from the concept list by double clicking on them. This action made users lose their focus of attention on message writing. To overcome this problem, an automatic word completion feature was added: the interface recognizes when the user is writing a concept and completes it only after a few characters are typed. In this way, users are bothered less by possible erroneous completions. This functionality has been realized adapting the partial matching proposed by [8].
3 An experiment to test the Reconciler Our goals were to understand how the Reconciler might affect articulation and communication. Our interest was to address global articulation problems, i.e. the use of shared meanings in distributed settings. When people cooperate in collocated settings, they have more opportunities for negotiating shared meanings of terms, e.g. by meeting face-toface and discussing them. However, in distributed settings, even if the shared meanings can be negotiated, e.g. through formal meetings, it is a challenge for people to integrate them into their communication. The different contexts of the distributed settings, along with the lack of informal communication, make it difficult for people to "remember" these shared meanings, and use them in their communication [8]. We focused on this latter problem in the experiment. To do this, we developed a process whereby people initially developed shared meanings for terms. We then focused on the use of the communication interface to determine how well it could support people in applying these shared meanings in their communication. We conducted an experiment based on a similar communication scenario as described in the last section. We developed hypotheses on how the Reconciler should affect articulation work:
G. Mark et al. / Reconciling Different
Perspectives
Figure 3. The Communication interface for A (technical).
•
H/: Articulation should be more explicit when not using the Reconciler. Since both actors know that the other understands the term being sent, we would expect fewer messages that concern articulation when the Reconciler is used. Conversely, without using the Reconciler, we expect more explicit messages concerning articulation. We consider messages concerning explanations, asking for clarifications, and confirming the other (establishing common ground) to be articulation messages. As a consequence of H1, our second hypothesis states: • H2: Communication should involve a higher proportion of task-related messages when using the Reconciler. We expect that the use of a system to support the shared meaning of terms would result in actors exchanging a higher proportion of messages that are taskrelated, compared to without such technical support. In addition to these hypotheses, we investigated the use of the interface, and the application of the terms in the communication. To examine the hypotheses and system usage, we designed a task which involved negotiating shared meanings of technical computer science terms. In order to have an experimental paradigm that more realistically replicated the idea that organizations have different specialties that must communicate with each other, we paired together students from different university specialties: computer science and from a non-technical discipline. Articulation was needed for the two people to understand and make sense of the technical terms so that together they could perform the task.
G. Mark et al. / Reconciling Different Perspectives
29
Figure 4. The Communication interface for B (non-technical).
We performed a pretest survey of terms with technical and non-technical students, and based on this, constructed an initial list. We conducted a series of pilot experiments to select terms from this list that we felt confident would be understood by an upper level computer science student, but not by an upper level non-computer science student. 3.1 Experimental design The experiment used a 2x2 between-subjects factorial design. One independent variable was System Use: • R condition: the Reconciler was used to include terms and corresponding definitions in chat messages. • W condition: chat messages were exchanged without the aid of the Reconciler. A second independent variable was Role: • Chief Technology Officer (CTO) • Manager (M). 3.2 Subjects Subjects were upper level undergraduate students at the University of California, Irvine. Forty subjects participated in the experiment, 10 dyads in each System Use condition. Subjects were recruited for the CTO Role if they were computer science students, had programming experience, and used the computer for specialized technical tasks. Subjects were recruited for the Manager Role if they were from a non-technical field, had no computer science course work, no programming or other specialized computer experience, but did have
30
G. Mark et al. / Reconciling Different Perspectives
a basic understanding of using a keyboard and mouse. Technical and non-technical students were randomly paired together, and randomly assigned to the R or W condition. Sixteen females and 24 males participated. Only three females were in the role of CTO. Subjects were paid for their participation. 3.3 Task and Roles The task intended to recreate a common situation in the IT industry: planning a proposal for a new product: Personal Organizer software for a Personal Digital Assistant (PDA). Since our goal was to study the articulation process, the task focused on the initial phases of planning: selecting general technical features that the software will include or support. These were: (1) Interface: How should information be presented with the interface? (2) Information Organization: How should information be organized to fit the PDA screen, (3) How should input be given? (4) Support for peripherals: What kind of external devices should the software support? (5) Security: What kind of security systems or methods should be implemented, (6) Threat: How can the product resist threat, (7) Communication: methods for the PDA to communicate with other devices. Examples of technical terms that needed to be discussed as options were: RGB, XLST, OCR, Jaz drive, Public Key, SSL, USB. Each student played the role of an employee of a software company. The computer science student played the role of the Chief Technology Officer (CTO) and the non-technical student played the role of the Product Manager (Manager). 3.4 Experimental setup The experiment was conducted in a large single room lab. Students sat at workstations facing each other, but were visually separated by partitions. The goal was to simulate distance collaboration. Each computer was equipped with a small web camera, noise-cancelling headphone and microphone. For the first part of the experiment the students communicated by using audio headphones and video. For the second part, they wore noise cancelling headphones, communicating only with online text chat. 3.5 Experimental procedure The task was split into two consecutive days, 45 minutes each day. On the first day the students defined the list of terms. They returned a second day to use the terms in discussing the proposal. Our intention was to create a time separation so that subjects would have less reliance on their memory of their predefined terms from Day 1. Day 1. The experiment was explained to the subjects as well as the role they would be playing. Both conditions used audio and video to communicate to define the terms. In the R (Reconciler) condition, the subjects used the definition interface of the system and worked together to define the technical term so it would be understandable to the Manager. In the W condition (without Reconciler), subjects defined the same terms as in the R condition, but wrote the results on paper. An example was for "HTML" to be defined as "web language" for the Manager. They had to define 39 technical terms. Day 2. AH subjects were trained for five minutes on using the chat functionality of the Reconciler. In addition, in the R condition, subjects were trained on how to use the communication interface of the system. In R, subjects were given examples about how the system automatically recognizes terms, how to introduce them directly, and how to check for the definitions in the screen. All subjects were told to analyze advantages and disadvantages of the different options that the proposed "Personal Organizer" software would include. Each category (e.g. Interface) involved discussing and selecting among the set of technical terms that they defined on Day 1. Groups had 35 minutes for the task. Thus, in the R condition,
G. Mark et al. / Reconciling Different Perspectives
31
groups used the communication interface of the Reconciler, and in the W condition, groups used the Reconciler interface without the translation mechanism (i.e. chat). The W groups could refer to their defined terms on paper from Day 1. Groups then filled out a questionnaire. 4 Results All R groups completed the task on Day 2, but three W groups did not finish the last topic. Most groups in both conditions chose the same options for the Personal Organizer. Due to the simple functionality, training was easy, and no one had difficulty using the system in either condition. 4.1 Use of terms and definitions An ANOVA showed no main effect of System Use for total number of terms used in the discussion F(l,36)=.08, p<.8, and not surprisingly, CTOs (mean=26.7, sd=9.6) used significantly more terms than Managers (mean=20.5, sd=8.7), F(l,36)=4.4, p<.04, with no significant System Use x Role interaction. 4.2 Messages used with the system We next compared the total number of messages sent by the groups. However, we noticed that some messages seemed to be artificially broken onto two lines, which is indicative of a frequent chat user. All messages that appeared to be broken were combined by two coders, using a criteria of: a) if two messages made up a complete sentence, b) if the two messages were not split up by the partner's message, and c) if the messages were contiguous in time, no more than 15 seconds apart. A total of 2,145 messages were exchanged for all groups. An ANOVA showed that those groups in the R condition (Mean=45.9, sd=22.0) exchanged significantly fewer messages than in the W condition (Mean=61.4, sd=16.2), F(l,36)=6.08, p<.02. There was no main effect of Role and no significant interaction.1 Thus, we found that using the Reconciler was associated with fewer total messages. Yet fewer messages could be due to different reasons. For example, perhaps there could be fewer clarifications about the meaning of the terms. To explore this further, and to test our hypotheses, we developed a coding scheme.
Note that less than 5% of the messages were split, and an ANOVA performed on the original messages, before combining them, produced the same significant results.
G. Mark et al. /Reconciling Different Perspectives Task related: any comment related to the task—choosing the options, analyzing advantages and disadvantages, proposals. etc. Articulation messages Confirmation: agreeing with the response of the partner on a component to use, making sure both are on the "same page", e.g. let's go with it, OK, that's cool. Clarification: any comment that concerns asking partner about term to explain it, that indicates lack of understanding, and feedback after the term is explained. Explanation: any comment that concerns explaining what a term means, may or may not be in response to a question. Asking partner if they understand a term, or how it is used. Other messages: Group Interaction Opinion Solicitation: asking the partner's opinion about a term. Process; comment on the process, e.g. moving the process along, announcing a topic is completed, questions on process. Non-task related: any comments that were not related to the task or process, e.g. greetings. LOL, social comments. Table 1. Coding scheme used in messages.
There are many ways to analyze conversation, and for the purposes of testing our hypotheses, we chose to identify different types of arguments. We deliberately chose to code messages that we felt 1) were task-related, to test our first hypothesis, and 2) that concerned articulation, in order to test our second hypothesis. These latter messages were: Confirmations (as a way to check understanding), Clarifications, and Explanations of terms. We also coded other messages that were basic group interactions to see if other aspects of the group process were affected. These were: Opinion solicitation, Process, and Non-task related comments. Table 1 shows this coding scheme. The coding scheme was mutually exclusive and exhaustive. Two coders coded the messages independently, with 88% agreement. Where there were disagreements, the two coders discussed the codes until agreement was reached. Most errors were due to oversights, and it was straightforward to reach agreement in these cases. Table 2 shows the means and proportions of messages for each of the codes.
Means (s.d.) Code
Hyp. tested
System Use: R 23.7 (10.4)
Task
1
Confirm. Clarific. Ex plan. Total Articul.
2 2 2 2
7.2* (6.6) 1.0*(1.1) 1.3* (1.7) 9.5** (7.1)
Process Opinion Nontask
--
6.6(6.6) 3.9**(3.1) 2.4(3.7)
System Role: Role: Use: R CTO M 22.2* 54%* 29.0* (7.2) sd=(.15) (11-4) Articulation messages 10.7 (6.3) 16% 7.5 (5.2) 10.9*(4.6) 3.3** (4.1) 2%* 3.2* (4.21 .8** (1.3) 2%* .4**' (.7) 2.7* (4.4) 3.6**' (4.2) 21% 14.3** 11.9** 16.7** (7.9) (8.0) (7.1) Group inter*action COther messages: 6.3(5.1) 13% 6.6(3.6) 7.0(5.5) 9% 7.1** (4.2) 6.6**(4.4) 3.4** (2.9) 4% 2.7 (3.0) 4.0 (3.5) 3.6(4.2) System Use: W 27.4 (9.4)
Proportions System Use: W 45%* sd=(. 13)
Role: CTO 54%* sd=(. 15)*
Rote: M 44%* sd=(.l3)
18% 5%* 4%' 27%
14%* 2%**
21%* 5%** .06%'
11% 11% 6%
Table 2. Mean (s.d.) and proportions of coded messages for all conditions. *=p<.05, **=p<.01,' =System Use x Role interaction, p<.05. N=40.
6%' 21%
27%
12%
11%
7%**
13%** 4%
5%
G. Mark et al. / Reconciling Different Perspectives
33
H1: Articulation more explicit without System In our first hypothesis, we expected that articulation would be more explicit in the W condition. We were interested to compare the actual number of articulation messages that were sent in the 35 minute task period, so we combined the totals of the messages coded as Confirm, Clarification, and Explanation (see Table 2). We found more articulation messages sent in the W groups (mean=16.7, sd=7.1), than in the R groups (mean=9.5, sd=7.1), F( 1,36)= 10.1, p<.0032. The effect was highly significant. Thus, fewer messages concerning articulation were sent with the Reconciler, which supports our first hypothesis. H,: More task-related messages with System To test our second hypothesis, we compared the proportion of messages coded as Task (see Table 2). An ANOVA showed a significantly higher proportion of messages sent with the Reconciler (54%) that were task-related, compared to without the Reconciler (45%), F(l,36)=4.2, p<.05. CTOs (54%) sent a higher proportion of task messages than Managers (44%), F(l,36)=5.3, p<.03, with no significant interaction, F(l,36)=.17, p<.7. Thus, we found support for our second hypothesis, that a system that can aid people in communicating terms with shared meanings would increase the proportion of messages focused on the task. We were interested in how the Reconciler supported the shared meanings for the Managers in the task discussion, since this is central to the Reconciler use. Terms and their definitions (agreed upon in Day 1) used in task messages were counted by a coder. We considered both the definitions and technical terms that Managers used. Managers in R used a higher proportion of terms/definitions in their task-related messages (mean=52%, sd=.24) than Managers in W (mean=34%, sd=.13), t(18)=2.13, p<.05. With respect to the other coded messages, we found only significant differences with Opinion: Opinions were more likely to be asked without the Reconciler use (F(l,36)=6.6, p<.01), and Managers were more likely to ask CTO opinions than the reverse (F( 1,36)= 12.4, p<.001). There was no interaction. 4.3 Satisfaction of communication and confidence In order to determine satisfaction with the Reconciler use, we analyzed the results of a questionnaire that we gave at the end of Day 2, shown in Table 3. There were no significant differences in System use and Role on the questionnaire, so we report total means. The fact that no significant differences were found in understanding their role, or in understanding how to use the terms in communication serves as a check across conditions. In general, subjects were satisfied with their process, result, and in the R condition, use of the system. Everyone seemed satisfied in the communication task. In the questionnaire, we did not specifically ask about confidence in the task. However, we examined the qualitative nature of the responses. Two independent coders identified all responses that concerned Mack of knowledge or confusion'. Examples are: But I am a technology dummy so I don't know how dependable my answer is. This one is purely your territory cuz / still am kind of confused. Well, I don't know enough about this. Whatever you think best, dude.
Comparing the proportions of articulation messages yields results that are consistent, approaching a significant difference.
34
G. Mark et al. / Reconciling Different Perspectives
Nine of the Managers in the W condition expressed at least one statement of this nature, but only three Managers in the R condition did. A chi-square test shows this difference to be highly significant, X2(1)=7.5, p<.004. I understood how to use the terms in our comm. I liked working with my partner on the task. I understood the role I was playing. I was motivated to perform the best I could. I understood the meanings that the terms had for my partner. I found it useful to define the terms in Part 1 . I felt that our communication was productive. It was easy to express myself to my partner. It was easy to understand my partner. I was satisfied with our team process. I am satisfied with our team result. I would want to work with my partner again. R only: The messages from the system were easy to understand. R only: I found the system useful for comm. R only: I would want to use the system again. R only: It was easy to use the system for comm.
5.1 (.7) 5.3 (.6) 5.1 (.8) 5.3 (.7) 5.0 (.8) 5.2 (.8) 5. 1 (.6) 5.1 (.7) 5. 1 (.8) 5. 1 (.6) 5.0 (.6) 5.3 (.6) 4.9 (.6) 4.8 (.7) 4.9(1.0) 4.9 (.8)
Table 3. Questionnaire results from Day 2 of experiment, mean (s.d.). (l=strongly disagree, 6=strongly agree).
5 Discussion and implications In summary, we tested the effect that the Reconciler system had on articulation work. We found that the system effectively served to help people apply negotiated shared meanings in their communication. With the Reconciler, Managers used a higher proportion of definitions in their task-related messages. Managers used technical terms with and without the Reconciler, but they were used differently. Although Managers in W had the definitions on paper, they did not receive the defined terms from CTOs in context, i.e. in chat messages. They did in R (e.g. HTML [web language]). We argue that Managers who sent technical terms using the Reconciler (e.g. HTML) were more likely to understand their contextual meanings than Managers not using the Reconciler. Even though only a single instance of a definition was used in the chat by Managers without the Reconciler, this does not imply that they understood the technical terms. This is supported by the results of testing our first hypothesis. We found support for our first hypothesis, that groups without technology support exchanged more messages explicitly concerning articulation. We considered articulation messages to concern confirming the other, clarifying a term, and explaining a term. This result suggests to us that with the Reconciler use, articulation is implicit. When a group sends a term with the Reconciler, it has already been predefined, and we interpret the results to mean that by using the term in the communication interface, the concept of articulation is sent implicitly with the term. The users are "reminded" of their articulation work that they had already accomplished. The articulation is embedded in the message, so to speak. We found support for our second hypothesis, that a higher proportion of task-related messages were sent with the Reconciler. This result is interesting in relation to another: fewer messages were also exchanged using the Reconciler, compared to without the system. We
G. Mark et al. / Reconciling Different Perspectives
35
can explain the extra messages as being due to more articulation messages. Taken together, the results suggest that communication with the Reconciler was more task-focused, i.e. in the sense of utilizing the terms to discuss advantages and disadvantages of the Personal Organizer task, and less focused on clarifying, explaining, and confirming what the other had said. Moreover, three groups without the Reconciler did not even complete the task. One might be tempted to propose that we found fewer total messages in the R condition because more effort was required to include a term or definition into the message. The Manager needed to perform an extra step, i.e. to double click on the interface. However, the fact that we found no differences between System Use conditions in number of terms and definitions used argues against this. Thus, extra clicking on the interface is not a feasible explanation for the differences. There appears to be some inherently different behavior that results from being able to easily integrate definitions in communication, plus the common knowledge that the meanings are shared. Satisfaction did not differ among conditions. Users did not express difficulties in working with the system, or in understanding the definitions through the system. They agreed that the system was useful, and easy to use for communication. Thus, despite the extra effort of predefining terms, and in selecting definitions during the task, people were quite satisfied. 5.1 Implications of technology support for articulation Although terms were defined by all groups on Day 1, what we were really testing is how well they are applied in context. By receiving the definitions in the communication (e.g. web language=HTML), the Manager is understanding the term in her own context. The articulation is occurring as the Manager is reading the message. Without using the system, the Manager must do the extra effort to understand what HTML is, and how it fits into the message context. The system was used to offer a one-way translation, from technical to non-technical. However, we argue a two-way learning process occurred, although different for the two roles. For the Manager, she is learning what the term means, e.g. that HTML is web language. For the CTO, she is learning how the Manager understands the term in her context. In other words, the CTO can understand better how the term is made meaningful for the Manager. Further research can explore how one specialist may even change communication, as a result of knowing how a different specialist or non-specialist understands a term [9]. In our experiment, the CTO did not have to put in extra effort to make the terms clear for the Manager because of the automatic word completion. What surprised us was that sometimes the CTO would type in the term, realize a typo, and then retype the term. This extra effort was to make sure that the Manager received the definition, i.e. to ensure that they were both "on the same mental page" with respect to the concept. Perhaps the CTOs did this as much as they did because they reaped the benefits of having their partner understand. We had hoped that if a technical term were translated into a meaningful idea for the non-technical person, it would "empower" that person to use the terms and definitions more in the task discussion. Our results suggest that the Reconciler may support this. First, with the Reconciler, a higher proportion of the Managers' task-related messages contained terms. That is, we are seeing that the Manager used the terms and definitions more in the discussion of the task, compared to without the system. Second, the fact that fewer Managers in the R condition expressed "confusion" or "ignorance" suggests that they may have had more confidence in discussing the technical terms. Or, they may have been less confused. Managers asked more opinions without the Reconciler (see Table 2) which could also suggest that these Managers were less confident. There were limitations of the experiment. First, we were limited by the roles we designed. The translation was one-way. In future research we would like to investigate twoway correspondences, where both parties use specialized terms. Because subjects were
36
G. Mark et al. /Reconciling Different Perspectives
visually separated, and communicated by chat on Day 2, we believe that gender differences were minimized, yet we cannot rule out that fewer females in CTO Roles may have had an effect. 6 Conclusion The communication interface of the Reconciler that we tested is designed to promote shared meanings and reduce the effort in re-explaining and re-clarifying. Our results suggest that the system can indeed do this. We deliberately employed a process of having people articulate terms on Day 1, as this is a necessary component of collaboration. In this experiment, we focused on how the articulation can be applied in communication, e.g. as with distance collaboration. In future research, it would be beneficial to test how the initial articulation process can be better supported. Our view of the Reconciler is that it supports a learning process. Over time, we would expect that the shared meanings could be learned. We envision new functionality in which the Reconciler plays a reinforcing role, identifying discrepancies. Is it a benefit that less re-explaining is done using a system such as the Reconciler? We cannot say. In fact, it is more of a philosophical question whether communication is enhanced or impaired by reducing re-clarification. We argue that by reducing the repetition of previous articulation, the system is reducing the requisite effort in communication, and thereby is a benefit. Our view is that technology support should be conceived of as a guide to support articulation, and not as a replacement for articulation. For example, we can integrate the definition and communication interfaces to enable ongoing articulation. We are continuing to work on conceiving new ways for the interface to more seamlessly support articulation. References [I] Bowker, G. C. and Star S.L. (1999): Sorting things out: classification and its consequences. Inside Technology, W. E. Bijeker, W. B. Carlson, and T. Pinch (eds.), Cambridge, MA: The MIT Press. [2] Brown, J. S. and Duguid, P. (2000): The Social life of Information. Boston, MA: Harvard Business Press. [3] Carstensen, P.H. (1997): Towards information exploration support for engineering designers. In S. Ganesman (ed.) Advances in Concurrent Engineering (CE97), Oakland (Mi), 26–33. [4] Conklin, J.and Begemann, M. L. (1988): gIBIS: A Hypertext Tool for Argumentation. ACM Transactions on Office Information Systems, 6(4):303–331 [5] Divitini, M. and Simone, C. (2000): Supporting different dimensions of adaptability in workflow modeling. CSCW-Special Issue on 'Adaptive Worflow Systems', vol. 9, no. 3/4, 365-397. [6] Dourish, P., Lamping, P., and Rodden, T. (1999): Building bridges: customization and intelligibility in shared category management. Proc. of Group99, Phoenix AZ, New York: ACM Press, 11– 20. [7] Fischer, G., Lemke, A. C., McCall, R., and Morch, A. I.(1991):Making Argumentation Serve Design. Human-Computer Interaction, 6, 3&4 , 393–419 [8] Flint, W. (2001): Plant your data in a ternary search tree (create an English dictionary that checks spelling and matches words as you type). Java World, February, 2001. Available at http://ad.doubleclick.net/ad l/idg.us.idgnet.javaworld/articles;pos=top3z=468x60;tile=l;ord=844786? [9] Fussell, S. R., and Krauss, R. M. (1992): Coordination of knowledge in communication: Effects of speakers' assumptions about what others know. Journal of Personality and Social Psychology, 62, 378–391. [10]Karsenty L. and Brezillon P. (1995): Cooperative problem solving and explanation. International Journal of Expert Systems With Applications, 8(4), pp.445–462 [ I I ] Mark, G., Fuchs, L., and Sohlenkamp, M. (1997): Supporting groupware conventions through contextual awareness. Proceedings of ECSCW'97, Lancaster, England, Dordrecht: Kluwer Publishers, 253-268. [12] Schmidt, K. and Simone, C. (1996): Coordination Mechanisms: towards a conceptual foundation for CSCW systems design. CSCW, 5, (2-3). 155-200.
G. Mark et al. / Reconciling Different Perspectives
37
[13]Simone, C., Mark, G, and. Giubbilei, D. (1999). Interoperability as a means of articulation work. In D. Georgakopoulos, W. Prinz, and A. Wolf (eds.) Proc. of ACM WACC'99, San Francisco, New York: ACM Press, 39-48. [14]Simone, C. and Sarini, M. (2001). Adaptability of Classification Schemes in Cooperation: what does it mean? Proceedings of ECSCW'Ol, Bonn, Germany, September 18-20, Dordrecht: Kluwer. [15] Trigg, R. H., Blomberg, J. and Suchman, L. (1999): Moving documents collections online: the evolution of a shared repository. Proceedings of ECSCW99, Copenhagen, Kluwer Academic, 331–350. [16]Winograd, T. and Flores, F. (1986): Understanding Computers and Cognition: A new foundation for design. Addison Wesley, Reading, MA 1986 [17] Wulf, V. (1997): Storing and retrieving documents in a shared workspace: experiences from the political administration. Proc. of INTERACT'97, London: Chapman & Hall, 469-476. [18]Zacklad, M., and Rousseaux F. (1995): Modelling Co-operation in the Design of Knowledge Production Systems: the Madeln'Coop Method - An example in the field of C3I systems, in Proceedings of COOP'95, Juan-les-Pins, 24–27 January 1995, pp 1-19.
Cooperative Systems Design M. Blay-Fornarino et al. (Eds. I IOS Press. 2002
Growing Networks: Detours, Stunts and Spillovers Margunn Aanestad, Ole Hanseth Department of Informatics, University of Oslo, P.O. Box 1080, NO-0316 Oslo, Norway
Abstract: Designing and developing technologies used to support distributed cooperation across organisational boundaries is challenging. Traditional design strategies do not appear to be suited to the situation at hand, as we will have to deal with vague visions rather than defined requirements, and most importantly, the development has to happen in a complex setting with a multitude of stakeholders, over whom no single actor has control. This paper describes the exploratory and iterative development of a telemedicine infrastructure in a hospital department. The need to enrol partners for development as well as for use required detours from the original goals. In this case the detours took the form of stunts, individual transmission events that were not parts of a grand plan, but which emerged in an ad hoc manner. Even if the main objective of the stunts was on an immediate nature, related to their successful accomplishment, the experience provided spillover knowledge that accumulated and proved valuable. Thus, far from being just a secondary and nonproductive activity, the stunts provided the core learning situations on the project. Keywords: CSCW, video conferencing, field study, design methodology, video communication.
1
Introduction
New multi-media information and communication technologies are widely assumed to enable a wide range of new medical services, usually labelled telemedicine. Telemedicine services can be developed based on a wide variety of technologies, from video conferencing technologies supporting synchronous communication to email technologies supporting asynchronous communication. The use areas also varies widely; telemedicine services may support diagnostic as well as treatment processes, they may be used in emergency as well as non-emergency situations, and in non-patient-related activities like teaching and meetings. New services may cut across levels and boundaries; they may support work and communication both within a medical profession and between different professional groups, as well as communication and collaboration between primary care centres and hospitals, and between hospitals on a national or global scale. Most users should share the specific technological solutions, for instance, standard infrastructural technologies and services like e-mail and real time video conferencing. In addition functions designed and implemented because of specific needs of larger or smaller groups are needed. This implies designing new technological solutions; it is not just about organizational implementation of e-mail or video-conferencing solutions. If these grand visions shall become reality, an extensive, global and highly heterogeneous information infrastructure has to be built. The challenges associated with this are addressed in this paper, and this should have relevancy for other attempts at design of similar infrastructural technologies to support distributed communication and collaboration. In such cases both the technical base as well as the user groups and actors may be highly heterogeneous, and the design should allow for novel, innovative and unanticipated use. The first challenge is to choose the most appropriate development strategy for these kinds of projects.
M. Aanestad and O. Hanseth / Growing Networks
2 Challenges to design 2.1 Design strategies Should we adopt the most widely used design strategy and select a method based on the waterfall model, i.e. first specifying the user requirements, then designing the system, and finally implementing it in the user organization? Obviously this does not make sense. Working out user requirements for an infrastructure that should support the work of and collaboration between all medical doctors in the world and then implement it in a big bang is impossible in practice. We will at least have to adopt an evolutionary approach. Should we develop a prototype first and then the full system [1]? Should we follow the spiral model [2] and identify the most critical risk first, resolving this and selecting the next one? Or should we follow the traditional path of development of communication technologies, which is through processes leading to agreement about standards. This is for instance, the path chosen by of the largest standardization efforts for IS in health care [3,4]. The problem with all these IS methodologies as well as the telecommunication strategies for defining standards, is that they focus on the development of one shared, unified solution for all users which are designed by one body (group or project) in a coordinated manner, and managed from one central point. This is simply impossible for several reasons: the complexity of the solutions to be designed, the novelty of the work practices into which the technological solution will be integrated, the unpredictable dynamics of the health care sector (partly to be generated by telemedicine technology itself), the high degree of conflicting interests among various groups of health care professionals, etc. These issues (i.e. complexity, lack of knowledge, dynamics, conflicts) are also present in ordinary IS development projects, often making the traditional IS development methodologies inappropriate. We will in this paper primarily discuss issues and challenges related to two of these aspects, i.e. complexity and novelty. 2.2 Complexity and novelty The general strategy for dealing with complexity is to use modularization, i.e. split the overall system into modules, implement the modules individually, and integrate them. In case of telemedicine (or other large scale networks or infrastructures) this means that we are designing solutions for specific use areas, user groups, or new medical services separately. We have to do this without having an overall design, which is usually supposed to ensure that the individual pieces will fit nicely together. Then the only feasible approach is to let the actors design their own solutions, but not independently, only in collaboration with partners so that everybody develop solutions that fit their partners'. The actors can do this by selecting one of - or better: combine - the following two strategies: The first strategy contains the following steps: 1. finding out what kind of telemedicine services and solutions they want to establish, 2. finding partners, 3. make solution in collaboration with partner(s), implement and use it. The second strategy is to 1. check out what potential partners are planning
39
40
M. Aanestad and O. Hanseth /Growing Nerworks
2. 3.
join forces with those most relevant, and collaborate about the development and implementation of a shared solution.
A second major challenge is to design for novel and unanticipated use. Before we have some experience with the use of a new technology, we cannot specify requirements for general solutions. This may be a general feature of introducing generic and customisable network technologies into complex work practices [5]. Communication technologies are open and generic in the sense that the usage is not very much specified by the technology designers and producers. This means that the usage must be designed when the technology is incorporated into the specific work practices, e.g. surgery. This is more than merely designing the technological solution. Many personnel categories need to collaborate around the technology, i.e. also in the design of the new work processes. This can only be done during an experimental process related to real work practices. 2.3 Key concepts: Detours, stunts and spillovers The challenges from complexity and novelty need to be handled, and services and applications therefore must be developed iteratively and in cooperation with other partners. If partners with similar interests are hard to find, or for other reasons, like lack of knowledge or funding, it will often not be possible to approach the primary goal directly, and a detour will be necessary. The concept of detour (from ActorNetwork theory) emphasises the need to form alliances with other actors in order to be able to act, and consequently holds that action is a collective achievement. In order to enrol other actors, one needs to translate their interests, as well as one's own (the original goal). This necessitates a detour, and the consequence of the alliance is thus often a deviation from the original plan. A citation from Bruno Latour may illustrate the concept of detour: "... if the accomplishment of the agent's goal is interrupted for whatever reason (perhaps the agent is not strong enough), then the agent makes a detour, a deviation.... Agent 1 falls back on agent 2... and a third agent emerges from the fusion of the two. The question now becomes which goal the new composite agent will pursue.... [6: p. 178] In this paper we tell a story of a telemedicine project with a focus on the detours the actors made. We argue that the detour of creating telemedicine "stunts" (singular transmission events) for other partners was necessary and useful. The word "stunt" in colloquial use are most commonly related to stuntmen or -women; stand-ins that do a famous actor's physically demanding and dangerous parts of the acting. A stunt denotes a demanding and somewhat risky achievement, as well as a show-off of ability. In this paper we use it to denote a unique, singular telemedicine event, for example a live transmission of an operation to a large medical conference. These kinds of events are highly visible, one-off events with a short-term horizon. Within telemedicine it is commonly agreed that one of the challenges is going from the phase of such stunts and limited projects, to the everyday routine utilisation. We will argue, however, that the stunts were necessary for exploratory and evolutionary development in this context because they provided concrete use situations where using the technology provided immediate benefit without jeopardizing the primary medical activity.
M. Aanestad and O. Hanseth / Growing Networks
In Steinmueller's description of the role of the knowledge infrastructure in the early computer industry [7], the concept of spillover (from the field of economy) is used to describe the unintended dissemination of knowledge, which he claims was a significant stimulus for the innovation, competition and technical progress that happened. In this paper we use the concept to emphasise the unintended and unplanned benefits and consequences of the stunts. A stunt was usually staged for a specific purpose (e.g. to contribute to a professional meeting), but in the process of performing the transmissions, experience was gained e.g. with the technical solutions and with logistics. Experience from the specific transmissions "spilled over" and became generally useful knowledge about technical solutions and their use beyond the particular setting, and thus the stunts provided additional benefits beyond what was expected and planned for. 3 Method We present the empirical material through a chronological narrative that describes selected parts of the history of the Interventional Centre at Rikshospitalet (Oslo, Norway), related to telemedicine activities. The term "telemedicine" in this paper should be taken in its broadest sense as denoting use of communication technology within a medical context, including local transmissions and non-patient-related use. The material was gathered mainly through participant observation by the first author through three years (1998-2001) and through interviews with the head of department, the technology research coordinator, and the technical consultant on the history of telemedicine activities at the centre. Both authors were project members during the telemedicine project described, where our main tasks were related to logging of activities, evaluation and reporting. Relevant documents, e.g. plans for the centre, yearly reports, published papers, project documents and transmission logs have also been studied.
4 A History of Telemedicine Activities at the Interventional Centre This description does not pretend to be an explanation of driving forces or success factors for development and implementation. The material should also not convey the presence of a master plan, neither on behalf of the Interventional Centre nor any other actor. No matter which intentions and agendas an actor has, it is unable to see the whole picture and can only decide the next few steps. Therefore the focus of this paper is a description of one actor's (the Interventional Centre) navigation in a complex and changing world, and on the actual choices and actions that have been undertaken, rather than an analysis of the actors' strategic intentions and hidden agendas. 4.1 In the beginning... The Interventional Centre was established in 1996 at Rikshospitalet to do research and development on interventional radiology, image guided and minimal invasive procedures. Interventional radiology describes the radiology that goes beyond a purely diagnostic examination to also include treatment (i.e. an intervention) when the patient is on the examination table. Minimal invasive surgery is also called
41
42
M. Aanestad and O. Hanseth / Growing Networks
"keyhole surgery", and the idea is that instead of making a large incision in e.g. the abdomen in order to get access, small incisions are made and the surgical instruments are entered through these ports. The surgeon gets the visual feedback from a video image from the operation field. Experiences from similar research centres indicated that one could expect lots of guests, who preferably should not enter the operating rooms. This was the reason behind a wish to establish a local transmission facility where the guests could view the rooms and watch the medical procedures without entering. With the assistance from an external consultant firm, the cabling for this was included in the initial building plans for the centre. The room was equipped to provide two-way interaction with the team inside the operation room and with the possibility to select among a variety of image sources. 4.2 Looking outwards - thinking telemedicine The hospital's managing physician wanted videoconferencing facilities at the hospital, and the Interventional Centre was the only department interested in and willing to host and run the videoconferencing studio. ISDN videoconferencing equipment (384 kbit/s) was hired and later purchased, and the video room worked as the whole hospital's video conferencing studio. Several other departments had previously performed local transmissions of images and sound to meetings, as well as irregular telemedicine transmissions. The presence of the local transmission facility in the videoconferencing studio spurred thoughts of exploiting it for telemedicine. The local transmission network was connected to the equipment, and several "virtual tours" of the centre was conducted for remote viewers, mainly cooperating R&D centres around the world. Telemedicine was viewed on one hand as a tool that could be of benefit for the communication with cooperating partners in the R&D activities. But the centre also had as its mandate to do research on different tools for medicine and acquire competence on new technologies, and thus telemedicine was also seen as a research object in itself. A natural starting point for such research was to focus on the feasibility for the available ISDN technology for transmission from the procedures performed locally (radiological and videoscopic procedures). ISDN transmission at 384 kbit/s gave visible degradation of live video due to heavy image compression. The image quality was considered inadequate for image-guided procedures like minimal invasive surgery, and other technological alternatives were explored. Some potential and expensive broadband technologies were satellite transmissions or ATM (Asynchronous Transfer Mode) networks. In 1997 and in 1998 there were some of attempts of cooperation with technology providers (satellite companies and a telco) about a closed TV-channel for health care, but these projects did not take off. 4.3 Looking for partners The main partner in the early years was to be Ulleval hospital, the other large hospital in Oslo. All the three persons interviewed at the Interventional Centre, as well as one of the senior surgeons maintained personal contacts with central surgeons at Ulleval hospital. The persons at Ulleval were involved in a regional telemedicine project, and they wanted to test out different technological alternatives before decisions were made. The project had a focus on broadband technology so as to facilitate surgery, which was seen as the most demanding application in the network in terms of demands to bandwidth, image quality, and reliability. At the
M. Aanestad and O. Hanseth / Growing Networks
same time the telecom provider Telia (Stockholm, Sweden) was trying to get market share in Norway as the deregulation of the telecom market was approaching (January 1st, 1998). Telia was one of the potential network providers for the regional broadband network and had offered Ulleval a technology demonstration. Telemedicine naturally requires both senders and receivers, and the Interventional Centre was a natural partner for this demonstration in terms of medical content (video from minimal-invasive surgery). This resulted in a trial set-up of a broadband network (ATM) between the Interventional Centre and Ulleval hospital for 14 days during the summer of 1997, where the equipment and network access was provided by Telia. The tests were promising with regard to the technology's feasibility, and the cooperation was continued and turned into a formal project called Development of Interactive Medical Services (DIMedS). The project partners were the Interventional Centre and Ulleval hospital, with Telia as provider of network access, and Ericsson as equipment supplier, and with the Department of Informatics at the University of Oslo as a research partner. The activities in connection with this project account for the majority of telemedicine sessions and experience in the period of interest, and some of the project activities will be further described. 4.4 The activities of the DIMedS project The first sub-project started in September 1998 and was focused on using broadband technology for distant learning in surgery and radiology, but the actual activities expanded outside this use area [8]. A wide variety of transmissions were performed (meetings, demonstrations, seminars, discussions, live operations) for different groups, and even more was planned but not executed. Part of the reason for this drifting [9] was the difficulties with summoning the intended receivers (surgeons) at given points of time, as it was difficult to coordinate the work schedules of different institutions. In order to utilise the network access and video digitising equipment fully, the scope was expanded. Before the DIMedS project, the thoracic surgery department's meeting room had been connected to the Interventional Centre's video room, and thus the thoracic surgery department could utilise the ISDN facility to transmit from or receive to their own meeting room. Conversely, the Interventional Centre got wired access to this other meeting room. This was convenient on occasions where the group of viewers was too large to fit into the local video room (e.g. a class of students). A transmission from the Interventional Centre's operation rooms could then easily be forwarded to the Thoracic Surgery department's meeting room. This facility was used extensively in the DIMedS project. In order to enrol partners in the telemedicine activities, the Interventional Centre offered to arrange several sessions that would demonstrate the technical possibilities, ranging from whole-day regional seminars to half an hour's lunch meetings. At one instance early in the project (October 1998) the large weekly staff meeting at Rikshospitalet were conducted on-line with Ulleval hospital's staff meeting, with some speakers at each site. As the two hospitals were well known for their competitive attitude towards each other, this was seen as remarkable and attracted quite a lot of attention. The session was also transmitted to a nearby hotel hosting the yearly national telemedicine conference. This was the first high profile session where the equipment and network of the DIMedS project was used. The planning and support work was undertaken by the Centre's staff in cooperation with the technical consultants, and it provided valuable experience.
43
44
M. Aanestad and O Hanseth /Growing Networks
Other major high-profile transmissions were to Sweden's yearly medical conference (Rikstamman) in November 1998, and to a minimal-invasive surgery conference at the Sechenov Institute in Moscow (where the founder of the Interventional Centre had contacts) in April 1999. In addition there were several smaller transmissions. One significant event was a transmission for ear-nose-throat (ENT) specialists where two regional meetings were coupled together, one at Ulleval and one at Rikshospitalet. This was the first instance where the support technicians handled the planning and set-up of the technology on their own, and only needed to borrow some pieces of audio equipment from the consultant. Later all the equipment needed was acquired, and the technicians would do everything themselves. To prepare for this event the technicians carried out extensive planning. The lecturers were contacted in order to find out what technical aids they would use for their presentations. A cabling plan was sketched, indicating how the required equipment (e.g. VHS players, video projectors, laptops, camera on lecturer, monitor with incoming image, microphones and loudspeakers) would be connected to the network. Also the responsibility for operating these pieces of equipment was distributed among the technicians on a sheet stating the time schedule and detail for each speaker's presentation. The setup for this session served, with modifications, as a kind of template for later transmissions from lecture halls of this size. Later in the project there were arranged weekly lunch lectures for radiologists, which gave experiences from novel use areas (transmission of radiological images). Also lectures for undergraduate students, live operations and other procedures for post-graduate surgery students, and several demonstrations for non-medical personnel (newspapers, TV-channel, visitors) were performed. Some of the transmissions were forwarded onto the regional network that Ulleval was connected to as well. The expansion of the DIMedS project's activities was done in order to utilise the equipment fully, but also to analyse the feasibility of the present technology for different use areas. The learning process through gradual expansion of use areas is further described elsewhere [5]. Here we want to emphasise the growing experience with how to organise and perform production work in different settings. Much of the planning work (e.g. plans for cabling, placement of microphones and cameras) was very specific for the actual transmission, and most of the equipment would be installed before and dismantled after the transmission. Still the experience proved valuable for later planning work. For example, experience was gained with the suitability of a particular set-up of equipment for a particular kind of transmission (e.g. How many free microphones are needed for a meeting where 30 persons participate? How many support technicians is required for camera control, mixing and monitoring? How much time does it take to install and troubleshoot the equipment in a meeting room? Which devices are problematic and should be duplicated?). As more demanding productions were undertaken, more equipment was purchased and the set-up of the video room became more complex. This led to a more systematic approach to management of equipment, standardisation of technical set-up, labelling of cables and connectors etc. A crucial resource for this expansion of activities to be possible was the availability of temporary workers as support technicians. The technicians were conscientious objectors working for 14 months at the hospital as an alternative to army service, and an active recruiting policy resulted in employment of motivated and technically skilled personnel in this function.
M. Aanestad and O. Hanseth / Growing Networks
4.5 Events beyond control, changing environment The Interventional Centre was located in temporary premises while the new Rikshospitalet was planned and built. This precluded large investments in technical infrastructure that would not be moved. The original plans were to move to the new premises during the winter 1998/1999, but this move was postponed several times, and it eventually occurred in May 2000. The time scope of the first sub-project in the DIMedS project had to be expanded, both due to delayed start and in order to exploit the time until the hospital actually would move. When this sub-project was ended (June 1999) no further sub-projects were initiated, as the date for the actual move was still undecided. The overall DIMedS project was discontinued in the autumn of 1999 even though the initial plans were a three years project (until 2001). The industry partners' hesitation to continue the project may have several explanations, but one significant factor at this point in time was an expected merger between the Norwegian and the Swedish telcos, Telenor and Telia (the project partner). This led to postponing of strategic decisions regarding telemedicine activities within Telia. Although the attempted merger eventually failed (autumn 1999), the project was not reactivated. As telemedicine has not been a core activity, the department has had the time to wait and see while going along with its other activities. When the DIMedS project was discontinued, no alternative broadband connection was in place, and the telemedicine activities were restricted to occasional ISDN transmissions. The planning for the communication infrastructure at the new Interventional Centre has been based on experiences from the DIMedS project, and the technical set-up has been expanded and changed after the move into the new premises. The control room now functions as the technological hub for external and internal audio/video transmissions for the whole Rikshospitalet, and several departments use the facilities frequently.
5 Cultivating Networks As we argued in the Introduction, open and unspecified network technologies call for evolutionary, prototyping approaches to design and development. Our description of one such process clearly shows the open-ended and unplanned character of the process. Telemedicine was not an intended activity of the Interventional Centre from the start, and the way forward has been planned one step at a time. This does not mean that the process was random, neither that it was passive. 5.7 Detours The telemedicine activities reflects an ad hoc exploitation of opportunities that was considered relevant and interesting: "We have used possibilities that we have seen, and there have also been many that we haven't managed to exploit. This has not been a programmed or planned process." (Technology Research Coordinator) "We had many antennas out, we were open to many things and we exploited possibilities." (Head of Department)
45
M. Aanestad and O. Hanseth / Growing Networks
As telemedicine implies cooperation, a shared vision and common activities (e.g. services and applications) were central. This happened through a combination of the strategies mentioned in section 2.2., both cooperation with partners with similar interests (mainly Ulleval Hospital) and also through mobilisation and cooperation with "different" partners, both within and outside the hospital. To mobilise other partners was not trivial, however, and required a detour through arranging stunts that were not directly relevant to the Interventional Centre's primary goals. Usually medical personnel will prioritise the immediate medical concerns higher than participation in development projects with long time horizons. Any non-primary activities have to provide immediate benefits. The short-term time horizon of the stunts (no large commitments), the reduction of the "costs" involved by the availability of support personnel, and the immediate benefits in augmenting existing work (e.g. planned meetings) were attractive. The stunt detour both provided benefits for the participants, but in addition they served as an opportunity for learning and development for the initiators. They are thus a successful translation of both parties' interests. 5.2 Stunts The stunts (the singular transmission events) were planned in an ad hoc-manner, seizing emerging opportunities to ally with potential partners. Typically the event was scheduled, and the partners agreed on a date and the content and scope of the transmission. Then the scheduled event would serve to mobilise the local organisation (the participants and the support personnel). The necessary amount of planning, preparations and tests were performed, with the aim to manage the performance well. Some partners were interested and able to cooperate, and a transmission could be planned rather directly. The learning focus would then be on the technology's quality and feasibility for this specific use area and on how to organise the support and production work. Other partners were not convinced about the benefits of the technology, and an additional detour was necessary to enrol them. This was attempted through demonstrations of the technology's potential, and the stunts served as such a demonstration for potential users. "We need to tell the clinicians about what is possible to achieve. Ideally, it is the clinicians that should come with a definite need, a problem that should be solved. Here however, we need to tell them what is technically possible to achieve. When we demonstrate the possibilities, they get ideas about what this technology can be used for." (Technical Consultant) Partners and stunts could not be chosen arbitrarily, but had to fit with the fundamental (or eventual) goal. An added restriction was that in order to attract partners in general, and in particular the sceptics, the ,,useful" component of the stunt should be strong. This usefulness is related to common professional interests, e.g. to cooperate around a given case, or to disseminate knowledge. The importance of this professional content is illustrated in the Head of Department's reflections on choice of partners: "Ulleval Hospital was the natural partner for our cooperation, because we both wished the same thing. Even though Tromso has organisation and structure in place, which Ulleval didn't, Tromso did not have the same needs as we did, so it was not a natural partner, other than for demonstrations." (Head of Department)1 1 The National Centre for Telemedicine, which is located in Tromso, has mainly focused on low cost, mass technologies and telemedicine applications for primary and secondary health care.
M. Aanestad and O. Hanseth / Growing Networks
47
The high visibility of a stunt made it risky, as a failure could reinforce opposition and scepticism. A strategy that focused on high technical quality was paramount, both to enrol the allies in the first place and to avoid loosing them: "I have always had a focus on ... the user interface and the quality of the equipment. We should go for quality, and not compromise on this. If the technology does not deliver high enough quality, people will loose interest. Both the use value and the educational value will disappear. " (Technical Consultant) The recognition of this is also the basis for the strong focus towards a well-performed session, and explains the large amounts of manpower used for planning and execution of these events (more than 400 hours through the most intense 9 months, or above 50 hours per transmission at some instances). 5.3 Spillovers The stunts were demonstrations of broadband network technologies' potential for several different use areas. This is a development of a general rather than a specific technology, and it would involve users with different requirements to telemedical cooperation. Such an approach therefore required an open and flexible strategy as regards technology. It should allow easy change, reconfiguration or extension: "...I've focused on modular technological choices. One should not be restricted to one specific system. It should always be possible to build on, replace or change the technical equipment. Stepwise improvements are central." (Technical Consultant). The approach also requires quite a lot of support work, both technical and logistic support, which was undertaken by the staff at the Interventional Centre in cooperation with the partners. The staging of the stunts led to a cumulative and incremental growth of experience. This learning relates to the use areas' requirements, the different partners' strength and weaknesses, the organisation of production and support work, the technology set-up, and the requirements to mediated communication. This specific experience contributed to a knowledge base that could be drawn on also for entirely new use areas. Based on practical experience, the technology set-up was modified and standardised, and planning, production and support work was routinised (logistics, cabling plans, distribution of tasks). The knowledge generated in different and specific settings (stunts) becomes accumulated and "spills over" to become useful also beyond the particular setting. Some of the stunts were high profile and as such risky (a lot of people were involved, and a technical or logistic failure would be public), but there was no simple correlation between size and importance in terms of learning benefits. The complexity of the production were at its highest for some of the small-scale transmissions, depending on the medical content's demands to technology and the interactivity of the session. 6 Conclusion The growth process of the Interventional Centre's telemedicine activities has not been a pre-planned and linear process of implementation. It has been unplanned and exploratory, constantly open to evaluation of new opportunities, technologies and
48
M. Aanestad and O. Hanseth / Growing
partners, inspired by (unclear) visions and exploiting the synergy effects of technology and a wide contact network. The complexity of the task, as well as the novelty of the usage, dictated an evolutionary and iterative approach. The active creation of events ("stunts") provided a way to perform this exploratory development. Thus the stunts, far from being just a secondary and non-productive activity, provided the core learning situations in the development project. Developing technologies with infrastructural characteristics may have characteristics similar to the case described here. They may be large in scope, highly heterogeneous in terms of technology, use areas and users, and they may be expected to support novel, innovative and unanticipated use. In such situations characterised by high complexity and novelty, there are no shortcuts. The shortest (in fact the only possible) way to reach your goal may be through detours. Here the detours of stunts was important, as it allowed other partners to become involved in activities that all parts benefited from. The actual "route" for the detour is not possible to generalise on, but having a short-term useful result might be advocated as a general strategy for building a network technology where the challenge is to get started and cooperate with others. We could argue that the "stunt strategy" might be beneficial also in other cases if the technology and the setting are of a comparable nature, i.e. an interactive, open and customisabie network technology to facilitate novel collaboration between (relatively) independent partners.
7 Acknowledgements This work has been made possible through funding from the Norwegian Research Council's grant no. 123861/320. The input from the staff at the Interventional Centre and the partners in the DIMedS project is gratefully acknowledged.
M. Aanestad and O. Hanseth / Growing Networks
49
References [1] Budde et al. (1991), "Prototyping - An Approach to Evolutionary System Development". Springer Verlag. [2] Boehm, B. (1988), "A Spiral Model of Software Development and Enhancement", IEEE Computer, May 1988, pp. 61-72. [3] De Moor, G. (1993), "Standardization in Health Care Informatics and Telematics in Europe: CEN TC 251 Activities". In G. De Moor et al. (eds.): Progress in Standardization in Health Care Informatics, IOS Press. [4] McDonald, C. (1993), "ANSI's Health Information Planning Panel (HISPP) - the purpose and progress". In G. De Moor et al. (eds.): Progress in Standardization in Health Care Informatics, IOS Press. [5] Aanestad, M. and Hanseth, O. (2000), "Implementing Open Network Technologies in Complex Work Practices: A Case From Telemedicine", in R. Baskerville, J. Stage, and J. DeGross (eds.) "Organizational and Social Perspectives on Information Technology", Kluwer Academic Publishers, Dordrecht, The Netherlands, pp.355-369. [6] Latour, B. (1999), "Pandora's Hope. Essays on the Reality of Science Studies", Harvard University Press, Cambridge, MA. [7] Steinmueller, W.E. (1996), 'Technological Infrastructure in Information Technology Industries", in M. Teubal et al. (eds.), Technological Infrastructure Policy", Kluwer Academic Publishers, The Netherlands, pp. 117-139. [8] Hanseth, O. et al (1999): "Project report: DIMedS- Development of Interactive Medical Services, Distant learning in surgery and radiology using broadband networks". Telia, Norway. [9] Ciborra, C. (1996), "Introduction: What do Groupware Mean for the Organization Hosting It?" in C. Ciborra (ed.) "Groupware and Teamwork, Invisible Aid or Technical Hindrance", Wiley & Sons, Chichester, UK.
This page intentionally left blank
Tools and Techniques for Cooperative Systems Design
This page intentionally left blank
Cooperative Systems Design M. Blay-Fornarino el al. (Eds.) IOS Press, 2002
53
Modelling Cooperative Work: Chances and Risks of Structuring Thomas Herrmann, Marcel Hoffmann, Gabriele Kunau, Kai-Uwe Loser Informatics & Society , University of Dortmund, Germany {herrmann, hoffmann, kunau, loser}@iug.cs.uni-dortmund.de Abstract. We found evidence in several cases that semi-structured modelling methods are quite helpful to model cooperative work. How can our positive findings with modelling be related to those publications which emphasize a sceptical view on representing work with models and list a number of problems, risks and inappropriatenesses with respect to the explicit representation of cooperative work? In this paper we evaluate the chances and risks of modelling based on three case studies. Various arguments about impacts of modelling are assigned to a framework differentiating the following three levels: A) perceiving and reflecting of structure, B) explicating, articulating and negotiating structures and C) contributing to the developing of structures. Within this framework we discuss possible chances and risks from literature and from our case studies. We argue that improved methods, notations and tools can help to reduce or even avoid the risks and use the chances for cooperative work.
1
Introduction
We found evidence in several cases that semi-structured modelling methods are quite helpful to model cooperative work [10]. This semi-structured method avoids the equalization of modelling and formalization (e.g. [3], referring to [15]). However there remain some central questions: How can our positive findings with modelling be related to those parts of the literature which emphasize a sceptical view on representing work with models (e.g.[14]) and list a number of problems, risks and inappropriateness with respect to the explicit representation of cooperative work. And furthermore, if the answer about the usefulness of explicit representations is not just "yes" or "no", what are the criteria according to which we can compare the benefits with the risks and costs of developing explicit models of cooperative work - and how relate these criteria to the literature as well as to empirical case studies. To answer these questions in this paper, we start with a framework (fig. 1) which differentiates between three levels of structuring. The three levels are helpful to assign the different orientations of the arguments on explicit representation of cooperative work which are given in the discourse: A. Individual mental models are developed in interaction with the perception of the structure of parts of the reality (such as work processes, organizational units) B. Mental models are made explicit and structure is developed in interrelation with the method of representation and by combining the viewpoints of different individuals C. Implicitly and explicitly represented structures are reproduced or reformulated in practice and, on the other hand, guide and constrain practice, too. Purposefully attempting to shape reality with recourse to an explicit model gives rise to special chances and risks.
54
T. Herrmann et al. / Modelling Cooperative Work
Figure 1: Three Levels of Structure According to this framework, structuring and modelling are highly interrelated. We use the term "structure" to refer to the mesh of relations between the elements of a system and between the properties of these elements Ad A) The activity "structuring" is already taking place when a vague mental model of a system emerges, since the structure of an entity can only be relevant in interrelation to an observer. From the viewpoint of the general systems theory, every system has structure which determines its borders, unity and identity. In the Seventies and Eighties, system theorists such as Maturana [13] or Luhmann [12] supported the constructivist approach by emphasizing that any structure is always the result of an interrelationship between an observer and an observed system. The supposition that every type of system is structured is also supported by Giddens' structuration theory [7], which is widely adopted by CSCW researchers. Human actors build theories about the systems [1] which we call mental models of the system's structure. However, also in the case of these mental models, too, Bannon's [2] question of "what can, in principle, be captured in any model of the work process" has to be applied. In the special case of cooperative work it might be an appropriate assumption that the overall structure of the cooperation is not replicated in any of the individual participants' mental models. This concern can be related to Suchman's ([19], ref. in [17] p.323) statement that the procedural structure of organizational activities is the product of the orderly work of the office rather than the reflection of some enduring structure found behind the work. According to this, the overall structure of a cooperative work process is only present in the work itself or - in other words - human action cannot be described by "plans" that are separated from the action itself [21]. Consequently, the issue of whether mental models can represent a relatively permanent reflection of the structure of cooperative work is questionable. We will have to study these problems in the light of our case studies by analysing whether modelling contributes to an higher transparency of the overall organization of cooperative work and how far workers can be supported to continuously reflect on their
T. Herrmann et al. / Modelling Cooperative Work
55
own practice. These potential benefits can only occur in the context of the next level when models are made explicit. Ad B) Schmidt and Bannon [16] claim that cooperation needs articulation work which supports the coordination of cooperation. Articulation work can be facilitated by explicit models of work and it can help to create such models. Thus it becomes obvious that it might be sensible and supportive to make mental models of the structure of cooperation processes explicit. However, explicit models share certain characteristics which give reason for sceptical considerations: • Explicit models are more permanent compared with the potentially more ephemeral mental models. • Explicit models are more standardized and generalized. They have therefore a greater distance to real work. Thus, it might be the case that the problems of work are ignored in procedural expressions modelling its structure ([17] p. 320, ref. to [20]). It is argued that the greater distance to the ongoing work, the more stereotyped are the models [23]. However, it can also be stated that the problems of stereotyping and idealizing disappear if the appropriate tacit background assumptions can be meshed with the models [17]. • Explicit models are the result of a political process: different individual perspectives are merged; certain aspects are sorted out. The developing of explicit models is always combined with selection and this is guided by interests [23]. Since there are different interests involved and related to representations, some actors might want to push their personal beliefs into the model. Others might want to keep their working practice a secret for varying reasons: it might be "... valuable that members of an organization know their own work in ways that others positioned differently in the organization do not" ([23], p 56). The problems with making models explicit heavily depends on the question whether the models are developed within the modelled system or whether they stem from outside ([23], [4]). The potential problems with explicit models do not provide sufficient reasons for totally neglecting the potential benefits of explicit work representations. Schmidt [17], who values the work of Suchman, states that her investigations do not provide sufficient insight in how standard procedures defined as pre-defined stipulations are applied in routine daily work. Thus, it is sensible to ask the question of under which circumstances it might be an advantage to develop explicit models. They might help to disclose the influence of different interests and they can serve as a basis for negotiation. Therefore we are investigating in our case studies the question of how far the models support dealing with conflicts, participation or continuous changing of work presentations in mind. The success of or problems with explicit models depends also on the modelling method being used: mainly the type of notation [8]. We are especially interested in graphical notations. Explicit models provide structure on two levels: the structure of the modelled reality and the structure of the representation itself. If the modelling method allows the modeller to consciously leave parts of the modelled reality unconsidered and to make this decision transparent in the model itself, we call the modelling method semistructured. All in all we agree with Suchman [23] that the work of modelling Figure 2: Criteria of case study analysis has itself to be reflected on.
56
T. Herrmann el al. / Modelling Cooperative Work
Ad C) Generally, the development of explicit models pursues a certain purpose which also reflects the interests of the participants being involved. Explicit models are often used to anticipate how the modelled system or process can be improved and to model the im proved constellation. With the model of the improved "to-be"-situation, managers and system designers try to shape and to change the structure of reality. The activities of these attempts as well as all unintended influences on the reality's structure have to be assigned to level C. The intentional attempts to reshape structure mainly serve the following purposes: • Guiding activities: Explicit representations are often developed with the aim of controlling human activity, especially the coordination of cooperative work [23]. However, it has to be taken into consideration that plans can be seen as a resource for situated action, but do not in any strong sense determine its course [21]. If this statement can also be applied to the role of explicit representations, it is questionable whether they can be used appropriately to guide human activities. To overcome this problem, Bannon and Schmidt [14] propose the differentiation between models as maps or scripts. While scripts describe exactly how a work procedure has to be carried out, maps only provide orientations if needed. Modelling methods should support this differentiation. Furthermore, it is stated that the guidance of human activity should never rely completely on explicit representation but should always be accompanied by a channel for informal communication ([16] ref. to [14]). • Designing technology: the designers of CSCW systems in particular try to build models of how cooperative work will be conducted with their system. The question arises [16]: "How can designers unravel the essential functions of the cooperative work relations to be supported as opposed to ephemeral or accidental cooperative work practices that may be observed"? The answer to this question leads to a more general and permanent model of a certain situation of cooperative work. However, all the problems with an explicit model, such as interest oriented selectivity or generalization, have a considerable impact on the technical system which is designed in accordance with this model. This impact is even more relevant if the system is used to control human activities as in the case of most workflow management systems. It is obvious that the problems with guiding human activities by explicit models are clearly intensified if technology is used to enforce this guidance. In the first two case studies, we did not investigate such enforcement strategies but used models to make people aware of the potentials and limits of using structure or shaping the structure of their work and cooperation. However, technology can be very helpful in this context if it reflects the heavy constraints of work processes without giving up the possibilities for flexibility (e.g. Freeflow, [6]), or if it is used to provide data which can help to complete the explicit models. To understand the relevance and the impact of these problems, we analysed in our case studies the impact of explicit models on the coordination of activities, on quality, on the enactment of designed solutions and on organizational change and its anticipation. In the following sections we introduce three case studies which are our basis to evaluate the chances and risks of modelling (Section 2). Sections 3, 4 and 5 describe the results of this evaluation on the three levels of our framework by referring to different criteria as shown in fig. 2.
T. Herrmann et al. / Modelling Cooperative Work
2 2.1
57
Case Studies Case: Developing new business processes in a university library
A new software-system for their work on the acquisition and cataloguing of books was to be introduced in two departments of a university library. Since there was a fundamental difference between the current organizational practice and the type of processes being required by the new technology, we were asked to help the group to redesign the work processes. The traditional practice in the library was to start with the acquisition and to do the cataloguing afterwards. Now, with the availability of software system based catalogues, the simplification of the cataloguing reduces the required qualification for most of the new books. With the new system, it can be done in the early steps of the process and guarantees a high quality of data. Another reason to introduce this new system was the seamless integration into the software-platforms of the other departments. However, the software system provided more functions than needed and was not sufficiently adaptable to the situation in the library. The project with the library developed a diagrammatic representation of the future practice (Figure 3). The complex structure of interrelated activities for performing the acquisition and cataloguing of media was represented. These tasks were supposed to be supported by the new software system. Therefore it was necessary to develop a picture of the activities with respect to the system's functionality. In parts there was a clear relation between tasks in the diagram and forms in the system. The scope of this project was the tasks of the involved groups doing the acquisition and cataloguing and the goal was to show how the software is employed for these tasks. At first, the main focus was the standard procedure for common books, but most of the special cases were also integrated later. There were six practitioners from the two groups participating in the meetings. We used the modelling method SeeMe [10] [11] to represent the process. The diagrams were represented on pin boards first and then transferred and aesthetically improved using presentation software. 2.2
Case: Web Interface for Ordering Business Cards
The Medienzentrum Duisburg (MZD) offers services and products associated with digital printing. Amongst other things they print business cards for insurance companies. In order to improve the efficiency of the process of ordering and printing business cards for major customers the MZD is developing a web interface. The customer will be able to place or-
Figure 3: Overview diagram of the acquisition and cataloguing tasks
58
T. Herrmann et al. / Modelling Cooperative Work
Figure 4: Production of Business Cards - a Diagram created during a Case Study ders using the web interface without the need for contacting a sales person. Software automatically generates the files needed for production from the information the customer enters through the web interface. We helped one of the responsible managers on the sales team to model the work processes as they were without the web interface and how they might be using the web interface. The people at the MZD had not yet systematically identified the processes that would be affected by the usage of the web interface. They knew that changes would be necessary but they did not exactly know where and how. The diagrams helped to identify those parts of the current working process that had to be reorganized. Furthermore, the representation of the future work processes with the web interface was used to discuss alternative options for the organization. The goal was to ensure that the new process will be as efficient as possible without missing necessary steps. Therefore the scope of the diagrams is the production of business cards starting with the order placed by the customer, then the printing, and ending with the delivery of the cards as well as the invoice (Figure 4 shows the diagram representing the future working processes). We included only those activities, persons and entities in the representation that were either affected by the introduction of the web interface or that had to be included in order to create a coherent model. The modelling language SeeMe [10] [11] with its editor was used to create the diagrams. An extensive visit to the production site was the beginning of the work on the external representation. The responsible manager showed us around and explained the working processes as they were. She also demonstrated a prototype version of the web interface and explained how she intended to use it. Afterwards we - the researchers - drew up first versions of the two process diagrams which we then discussed and modified together with the manager. This manager, who had been trained in the usage of SeeMe before, was the only one who participated in the whole process of creating the diagrams. We did, however, meet and talk to some other employees during our visit of the production site.
59
T. Herrmann et al. / Modelling Cooperative Work Advise Agenda Approach Article Bidding Book
Diagram Diary / Journal Directory Dissertation Documentation Draft
Guideline Handbook Information Instruction Invoice Layout
Contract Correspondence Description
Expenses Figure Form
List Livelink-Element Logo / Icon
Map Masterthesis Method Offer Order Organizational chart Overview Photo Presentation
Profile Proposal Protocol Settlement Software Study Summary Template
Table 1: Scheme for document classification 2.3
Case: Design of a Knowledge Management Application
In a medium-sized consulting company with about 150 consultants and some smaller affiliated companies, we are taking part in the process of designing and introducing a knowledge management (KM) application. Providing an efficient platform for storing and retrieving organizational information, the applications aim at supporting knowledge sharing among the consultants, collaborative knowledge building and qualification, as well as the establishment of new standards in project documentation. The system is designed to match certain envisioned use cases, especially certain information needs occurring during the acquisition and execution of projects. The application builds on the Knowledge-Management-System Livelink® (www.opentext.com), which provides web-based document management services enhanced with several features for the management of tasks, workflows, search and retrieval, categorization of contents etc. The software constrains the design to arrange the contents in one hierarchy. All users have to deal with this hierarchy and it therefore constitutes one critical success factor for the whole project. It became one major issue of the design process. In November 2001 the application runs for about 6 months with about 150 users and consists of a few hundred folders extending into the 6th or 7th level of substructures at many places. At the top level, the hierarchy includes sub-areas for news (1), internal projects (2), internal services (3), knowledge communities (4), organisational units called
Figure 5: Excerpt from the content hierarchy of the KM application
T. Herrmann el al. / Modelling Cooperative Work
competence centres (5), customers and projects (6), process knowledge (7), locations of the company's facilities (8), and affiliated companies (9). In addition to the primary hierarchy of contents, the design process generated templates to replicate substructures and classification schemes for structured descriptions of documents, customers, projects, and consultants' skills. Figure 5 shows a mind-map diagram of the customers and projects branch (6) and the template for retention of project information. What?
Why?
To what extent?
Techniques and methodologies
Process of creating the represented structure
Participation
Library Case A diagram representing the procedures and structures of the future practice of two working groups
Business Card Case Two diagrams: one representing the current procedures and structures; and another one representing the future procedures and structures based on the usage of the web interface. It was the goal to identify those parts of the current working practices that would be affected by the usage of the web interface.
KM Case A hierarchy of existing and to be generated contents and a categorization scheme for contents
Regulating and standardizing the documentation of projects; Facilitating the localization of contents in an extensive archive; The external representation A categorization scheme is The diagrams cover the (planned) practice of the two shows the complete procmeant to cover all contents working groups for acquiring ess of producing business by providing around 50 content-types. The hierarand cataloguing new books. cards: starting with the order placed by the cuschy contains different levels They show tasks and work tomer, the printing, ending of detail. For project docuprocesses employing the mentation detailed hierarwith the delivery of the system, the division of lacards and the invoicing. chies are stipulated. For bour among the users, and building knowledge comroles which are affected by munities the design is less the introduction of the sysdetailed. tem. In some places internal structures and procedures of the system became part of the description. Content hierarchies were The diagrams were develThe diagrams were developed using SeeMe (notaoped using the notation constructed using mindtion as well as editor). SeeMe. mapping techniques. The categorization scheme was represented in a simple list of content types. In a first phase an overview The researchers created Content hierarchies and schemes the diagrams after they had categorization of the main steps of the were compiled by an initial been informed verbally by process was developed. brainstorming and subseone of the managers. During the following sesquently revisited in four sions for each main step one workshops. Since the introor more detailed descriptions duction of the system in were developed. Special spring 2001 both, the hiercases were collected during archy and the classificathe process and detailed later on. During the last tions have undergone minor changes. A more substansession the whole model tial redesign is planned. was checked. In addition to the core projThere were six practitioners Two researchers and the with different roles from the responsible manager were ect team, which consisted actively involved. Other two groups participating in of the project leader, one the meetings. The diagrams employees gave input on a consultant, and two researchers, the chief intranet case-by-case basis. were represented on pin boards first and then transeditor took part in the creaferred and aesthetically tion of the hierarchy. The improved by an experienced hierarchy was discussed modeller. with the project's steering committee including the managers of four other departments. A new software system was to be introduced. At the same time the two working groups should be reorganized.
Table 2: Summary of the three Case Studies
T. Herrmann et al. / Modelling Cooperative Work
61
Mind mapping was one of the favourite representation techniques employed during the design stage. After the system had been introduced, the mind-map lost its function as a tool for planning changes in the hierarchy. However, the project-team keeps the mind-map up to date and uses it as training material. Table 1 shows the scheme for classification of documents. Such metadata schemes were documented using Excel. In the design process, the heads of some influential competence centres, the core project team, consisting of academic researchers and consultants from the knowledge management unit, and a former intranet editor as a delegate from the internal services were involved. The processes stretched over three months and included several smaller and bigger workshops. During the workshops, brainstorming techniques, wallchart print outs and projections were used both to visualize and to discuss the hierarchy and the accompanying classification schemes. 3 3.1
Chances and risks of supporting the reflection on structure Transparency of Organizational Structure
As members of an organization reflect on their practice and their organisation using an externalised representation of structures and processes they become aware of certain properties of their practice and the organization. In particular the members of a transparent organisation create an understanding of the following: (1) The goals of the organisation. (2) The members of the organisation. (3) Their place in the structure of the organisation. (4) Their tasks and responsibilities. (5) The prerequisites of their actions: upon whom and what do they depend? (6) The effects of their own actions: Who is affected by their actions? Who depends and in which way on their actions? (7) The importance of their actions for the organisation as a whole. Two of the case studies in which we used a modelling technique to explicate organizational processes support the argument that modelling activities increase the level of organisational transparency. Using the modelling language SeeMe we modelled roles, actions and entities and thereby created awareness for all of the seven aspects of organizational transparency mentioned above - although the level of clarity is not the same for all aspects in both case studies. In the case of the MZD the focus was on the dependency between actions. One indicator of the success of the modelling was that we did find a need for coordination that had not been thought of before: using the web interface an order will be passed through to production immediately without informing the sales people. The sales people, however, are responsible for preparing the delivery documents and the invoices. This implies that the information about a new order needs to be given to sales people nearly at the same time as to production. In the case of the library, members of two different working groups had to exchange information about their work and then create an integrated work process. They focused on tasks and responsibilities. The fact that the participants understood each other's work in the end, and that they were able to create an integrated organization of their work, supports the notion that modelling can enable organizational transparency. One participant stated it this way: ,,OK, now that we have developed an overview of the tasks there [in the others field of work] you know now that this and that will be working." The current state of the case studies implies that the benefits of the modelling activities have, up to now, been limited to the participants of the process. The coming weeks and months will show whether and how the models will be used to increase the organizational
62
T. Herrmann el at. / Modelling Cooperative Work
transparency for other members of the organizations'. Creating organizational transparency using diagrams may cause problems for some participants who are not able to understand an expression with a new complex diagrammatic notation, so that for them this would not be a helpful tool for reflection. In the case studies for example some people were reluctant in using this kind of method, but they still were able to relate ideas and comments to the diagrams. It is hard to imagine these people sitting at their desk and trying to find personal solutions, by creating diagrams themselves. For them it might be helpful to provide alternative representations, so that these people still have the possibility to find a helpful documentation they feel comfortable with. In the case of the KMS, the aspect of organisational transparency is not as obvious as in the other cases. The question was whether or not a detailed structure for the KMS should be prescribed by the design team or not. One advantage of a precise structure given by the design team is that trainees as well as early users are informed about possible categories of information even though they may not yet be filled. This adds to the picture of the planned usage of the KMS and thereby increases organisational awareness. However, the content hierarchy's and the process diagrams' efficacy to support organizational transparency are difficult to judge at this point in the project. 3.2
Continuous reflection on practice
Organizations and their environment are changing continuously. To keep members' understanding up to date reflection is needed. Members should reflect on their practice continuously and develop a coherent and consistent understanding from their point of view. This can also be understood as maintaining a structure in members' minds that matches the experience made in actions. Doing this continuously is an idealized point of view as the practice in organization gives little opportunity for reflection. Extra effort is needed for this. Also in the field of Workflow Management Systems some authors already mentioned that models of work processes can be useful even outside the scope of workflow technology. Dourish (1996) writes that models can be useful "... for individual and group reflection on process, for documentation or organisational learning purposes." That graphical representations such as diagrams are superior to plain text when describing complex matters is confirmed by authors like Goguen [8]. There are certain issues that can arise if practice and personal understanding of practice are not actively kept in match. One might be that structures develop in different directions in different parts of the organization. As a result extra effort is needed to reintegrate work results again. One example can be found in the Knowledge Management Case: certain structures and categorization schemes were selected, some of which were mandatory in the beginning. In practice it became clear that especially the schema in particular was not complete or not completely understood by everybody. One may want to put extra effort into developing a better scheme. The other option would be to try to make everybody understand the scheme. In this case, the categorization was made open for new entries so that everybody may use the existing scheme, but if this is not sufficient, new categories can be introduced. This solution makes the individual work more efficient, because it is easy to categorize. This solution stresses the personal and situational viewpoint. From a wider organizational viewpoint this open and evolving structure looks different. The goal of having a consistent categorization scheme for all documents also has an efficiency aspect, looking at retrieval quality and costs. Therefore it is helpful, to revise the structures after a while. With both viewpoints we reflect practice and develop different solutions. In real organizations practice needs to be reflected with both a personal and situated viewpoint as well as an organizational viewpoint, and for both levels appropriate opportunities are ' In the library project there was a Kick-Off-Meeting recently where the whole library was informed about the plans. The project group decided to use the diagrams to give an overview of the future practice.
T. Herrmann et al. / Modelling Cooperative Work
63
needed. Diagrams as external representations of mental models can support both levels. As situations and perspectives of members are always different by nature, this may give rise to conflicts, which will be discussed in more detail in chapter 4.2. There are risks that personal reflection is reduced because of modelling projects. To change diagrams may be a too complex task so that individuals try to avoid touching the representations and even avoid thinking about possible and helpful changes of practice. Looking at representations as a dogma and sticking to the negotiated practice once and for all, might be an extreme result of this phenomenon. Due to the state of our cases this was not the case there so far. 4 4.1
Chances and Risks in the Explication of Internally Represented Structures Organizational Walkthrough
When planning to change the way an organization functions it is usually not feasible to run tests with various options and then choose the one that performs best. Visualizing possible future structures and processes by means of modelling can be a substitute for what in other fields would be done using simulations. We use the term "Organizational Walkthrough" to describe the activities of testing organizational structures and processes using artefacts like models by creation and reflection of cognitive artefacts namely diagrams. While Scenarios [5] tend to concentrate on one workplace, we try to take a broader view on the organization. Organizational walkthrough can thus be considered a pre-stage to "organizational change" - an aspect we discuss under category C. The difference between the two aspects is that organizational walkthrough does not have the immediate intention to change reality. Rather it is a method to evaluate whether an intended change will bring about the desired effects. Part of this method is to create artefacts on which the discussion is then based; therefore, organizational walkthrough belongs to category B. Once the evaluation has shown a positive result, organizational change would be the step which follows organizational walkthrough. Our case studies show that modelling can support organizational walkthrough. In the first two cases, the models of the work processes were used to evaluate the effects of an organizational change. The participants wanted to ensure that they knew what would happen if they changed their organization in the intended way. The models enabled them to identify critical points. In the library, the team found it difficult to explicitly describe the role for a certain person. The cause of these difficulties was the contradiction between the director's orders to fully integrate all workplaces on the one hand, and the intention of the group to find workplaces for all members on the other hand. The team felt that this person could not handle a fully integrated workplace where she had to be able to perform all tasks. Now the fact that this contradiction existed was probably known in the group even without the models. But by using modelling, the group was able to discuss various options for solving the problem without actually trying them out and thereby bringing the person in question into a bad position. In the MZD the modelling forced the responsible person to think about all aspects of producing business cards triggered by an order placed through the web interface. During this process it became obvious that there were aspects like the preparation of the delivery documents that had not yet been thought through. Although it would have been possible to run through the process using mock-orders in this case (other than in the library, nobody would have been harmed), we think that using models to visualize the process with all of it's different possible paths is more efficient and complete. In both cases the creation of the models was an integral part of the organizational walkthrough. However, we can also think of situations where existing diagrams are used to verify whether a new task can be fulfilled with an existing organization.
64
T. Herrmann et al. / Modelling Cooperative Work
When we say that organizational walkthrough supported by modelling seems to be a good way to prepare organizational change, we do not say that it is sufficient to sit in a meeting room and draw diagrams. The case study at the MZD showed that quite clearly: in designing the web interface, the first idea was that the production site would be informed about a new order by e-mail. The common means of notification by the e-mail client was planned to be used. When visiting the production site, it became quite obvious that this would not work: the PC was placed in the corner of a room on a desk so that it was not guaranteed that a visual sign such as an opening window would be noticed. A normal acoustic sign would also not be heard because the room was filled with the noise of the printing machines. We argue that models as artefacts can be very helpful when thinking and talking about organizational change - we do not say that models can replace a thorough knowledge of the real world. 4.2
Conflicts
When documenting structures and processes - be it for future or current activities - conflicts and discrepancies will become overt without even explicitly looking for them. In the following we try to shed light on some of the mechanisms that can take place. When creating a model of future working processes - e.g. in the frame of an organizational walkthrough - the participants have the chance to forestall and resolve conflicts that could arise when installing the new organization. The case study in the library provides us with examples for this. Since the group had to agree on one way to draw the diagram, any conflict about the organization of the new work had to be resolved while working on the diagrams. This gave them the chance to discuss matters without being in any emergency situation and without the need for an instant solution. For existing conflicts in an organization, the explication of structures and processes can be helpful to state the area of conflict more precisely, to find an objective level for discussion and to seek solutions. This is true for open as well as for hidden conflicts; but, of course, this method is limited to conflicts that are caused by organizational problems rather than interpersonal conflicts. Obviously there is the risk of becoming involved in discussions about hypothetical situations and possible conflicts about which nobody knows whether they will ever take place or not. There is also the risk that it might be difficult to find a solution for a conflict in a group meeting, whereas a solution would easily be found in the real situation. For both types of risk it is important that the method used for the explication of structures and processes leaves room for vagueness. The facilitator of meetings in which people work on the external representation of structure must help the group to find the right level of detail. On the one hand it is not helpful to be too imprecise. The chance to resolve conflicts before they really arise comes from the need to agree on one way to explicate an internal structure. On the other hand, it can be efficient to remain vague on certain aspects - the solution in reality might be much easier than an extended discussion on the explication. The fact that explicit structure contravenes tacit consent is the root of the problems that can be caused by using models to represent working practices. Our case studies do not contain an example of this; however it is easy to imagine that there are organizations that function on the basis of unspoken agreements. Lucy Suchman [23] describes a similar potential for conflict when she points out that there exists a kind of service work that will not be visible if it is performed well. She argues that "bringing such work forward and rendering it visible may call into question the grounds on which different forms of work are differentially rewarded, both symbolically and materially." In these cases one has to judge carefully whether conflicts that become open during the explication of the organization's
T. Herrmann et al. / Modelling Cooperative Work
65
structure can be dealt with in a positive way, and that this will eventually lead to an improvement. The case study in the library gives us an example of how conflicts can be caused by the activity of creating an externally represented structure. While working on and discussing the diagrams, the group who performed the acquisition of books felt that they were pressured not only to describe but to justify their working practices - even those who were not immediately affected by the introduction of the new technical support system. In the interviews that we conducted a few weeks after the group meetings, one of the participants said that there "were many things that had been fixed in the other group ... procedures that had so far nothing to do with this [the activities that were represented in the diagrams] and that met a lack of understanding because it could not be made plausible why things were handled in this way ..." Without the work on the external representation of the processes there would not have been the need to explain anything to anybody. The facilitator of the meetings had the responsibility to reduce the tension caused by this perception. 4.3
Participation
The explicit representation of organizational structures and processes is a good basis for consultative as well as representative participation because there are artefacts on which feedback and discussions can be based. On the other hand, the mere existence of such artefacts will stimulate discussions even if those who created the artefacts had not intended this. If these discussions are not embedded within a participatory process they can be harmful to the organization as a whole. In the context of this paper it is important to note that the explication of structure can be a very helpful means to support participation but that participation is more than drawing diagrams. Organizations that want to make their structures explicit but are not willing to embark on a participatory exchange, might be confronted with conflicts that stem from disappointed expectations. Creating and discussing external representations of structures requires much effort from the participants: in our case study in the library it took 11 meetings of about two hours. Employees who are willing to do this will be frustrated if this effort does not lead to visible results. It is not undisputable whether representations of structures and processes can be helpful in the design of computer systems. Robinson and Bannon [14] describe the effect of "ontological drift" which means that representations of reality are interpreted in different ways once they have been passed from one semantic community to another. They conclude that there is no "objective reality that can be usefully "captured" in a model and subsequently used as a sufficient basis on which to develop a computerized system." We agree with this statement - external representations of reality are always representations of individual interpretations of reality (fig. 1). However we still think that these representations can be very useful when designing socio-technical systems. Our case study in the library gives an example of how the explication of working processes can enable people to participate in the process of organizational change. The participants in the group meetings were qualified librarians who had no experience in modelling techniques. Not all of them were able to acquire a detailed understanding of the specific modelling language used. However all of them were able to express proposals for changes within the diagrams and thus they were able to participate in the design of their future working processes. In this sense external representation of structure in diagrammatic notation can become a shared tangible artefact that supports individuals to participate in the definition and development of the technical support of their working place. We argue that the explicit representation of processes and structures can help to overcome the gaps between semantic communities.
66
5
T.
Herrmann el at. / Modelling Cooperative Work
Chances and Risks in the reproduction of represented structures in design and activity
Depending on what diagrams or other representations describe and on the way they are employed in design and activity, the impact on practice varies to a great extent, In the library case representations were built to reflect on consequences of the introduction of a given software-system and to design organizational routines and the division of labour in accordance with the system. In the business card example, representations were part of an integrated approach to design work and technology at the same time. However, like in the library case, activities represented in diagrams (cf. fig. 3) were never meant to become steps of a formal workflow, they were represented to reflect and to (redesign organization only. In contrast, the content hierarchy and the categorization schemes designed in the KM case were completely translated into features of the computational artefact. We use the library and the business card example to discuss the impacts of building representations in general (5.1) and concentrate on the KM case to analyse the consequences of the transition into computational artefacts (5.3). Reflecting on the impact on design (5.2) and on organizational change we compare observations from all case studies (5.4). 5.1
Guiding and Coordination of Activity
The theory of structuration [7] clearly points out that individual as well as social representations of structures are an essential prerequisite for any kind of social activity. Implicitly represented as well as "formulated" structures like rules and resources are ubiquitous. They regulate turn taking as well as signification in communication, they legitimate sanctions, and they facilitate power, for instance. Accordingly, the represented structures from our case studies have an impact on practice in multiple ways. There is a general impact on activity as well as potential costs and benefits which are connected to special modes of using represented structures. At a personal level, expressing organizational relations in diagrams is a means to create clarity for acting in the organization. In the case of the university library, for example, participants expressed that the necessity of creating an expression in diagrams, forced them to be clear about the discussed issue. While discussing things without this necessity of creating one model, contradiction between different viewpoints remain opaque. Restricting the representations to spoken language, vagueness and incompleteness are less visible to the participants. They discuss an issue, without getting to the point of agreeing on one way of action. To create a model it is necessary to achieve a greater clarity about what should be expressed then in verbal discussions; in this sense using formal notations helps eliminating ambiguity (c.f. [9]). This support for clarity does not only support transparency and facilitates the resolution of conflicts as discussed in section 3.1 and 4.2. Pro viding agreed upon and less ambiguous grounds, representations become resources users can draw upon to execute and coordinate activity, too. In the mentioned projects there are many decisions about future practice. Most of the time they are not even visible to the participants, because there is no real divergent discussion about the decision. Everybody agrees instantly. On other issues, conflicts become visible while trying to express one or more variants of the model or propose changes to a model representing the different points of view. With the goal to create one representation everyone agrees upon, there is the need to discuss the visible conflicts and come to an agreement. Arising problems at the workplace can be avoided in advance or discussed at a forum away from the emergency situations that need instant solutions and where there is no time for the discussion.
T. Herrmann et al. /Modelling Cooperative Work
5.2
67
Quality of Design
Social construction and negotiation of explicitly formulated structures significantly influence practice. When represented structures are used, however, to configure computational artefacts or organisational routines additional consequences arise. In the library case, the process of modelling lead to a change in the goals of the whole project. Constructing diagrams to envision the future work procedures, the group had started to question the selected software's benefits and finally decided not to introduce this piece of software. If we agree with the participants' point of view, the reflection stage inhibited the introduction of a system that did not match the requirements of the users. The impact of the adoption of representations in design is not always as obvious as in the library case. In the business card case, however, we found some evidence that representation work contributed to the quality of design, too. For instance, the participants of the modelling stage mentioned that the diagram helped them to identify gaps and inconsistencies in their plans for the future service. In the KM case, both chances and risks of using design-representations become visible. On the one hand the adoption of the mindmapping techniques and the application of a drawing-tool allowed flexible handling and presentation. On the other hand, the techniques applied were kind of hypothetical or theoretical in that they only reflected on a few examples of the to be organized contents. As mentioned earlier when the categorization scheme was put to practice this emerged as a significant drawback. Many users came up with contents that did not match the categorization scheme. If the categorization had been developed in closer connection with practice, perhaps this problem would have been recognized earlier. 5.3
Enactment of Designed Solution
Embedding structures in software systems can enhance the effectiveness and the efficiency of an application by automating specified procedures, by providing specialized functionality for pre-specified requirements, and by establishing standards which users can draw upon to coordinate activity. However, when an explicit structure is brought to practice so that it has relevance for activity, specific risks and costs apply, too: unfamiliar, complex, or poorly designed structures cause cognitive burden especially for sporadic users and can lead to inappropriate use causing exceptions and violations of rules other users have to cope with later on. For instance, users who have to solve the violating situation lack the support they are accustomed to and have to develop new strategies. Obligations to match their solution with standards or requirement ensuring the continuation of dependent activities cause additional effort. Or elsewhere users are demanded to contribute to the adaptation of the structure so that it will incorporate future occurrence of similar situations. Even in situations where users manage to bring a structure in coordination with their situation, structures can induce learning and adoption costs, too. In order to be efficient the structure's benefit must outweigh all of these costs. In the KM case, some substructures like, for instance, the structures for project documentation, organizational units, and for process knowledge were designed in more detail than others. Similarly to workflow models the details of the structures were meant to support recurring activities anticipated by the design team, e.g. searching for materials from previous projects, looking for contact persons, etc. This reasoning envisions certain usage scenarios that should benefit from the detailed design. Not surprisingly, the detailed structures violated requirements that were not considered during the design stage, too. However, the pre-specified content hierarchy was generally accepted. Most users reproduced the suggested hierarchy for projects at least partially and some adapted it (fig. 4). During the training users' feedback on standardizing the documentation of projects in this way was positive.
68
7. Herrmann el al, /Modelling Cooperative Work
Many significant problems came from the classification of documents scheme's side (table 1). The pursuit of completeness in the specification of document types evidently annoyed users. Users constantly complained that the list did not match their requirements and suggested the introduction of additional categories. However, subsequent adaptations lead to a list causing even more objection than the first one. Now users complained that they could not decide on an appropriate category. As a consequence, the previously mandatory categorization routine was temporarily set to discretionary mode. 5.4
Organizational Change
Referring to researchers in the CSCW community, Bannon [2] points out that nobody "is arguing against the need for the representation of work per se, rather the argument is over what it is that you are doing when you build representations initially, and how they are to be used in subsequent stages of the design process." Once a structure has been defined and even documented using modelling techniques, one could argue that the existence of these diagrams fixes organizational habits and thereby inhibits organizational change and innovation. Defining a content hierarchy and a rigid content classification scheme for the knowledge-management-system, the design team in the third case study performed a creative process of organizing knowledge for the individual users of the system. It was the unspoken goal of constructing such an extended structure to prevent divergent processes that result in inconsistencies. However, providing such a detailed framework for content retention, incongruous knowledge will most likely not be entered into the system. Innovation concerning the structure might not come to pass because the users do not think about the organization of their knowledge - they rather try to fit information into the existing structure. Since innovation is always the result of a creative process, one could even assume that the number of innovative ideas in general will not increase because creativity is not supported by rigid structures. Several authors - e.g. Suchman [22] - point out the negative effects of structures prescribed by designers. Nevertheless, the users in this case surprised the researchers by asking for such a prescription. Later, we observed many subtle adaptations to the given hierarchy and some users even made detailed propositions to the initial design. One suggestion was for instance, to add functionality so that project documentations could be rearranged more easily when the project team agrees on new hierarchies. The adaptations and the suggestions showed that in certain environments an initial design may nurture or seed creative thinking, too. In the first two cases where work processes were defined using modelling techniques, it is likely that these models will be used to define tasks for employees and to train new members of the groups. Thereby the defined structures and processes will be reinforced in the process of their usage - they might even develop a sort of power like a written law. A danger could be that people will not think about alternatives even if the structures and processes described by the models turn out to be incomplete or even wrong. The diagrams could be used as a dogma by those unwilling to allow change. It could also be that changed circumstances will not be represented by updating the diagrams. This would lead to the effect that the actions described in the diagrams will still be carried out because people were told to do so, although they do not match the environment anymore. We think that such a scenario is one potentially negative effect of modelling. However we would argue that the diagrams could be used to promote innovation by enabling organizational walkthrough (see above). It is important how the diagrams are used within the organization: our goal is to use them in a more descriptive way - providing a group with a model of their interaction - than in a prescriptive way. We extend this position by saying that the usage of representations also has to be continuously reflected on during the course of change within an organization. We agree with Suchman when she writes:
T. Herrmann et al. /Modelling Cooperative Work
69
"The aim is a design practice in which representations of work are taken not as proxies for some independently existent organizational process but as a part of the fabric of meanings within and out of which all working practices - our own and others' - are made." [23].
6
Conclusion
Since the critique of modelling and explicit representation of work reached an intensive level of discussion (e.g. [14] or the Winograd/Suchman debate [24],[22]), a lot of progress has also been made with the improvement and flexibility of modelling methods. We think that some of the success we found in our case studies was possibly due to the usage of notations which are, on one hand flexible, and on the other hand standardized enough to serve as a homogeneous basis for communication. Furthermore, the software tools which are used to develop and to present models have also improved. With a semi-structured method, representations become possible without sticking to formalization and still having the advantage of supporting articulation work in the context of coordination and cooperation. It becomes obvious in our case studies that reasoning about organizational structures needs explicit representations since the overall relationships of an organization cannot be experienced and reflected on from the viewpoint of the activities which are carried out by a single actor. Our conclusion is that there are great opportunities to reap more benefits than costs from the application of explicitly modelled representations of cooperative work. From this point of view it is reasonable to summarize the characteristics of a modelling methods which can support the achievement of these benefits. The method should provide possibilities: • to deal with the dynamics of structures and to indicate it • to switching between maps and scripts in the sense of Schmidt [17] • to express uncertainty incompleteness and the limits of representing • to vary the level of detail and of abstraction. This provides flexibility concerning the distance or proximity between models and concrete work situations • to integrate information about the modelling work itself into the model (metareflection) • t o represent n o t only activities a n d resources b u t a variety o f aspects which a r e Another important point is flexibility which is provided by a software-based editing and presentation tool. Since we do not always know in advance what kind of aspects should be represented, it is sensible to offer electronic hide and show mechanisms concerning the level of detail and the level of abstraction. With the help of appropriate tools, not only the appropriate presentation of the structure but also aesthetic improvement is possible to facilitate the usage and understanding of explicit models in the course of training or participatory design. Furthermore, it should also be supported by tools to convert representations from one modelling notation to another. This can serve for the development of technical systems as well as for the flexible interaction between formal and informal communication. All in all, the modelling method and the appertaining tools should help to create models as boundary objects (in the sense of Star [18]) which are plastic enough to adapt to local needs and viewpoints as well as robust enough to serve for communication across sites.
We thank the participants of the E-CSCW 2001 Workshop on "Structure and Process: the interplay of routine and informed action" especially Paul Dourish and Havard J0rgensen for many valuable comments on previous versions of this paper.
70
T. Herrmann et al. /Modelling Cooperative Work
References [1] Argyris, Chris; SchOn, Donald A. (1996): Organizational Learning II Theory, Method and Practice. New York u. a.: Addison Wesley. [2] Bannon, Liam J. (1995): The Politics of Design: Representing Work. In: Communications of the ACM, Vol. 38, No. 9. pp. 66-68. [3] Bowers, John (1992): The politics of formalism. In: Lea, Martin (1992): Contexts of ComputerMediated Communication. Harvester Wheatsheaf, New York. pp. 232–261. [4] Bowers, J., G. Button and W. Sharrock (1995): Workflow from Within and Without: Technology and Cooperative Work on the Print Industry Shopfloor. In H. Marmolin, Y. Sunblad and K. Schmidt (eds.): Proc. European Conf. Computer-Supported Cooperative Work, Stockholm, 1995. Kluwer Academic, pp. 51–66. [5] Carroll, John M. (ed.) (1995): Scenario Based Design: Envisioning Work and Technology in System Development. New York, Wiley & Sons. [6] Dourish; P.; Holmes, J.; MacLean, A.; Marqvardsen, P.; Zbyslaw, A. (1996): Freeflow: Mediating Between Representation and Action in Workflow Systems. In: Proc. Computer Supported Cooperative Work '96 (CSCW). Cambridge MA USA. ACM. pp. 190–198. [7] Giddens, Anthony (1984): Elements of the Theory of Structuration. In: The Constitution of society: Outline of the Theory of Structure, Ed., University of California Press, Berkeley, CA, pp. 1–40. [8] Goguen, Joseph A. (1993): On Notation. Technical Paper. Download from http://wwwcse.ucsd.edu/users/goguen/pubs/ on 02/12/98. [9] Goguen, Joseph A. (1996): Formality and Informality in Requirements Engineering. In: Proceedings of the Fourth International Conference on Requirements Engineering. IEEE Computer Society Press, pp. 102-108. [10] Herrmann, Thomas; Hoffmann, Marcel; Loser, Kai-Uwe; Moysich, Klaus (2000): Semistnictured models are surprisingly useful. In: Designing Cooperative Systems. Proc. of Coop 2000. (Sophia Antipolis, France, May 2000), pp. 159–174. [11] Herrmann, Thomas; Loser, Kai-Uwe (1999): Vagueness in Models of socio-technical systems. In: Behavior & Information Technology: Special Issue on Analysis of Cooperation and Communication 18(5). pp.313–323 [12]Luhmann, Niklas (1993 (5th ed., 1st 1987)): Soziale Systeme. Grundriss einer allgemeinen Theorie. Frankfurt: Suhrkamp Verlag. [13]Maturana, Humberto R. (1975): The Organization of the Living: A Theory of the Living Organization. In: The International Journal of Man-Machine Studies, 7. New York. pp. 313-332. [14] Robinson, Mike; Bannon, Liam (1991): Questioning Representations. In: ECSCW'91,. Robinson, B. L.; Schmidt, M. (Ed.) Amsterdam, Uni of Amsterdam, NL. pp. 219–233. [15] Schmidt, Kjeld (1991): Riding a Tiger, or Computer Supported Cooperative Work. In: Bannon. L.J.; Robinson, M.; Schmidt, K.: Proceedings of the 2nd European Conference on Computer-Supported Cooperative Work. ECSCW "91. Dordrecht et al.: Kluwer Academic Publishers, pp. 1–16. [16] Schmidt, Kjeld; Bannon, Liam (1992): Taking CSCW Seriously. Supporting Articulation Work. In: Computer Supported Cooperative Work (CSCW) 1: 7–40, 1992. Dordrecht: Kluwer Academic Publishers, pp. 7–40. [17] Schmidt, Kjeld (1999): Of maps and scripts - the status of formal constructs in cooperative work. In: Information and software technology 41, 1999. Amsterdam, Elsevier. pp. 319–329. [18] Star, Susan Leigh (1989): The Structure of Ill-Structured Solutions: Boundary Objects and Heterogeneous Distributed Problem Solving. In: Huhns, M.; Gasser, L. (eds.): Distributed Artificial Intelligence 2. Mento Park, CA: Morgan Kaufman. S. 37-55. [19]Suchman, Lucy A. (1983): Office Procedure as practical Action: Models of work and System Design. In: ACM Transactions on OIS, Vol 1; No 4; 10/83. pp. 320-328. [20]Suchman, Lucy, A.; Wynn, Eleanor (1984): Procedures and Problems in the Office. Office: Technology' and People 2, pp. 133 –154. [21]Suchman, Lucy A. (1987): Plans and situated actions: The problem of human-machine communication. Cambridge U.K.: Cambridge University Press. [22] Suchman, Lucy (1994): Do Categories Have Politics? The language/action perspective reconsidered. In: Computer Supported Cooperative Work (CSCW) 211994. pp. 177–190. [23] Suchman, Lucy (1995): Making Work Visible. In: Communications of the ACM Vol. 38. No. 9. pp. 5664. [24] Winograd, Terry (1994): Categories, Disciplines, and Social Coordination. In: Computer Supported Cooperative Work (CSCW) 2/1994. S. 191–197.
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press, 2002
71
Understanding the Benefits of Graspable Interfaces for Cooperative Use Eva Hornecker Forschungszentrum artec, Universitat Bremen, D-28359 Bremen, Germany eva@artec. uni-bremen. de
Abstract: Although graspable interfaces are often developed as support for cooperative (design) activities, there is still little research on cooperative use aspects. This paper explains the concept of graspable interfaces and analyses their potential as tool for cooperative design. It presents a theory on positive effects of graspable media on cooperation and supporting evidence. After examining effects, the paper presents a model how these effects relate with key characteristics of graspable interfaces. This model is sustained with empirical evidence from video analysis of cooperative design using paper as graspable medium. Results stress the importance of parallel manipulative activity and of nonverbal activity for involvement and active participation. Gestures, talk and visible artefacts interact in producing meaning. The analysis suggests issues for further research, and demonstrates that video analysis is a fruitful method for analysing use of graspable design media. Keywords: CSCW, Graspable Interfaces, Tangible Interfaces, Theoretical Study, Semi-Realistic Field Study, Video Analysis, Cooperative Design, Interaction, Participatory Design Tools
Graspable interfaces are a new type of human-computer interfaces, diverging from the visuallycentred tradition of interface design. Although still in a state of prototypical implementation and exploration of design alternatives, this new approach receives increasing recognition and interest in HCI. Research on graspable user interfaces has focused largely on describing prototypical systems, categorising them, and investigating usability issues. Researchers share a rather intuitive belief that graspable interfaces are a valuable tool for collaborative design, being less intrusive, easier to handle and more amenable to cooperative interaction than graphical tools. This belief has been supported by user reactions to demonstrations and informal experiments with users [43, 44, 16, 40]. Yet closer investigation of this phenomenon is sparse (exceptions are e.g. [2, 41, 10]). If we want to exploit this potential, we must know what constitutes it. Otherwise system designers risk hampering properties of graspable interfaces accidentally, which are valuable for cooperative use. The goal of my research project is to contribute to an understanding: why and how graspable interfaces support cooperative design, which characteristics contribute to this and how these can be consciously exploited in system design. As the concept of graspable interfaces is still rather new, an extended introduction addresses some of the most often asked questions. After characterising graspable interfaces, presenting the state of research contributing to a theoretical understanding, and explaining 'cooperative design', the main part of the paper starts. It presents (the heart of) a theory on the effects of graspable media on cooperation. I will give an overview of positive effects of graspable media for cooperative use
72
E. Harnecker/ Understanding the Benefits of Graspable Interfaces
and relate these effects to properties and characteristics. This overview summarises main results of existing research about the influence of physical environments and artefacts, which can be held to apply to graspable interfaces as well. Many valuable results about functions of tangible (resp. Graspable) objects for cooperation and coordination can be found in ethnographic workplace studies. But these predominantly focus on affordances of paper as an artefact (e.g. the role of flight strips in air traffic control) and on cooperative tasks different from cooperative modelling (office environments, control centres). Thus empirical studies of cooperative design with graspable objects are necessary to support the theory and to concretise it. The last section of this paper presents results from a video analysis of a design session with scraps of paper (interpreted as a graspable medium). This evidence supports most elements from the described theory.
1 Basic Concepts /. / Characterisation of graspable interfaces The concept of graspable (resp. "tangible") interfaces evolved in close association with Augmented Reality from growing dissatisfaction with traditional HCI. Researchers searched for alternatives to desktop metaphor and Virtual Reality. The physical (life)world should retain its role as central reference, augmented with digital capacities. Soon research efforts differentiated into different directions, one of them being graspable interfaces. Fitzmaurice, Ishii, and Buxton [14] introduced the concept of graspable interfaces. It is pursued e.g. in MIT's tangible media projects [23, 43], Rauterbergs BUILD-IT system [15], the Envisionment and Discovery Collaboratory of L3D in Boulder, Colorado [2. 3. 12], and the Real Reality approach in Bremen [9]. These approaches differ in implementation and focus, while sharing main characteristics. Using various technical means physical objects are coupled with digital representations. Change in the physical arrangement is recognised and interpreted as controlling action for the digital information. Either spatial configuration, topology, sequence of actions or all of these can be relevant for interpretation. In graspable interfaces the physical objects integrate functions of representation and control for digital information [43]. They have representational significance for human onlookers. This representational function of graspable interfaces distinguishes them from other interfaces which are based on visual representations. Often additional information, e.g. results from simulation, is visually projected onto the physical work space. People thus can interact with physically and digitally represented aspects of the model. Input and output space coincide in the physical interaction space, the distinction between input and output devices is eliminated. /. /. / What does it mean for an object to be 'graspable''.' It means that it is of material nature, following physical laws, is situated in an environment and can be experienced by the living body. One interacts directly with graspable objects, touching and feeling them. The word tangible expresses the doubleness of touch — being touched by the same thing that one touches, being active and passive at once. From an anthropological viewpoint (or phenomenological), the sense of touch reminds us that we are embodied beings and forms the permeable border between outside and inside, enabling our primary experience of the world. Touch reassures us of our existence. In addition to the phenomenon of touch, the term graspable emphasises interaction with the hands. Whereas touching can be done with any part of the body, grasping refers to the act of enclosing something with ones digits, however partial. Our hands are the most sensorised part of our body and their representation occupies a rather large part of our brain. They To he specific, the idea of using graspable building blocks or construction kits as an input medium to create virtual models (for CAD or simulations) is older. Frazer and Aish [ 1) experimented with it in the early 80s But the idea did not get publicity in the HCI community and was not taken up by other researchers at that time
E. Hornecker/ Understanding the Benefits of Graspable Interfaces
73
have been our foremost and most versatile, creative instruments for ages until technology reduced their utilisation to clicking and pointing. Graspable interfaces, incorporating both real and virtual artefacts, are more than mere "physical props", which augment virtual environments to improve immersion. The real world is augmented and coupled with virtual structures, while remaining locus of control and activity. Different from the concept of token-based access [19] they are more than access points to digital information. Incorporating access, they enable the creation and modelling of structures or systems, thus creating new information. 1.2 Examples illustrating the concept of graspable interfaces The EDC environment (Envisionment and Discovery Collaboratory) has been built for participatory neighbourhood development [2] and utilises a touch-sensitive white-boards as horizontal workspace. Illuminating Light of MIT's Tangible Media group [44] is a learning environment for optical experiments and augments plastic model pieces laid out on a table with a projected, simulated light trajectory. Urp [43] supports planning processes of architects or urban planners and allows the interactive evaluation of design consequences on shadows, reflections and wind flow. BUILD-It [15] supports factory planning. Different from other graspable interfaces, the system relies on a few graspable elements only, which are temporarily bound to projected elements and serve as graspable "handles". Because physical blocks have little representational function, BUILD-IT should be interpreted as a combination of Augmented Reality with ideas from graspable interfaces. The 'Real Reality' concept [8, 9, 36] concentrates on the idea of synchronous modelling in real and virtual worlds. The virtual model can be used for simulation and gives access to complementary representations. 'Real Reality' places emphasis on the physical functionality (behaviour) of real models. The approach thus focuses on domains where functional models do exist, e.g. in factory planning, where small models allow execution of programs, or in the domain of pneumatics, where small versions of valves and pistons are regularly used for training. It seems that most projects up to now focused on domains with an obvious one-to-one representation of domain elements. This might be attributed to this straight-forwardness and to the existence of a long tradition of small scale models in many domains. Experience up to now indicates that design of graspable interfaces is easier if only standardised elements are used, known in advance and only slightly adapted.2 Some projects indicate that graspable interfaces can also be used for interaction within abstract domains. For example the MediaBlocks system of MIT [43] offers functionality of video cutting and sequencing. The earlier Log Jam [10] supported groups in logging and categorising videos. Triangles [16] supported interactive story-telling by connecting triangles. As this approach to Human-Computer-Interaction is rather new, its scope has not been fully explored yet. Therefore it would be premature to conclude on its limits. 1.3 Steps towards a theory of graspable interfaces Many implementations of graspable interfaces have scenarios of cooperative or participatory use (urban planning, learning optical experiments or pneumatics, participatory neighbourhood development, engineers and workers doing layout and configuration of factories). Most researchers report favourably, but only few case or field studies have been reported (e.g.[2, 10, 40]). Up to now, most research on graspable interfaces focused on implementation, although work contributing to a general understanding increases. This concentrates on defining concepts, building category systems [43], evaluating usability [15] or potential interaction metaphors. The focus is on single user interaction, questions of cooperative use are largely left out of consideration. As requirements for coop2
Standardised objects can be retrieved from an object library. Non-standardised elements complicate e.g. consistent coupling of real and virtual elements, recognition of real and production of corresponding virtual objects, and require sophisticated interpretation mechanisms for recognising users intentions.
74
E. Hornecker/ Understanding the Benefits of Graspable Interfaces
erative use are not identical with usability requirements for single user settings - often they are complementary or even incompatible (see e.g. [17]) - a deeper understanding seems essential in order to deliberately design for cooperative use. This motivated the research project described here. Two proposals [43, 7] seem most promising for an understanding of the characteristics of graspable interfaces. Both are relevant for my work presented here. Ullmer and Ishii [43] stress seamless integration of representation and control. a) The physical objects serve as interactive physical controls. b) The state of the ensemble of physical objects embodies key aspects of the systems digital state. Inspecting the physical representation only enables inferring a rough picture of the entire system state. c) Physical objects are computationally coupled with the underlying (digital) information and d) perceptually coupled to digital representations, which is often projected into the workspace. Whereas Ullmer's characterisation focuses on issues of representation and its computational coupling, Brauer's [7] perspective is one of human-computer-interaction, comparing GUI interaction with graspable interfaces. Brauer [7] defines as special qualities of graspable interfaces the following two key characteristics: a) Physical spatiality describes the co-presence of user, objects and other users in one interaction space. This space is a hybrid. Physical objects have a double affiliation to real/physical and virtual/digital space, but must still obey laws of the physical world. Real and virtual parts are each enhanced by the other. Because of co-presence of users and objects, interaction takes place IN the user interface. Therefore input and output space coincide. The user experiences a bodily shared space, his/her body is in the same space as the interaction objects. Following [17, 22, 35, 41] physical spatiality, by preserving physical laws and sharing of space, results in well-understood visibility of objects and of gestures. Strictly speaking this characteristic is a prerequisite for the next characteristic. b) Haptic directness denotes direct manipulation where the physical, graspable objects themselves are the interface. The user has direct contact with the interface elements and has an embodied experience of manipulation, using his/her hands and body movements. Interaction is unmediated and intuitive, leading to 'direct engagement'. Because hands interact directly with interface elements, two-handed or parallel interaction is possible. Unmediated, direct manipulation results in isomorphic and structure-preserving operations. 1.4 Cooperative Design The cooperative use scenarios mentioned above can be characterised as being processes of design, planning or model building. The term cooperative design is used to denote the common type of cooperation in these scenarios (independent from the particular medium used). Planning or designing in groups benefits of drawing from different expertise, comparing perspectives, developing shared views and putting arguments under close scrutiny. All participants learn from each other, exploiting the "symmetry of ignorance" [3, 34] and in result produce something that no-one alone could. Cooperative design is shared making of something that is new to participants. It is a creative and constructive activity, often concerning open-ended design issues. Because of differing perspectives and stakes it bears conflict potential. These conflicts must be handled constructively. Thus processes of developing understanding for different perspectives, evaluation of arguments, agreement and settlement are necessary in order to develop shared understanding and shared solutions. The process can be considered successful if participants agree on an "informed compromise", feel "shared ownership" and do understand the reasons (rationale) underlying decisions. In cooperative design there is no model monopoly of persons or subgroups, as all members contribute actively and scrutinise ideas. ' 3
Model monopoly [6] occurs when a subgroup dominates process and result, either because of political constellation or because of a head-start. The others are forced or tempted to adopt the given model and its inherent perspective, having
E. Hornecker/ Understanding the Benefits of Graspable Interfaces
75
This description denotes an ideal or special quality of cooperative processes and highlights differences to other cooperation types, which emphasise coordination, routine and division of labour. It focuses on conflict resolution and development of shared understanding in highly interactive and argumentative design processes. Cooperative design is related closely to participatory design. In PD, users and system designers build a shared practice of design, often using non-linguistic tools which allow for hands-on exploration and evoke tacit knowledge [13, 28]. The use of mock-ups in PD methods resembles cooperative design with graspable interfaces. Cooperative design is also related to the concept of collaborative knowledge building [38] and attempts for a dialogic theory of learning [27] in CSCL. Crutzen [11] argues that information systems tend to support routine activity. The notion of interaction held widely in computer science (and also CSCW) still relies on the information-transfer model and focuses on routine action. She describes another notion of interaction, situated, relying on human involvement and commitment in a situation, taking place in processes of constructing new meaning. Irritation and break-downs are a chance to interrupt routines and habits, generating doubt, leading to new understanding and the activity of change. Crutzen calls for information systems to enhance the visibility of differences and to support us in reviewing and changing our believes, creating new meaning and giving up routines. Cooperative design relies on accepting irritations caused by different perspectives or by the problem domain and on constructing new meaning. Without irritations routine, custom or habit will govern thinking. 2 The positive effects of graspable media on cooperation Whereas many studies show important characteristics (often termed 'affordances') of paper media due to its physicality (e.g. [29, 4]), only few explicitly investigate use of 3D objects. Nonetheless a lot of findings can be transferred, delivering several lines of argument explaining the positive social effects of graspable interfaces. I will now give an overview of these argument lines and present a model of how these effects relate to characteristics of graspable interfaces. Much interesting work is done in ethnographic studies, influenced by distributed or situated cognition and activity theory. Due to the space limitations, it is not possible to give reference to this multitude of research. A comment in advance: Graspable interfaces do not steer the structure of interaction. They are a tool or medium which supports cooperation, but does not guarantee for its success. Thus mediation, moderation or facilitation may be necessary, which uses graspable media as a tool, whenever appropriate.4 Often the most important part of a meeting is free debate, building new understanding. Graspable interfaces do not interfere in this discussion, but support it. 2.1 Description of effects Concrete graspable models allow for playful, intuitive and experience-oriented ways of interaction [8]. This is especially important for heterogeneous groups and people without abstract domain knowledge. This holds for learners as for workers, whose tacit knowledge is concrete and not abstract. One can discern two levels. Intuitive manipulation of graspable interfaces eases first access, reminding of children's play with bricks. Users do not need to concentrate on manual manipulation and feel less inhibited by low-tech manipulation [30, 13]. Intuitive manipulation concerns simple operations of manipulating objects. Experience-orientation refers to a higher level of semantic meaningful, complex and intentional actions, which depend on users' prior experience. In many domains where users have concrete experience in the real world, physical models help in expressing
4
no chance to produce independent ideas. The problem space is already framed. Because of too little discussion, scrutinising, evaluation and cross-fertilisation, design results often are poor (see for evidence [33]) Existing technical systems supporting group processes concentrate on brainstorming techniques, voting and rational argumentation structures. But the expertise of a moderator is to judge if and when such methods are appropriate.
76
E. Hornecker / Understanding the Benefits of Graspable Interfaces
and eliciting tacit knowledge [28, 13]. E.g. factory workers are able to show complex movements or process patterns manually, using spatial tacit knowledge, whereas they may not be able to explain it verbally.5 Thus active participation and contribution of knowledge by all participants is supported. Graspable interfaces can be manipulated by several people in parallel, not interfering with habituated interaction patterns. Social synchronisation/negotiation e.g. usually prevents people from grasping the same object. Parallel manipulation can also speed up modelling processes, as it allows interactive interaction and simultaneous work on subparts of the model. Experience in participatory design, especially with design games, shows that graspable models give focus to discussions [2, 30]. Abstract arguments must be concretised in face of the model, thus getting disputable. Discussions do not get stuck in abstract arguments and repetitions because the objects are a steady visible reminder of the problem. Many contradictions and problems are easily visible. This fosters consensus and pragmatic resolution of conflicts. The visual, public availability of physical objects makes them function as reminders and can heighten commitment. The physical environment constrains actions (via spatial layout, physical laws, embodyment of domain specific constraints) and enforces focus on the unconstrained remainders. These constraints are all well understood by the participants, because they are visible and familiar. Gutwin and Greenberg [17] analysed properties of physical workspaces (versus virtual ones) supporting awareness of partners and environment. Most properties are valid for graspable interfaces as well. In physical space and through size of the workspace peripheral perception is eased, supporting coordination of actions. People see announcing movements, the actions themselves and results of manipulation. Deictic actions [35, 18] augment oral communication, supplementing additional information or directing attention. Objects serve as shared visible reference for communication and resolve ambiguities [35]. In physical space embodied actions are visible for communication partners and thus have performative meaning besides manipulating objects [22, 35, 41]. Body movement can be used as familiar resource of interaction control [41, 18]. Sharing a space bodily also contributes to a feeling of social nearness and raises willingness to cooperate. Graspable models are visible externalisations. They act as anchor, as graspable symbol to point onto and to show something with. Graspable models serve as externalisation for both actor and listener. To the actor, they have two functions. Used as markers, they help in following a line of thought and visualising things to oneself, supporting individual cognition [32]. One interacts with the representation, using the "backtalk" as feedback [37]. While a mental image cannot be separated from its interpretation, externalisations can cue new interpretation and offer the possibility of doubt and ambiguity [26]. Graspable models also relieve the individual from some of the effort of verbalising, thus extending expression ability. At the same time artefacts (and gestures referring to them) are available to communication partners [41], enhancing understanding, often more easily understood than complicated verbal explanations. Graspable models thus can be a medium of communication across the borders of (professional) languages and can be understood before a common vocabulary is developed. Through shared experience of usage and showing things, meaning is associated with and ascribed to the artefacts, forming a new "language-game" [2, 3, 13]. In Susan L. Stars terms graspable models can serve as "boundary object" [39] in developing a common language. 2.1.1 What are the effects of the integration of virtual and physical ? Up to now I have described effects which hold for graspable interfaces due to their physical parts. But they integrate and blend real and virtual elements. The physical space of the interface is augmented with digital information and controlling abilities. For one physical model there may be many virtual representations associated. This offers additional transitions (or translations) in-between 5
What is experience-oriented thus differs between people of different experience. What is abstract to beginners may be very concrete for experts, who can augment the representations by imagination and experience. But to experts things on a higher level will be abstract again and thus be in need of being made concrete
E. Hornecker/ Understanding the Benefits of Graspable Interfaces
77
representations, where each kind of representation may serve different purposes or highlight different aspects. Augmentation can also "add attributes" to models, e.g. colour or behaviour. The digital part of the system can archive models and enables reconstruction of their evolution. Thus alternative solutions and design rationales can be analysed more easily [3]. Simulations allow analysis of results from the complex interplay of decisions and visualise behaviour of the resulting system. The virtual part of graspable interfaces thus compensates for some of the drawbacks of (non-functional) physical models [2]: adding behaviour and attributes which can not be manipulated as easily in the physical world. Properly designed, graspable interfaces may offer many of those facilities and positive effects which are usually attributed to computer support. Open questions concern which features of real and virtual environments can be integrated into one system without counteracting each other and how to balance this. It is naive to assume that a blend of real and virtual elements automatically brings all positive effects added up. As one needs kind of a 'zero point' (or 'backdrop') for comparisons, I chose to start by establishing an understanding of the effects of physical 3D interaction with objects for cooperation in order to know 'what not to destroy', before expanding the investigation onto mixed configurations of real and virtual components. 2.2 Relating effects to key characteristics of graspable interfaces From published research I extracted a non-comprehensive list of positive effects of graspable models and physical environments (see excerpts above). A closer look, searching for factors enabling these social effects, reveals that many result from common properties. This seems a promising solution to explain which properties make graspable interfaces a valuable tool for cooperative design. Constant visibility, bodily shared space, haptic direct manipulation and parallel access were the most frequent enabling factors. These factors obviously relate to the key characteristics of graspable interfaces (from Brauer [7]). The following graphic shows a subset of this net of relations. At the top are the key characteristics, in the middle the enabling factors, concretising characteristics, below some effects, connected with enabling factors. ('physical spatiality)
( haptic directness
Figure 1. Relation of effects to enabling factors and key characteristics.
Performative meaning of actions is enabled through bodily shared space (of actors and objects). Embodied actions [35, 18] are publicly available, meaningful actions that people rely upon in interaction. The living body is at the same time perceiving and being perceived. Thus constant visibility (in physical environments direct result of shared space) is a prerequisite for performative actions and gestural communication. Constant visibility also facilitates keeping focus. The objects act as steady reminder and invitation to experiment with alternatives. They ground communication and give shared reference; everything said can be compared with the visible model. Awareness relies on bodily shared space, constant visibility and haptic direct manipulation. Bodily shared space simply provides the physical properties of air and material. Constant visibility facilitates peripheral awareness. Haptic direct manipulation in bodily shared space makes actions on objects visible, that is
78
E. Hornecker/ Understanding the Benefits of Graspable Interfaces
announcing movements (opening the hand, moving the arm), the action itself (manipulating objects) and the final result. Because haptic directness implies isomorphic interaction, all of this can be interpreted easily. Externalisation heavily relies on constant visibility and haptic direct manipulation. The latter supports the actor by enhancing expression abilities, offering a medium of manipulation and supporting thought [32]. Constant visibility also supports individual cognition as objects serve as external memory aid [26]. For watchers and listeners constant visibility enhances understanding by grounding communication and ensuring availability of performed actions and representations. Artefacts are reference of communication, or medium of demonstrations. Gestures and performative actions can be interpreted as representations [11]. Intuitive use is facilitated by haptic direct manipulation, lowering thresholds for active participation. Parallel access and manipulation of the workspace ease active participation, because there are fewer time constraints and no artificial synchronisation procedures. This consideration of the relations between effects and enabling factors needs to be continued and sustained with evidence and empirical support. To achieve a deeper understanding of cooperative modelling, I started analysing videos of groups working with graspable interfaces. 3 Analysis of cooperative design with graspable media As mentioned before, there exist few studies of cooperative design using 3D material which can be manipulated. Most studies focused on design interaction with sketches. The empirical part of my project therefore consists of an investigation of design situations with different kinds of graspable media. As the examples illustrating graspable interfaces have shown, a variety of combinations of real and digital components - that is decisions about the design of these interfaces - is possible. Which elements are real, how they are coupled with virtual ones, which augmentations are made, which manipulations are possible, how the virtual components are controlled and accessed - all of this changes the properties of the graspable interface and affects interaction patterns. I will refer to this as 'configurations' of real and digital components. The methodological approach chosen can be described as triangulation across different configurations of real and digital components. Up to now one video analysis is fully completed, from which results are presented here. It examined a design session using a graspable, but purely physical medium - that is scraps of paper. Analysis confirms some of the claims made earlier about the positive effects of graspable media on cooperation. Starting with analysis of a purely physical medium may be questioned as research approach. But the aim is to explore the possibility and potency of graspable interfaces, not the limitations of current prototypes and systems. This study, by focusing on the effects of a purely physical medium allows a glimpse of this potency and a back-drop against which to compare phenomena in other configurations resp. prototypes of graspable interfaces. A second study is mid-way. I observed a weekend course on robotics using LEGO-Mindstorm™, which is related to graspable media (although not being one) because it is a combined real-virtual construction kit. A semi-realistic field trial of the EDC [12] has been conducted and is now being evaluated. These studies allow for comparisons and generalisations in the long run. The approach chosen for analysis of video data is oriented by Interaction Analysis as described by Jordan and Henderson [24]. The analysis of moment-to-moment interaction is deliberately kept free from pre-existing category systems, led by general questions (as explained in [24]) like: the structure of events, temporal organisation, turn taking, participation structure, use of space, influence of artefacts.... Through repeated viewing invisible phenomena become apparent and questions crop up (e.g. on how the design space orientation was negotiated), as repeated viewing alienates the familiar. All findings and hypothesis were cross-checked against the rest of the tape, as hypotheses and conclusions need to be based on evidence on the tape or on explicit knowledge about the context situation. Analysis was pragmatically restricted to topics relevant for the study context and in
E. Hornecker/ Understanding the Benefits of Graspable Interfaces
79
depth, not focusing on the ethnography of a specific setting, but on general behaviours and phenomena. 3.1 Paper prototyping as a graspable medium In a workshop on PD methods a group of six women used the design game PICTFVE [30]. PICTIVE is a low-tech prototyping method for participatory design of user interfaces, based on paper scraps, pens, scissors and transparent foil. Thus it can be considered a graspable medium of physical material only. The group designed the user interface (touch-screen) for the local transportation ticket machine. One person was assigned the role of technical expert, the others of users. Before starting, they only had ten minutes to get acquainted with the task and the fare structure. These six people sat around a table whose middle was reserved as design space and captured on video. On one side of the table the video tripod was mounted, taping the table in a birds eye view. The video is part of the design method and produces a design record. Material with some potential elements of the user interface had been prepared and distributed on the rim of the table. I transcribed about 40 minutes out of 50, including all visible gestures (about ten missing minutes due to changing the tape). The resulting paper prototype was used to disambiguate some of the verbal interaction and to check the content of unreadable text on scraps of paper. 3.1.1 Types of gestures and frequency distribution In comparison with prior studies of face-to-face design sessions which used paper primarily for writing and sketching [42, 31, 5], additional types of actions were found due to the possibility of manipulating the material (cp. Robertson [35]). There are gestures and actions - on the rim of the design space: cutting, scribbling, searching, sorting. These were rarely mentioned yet. - referencing the design space (mostly identical to those in interactions with sketches): simulation of interaction with the system ("kinetic, mimicking" gestures or "enactment" [5, 42, 35]), pointing, indicating an area by circling or waving, and communicative gestures. - manipulating the design space (rare when using paper as drawing surface only): laying scraps of paper, removing scraps, fastening them, rearranging them.
Figure 2. Frequency of gestures over course of the 50 minutes of session (1 bar represents 30 seconds, 10 minutes of tape change shown as negative value)
I analysed frequencies and types of gestures (averaged per ten second intervals) for these three categories. Gesture frequency and percentage of gesture types vary greatly. After ten minutes of low gesture activity, discussing needs of different user groups, frequency of gestures referring to objects rises rapidly once the group starts designing. During quiet phases with few gestures discussion usually centres on general topics (user groups and requirements, colour choice, text, dialogue flow). In 'burst' phases with up to five gestures the group usually rearranges elements and implements ideas, often with several people simultaneously active and creating new elements on the rim.
80
E. Hornecker / Understanding the Benefits of Graspable Interfaces
3.1.2 Parallel activity: interaction, synchronisation and orchestration Analysis shows that parallel activity has an important role in design activity, especially during those 'bursts' of activity when freshly developed design ideas or consensus is transformed into visible design. Besides of several instances of parallel pointing (Figure 3 a), there is parallel manipulation, which is either interactive or independent of each other, interlacing interaction, highly orchestrated, alternating interaction (figure 4) and - most of the time - parallel activity on the rim of the table (see figure 5).
Figure 3. Parallel work: (left to right): a) parallel pointing b) parallel, interactive manipulation. c) parallel but independent synchronous manipulation
Seven scenes show truly parallel manipulation (Figure 3 b & c). Four times two persons interact in rearranging paper scraps, manipulating highly interactive and synchronous. This seems to occur especially when there is some consensus (shared vision) about design. Usually one person begins to rearrange scraps and a second person assists. This is illustrated with transcript Nr. 1. Three times people independently but synchronously manipulate objects in different areas. In addition there are four scenes of almost parallel, alternating manipulation, which may either be interpreted as turntaking with artefacts and bodies (comp. [24]) or as just by accident not parallel executed. There are also several situations with interlacing interaction. While one person cuts off paper scraps and lays them into the design space, the other person positions them.
Figure 4. Example of almost parallel, but alternating manipulations (see frames from left to right)
Researchers working on interactive white-boards (personal communication) report no problems from the constraint that touch-sensitive white-boards can only be manipulated by one person at a time. They rarely noticed an impulse to work interactively. This may be due to the higher threshold of intruding into other people's personal space when standing. The results presented indicate that a horizontal workspace produces different interaction patterns. The table promotes parallel activity, because actors' bodies remain on the periphery of the table while only arms and hands reach into the middle. In recent publications Arias and Fischer [3] mention breakdowns due to the use of a touchsensitive white-board for the EDC system, which enforces a turn-taking and modal interaction. We could observe such breakdowns during our field trial of the EDC [12]. Jordan and Henderson [24] note that group problem solving before a screen differs from groups positioned around a flat table, as the physical arrangement strongly influences the structure of interaction. Analysis demonstrates the importance of parallel manipulative activity for quick and intuitive interaction and highlights its intricate social orchestration. If the technology used in implementing graspable interfaces prohibits parallel manipulation, these fast and effective social mechanisms are endangered. Technical constraints would destroy peoples ease, produce overhead work of coordinating interaction, and breakdowns would interrupt the flow of interaction.
E. Hornecker/ Understanding the Benefits of Graspable Interfaces
81
Transcript Nr. 1 gives an example of parallel, highly interactive manipulation while developing design ideas and testing them at the same time (scene similar to figure 3 b). In this scene U and S interact and cooperate in rearranging scraps and thus in developing design ideas. The ideas evolve in interaction with each other and with the design materials, exploring and extending vague ideas. Transcript Nr. 1 (lasting from 11.55 to 12.19) (talk translated) 11.55
Several 'buttons' have been laid down; discussion concerns what kinds of tickets
do exist
U: "But the 7-day ticket exists for several price levels ? "(points to the fare rates information) B: "That, because"/ (lays new scrap onto ist place)
U: "There is not just one 7-day ticket"
S: "That's true??? " U: "Yeah, there are several."
(points onto fare rates info)
Figure 5. Examples of parallel activity on the rim
D: "Yes, it should anyway " B & S: "Ah " -(S points to the middle, rearranges scraps and pulls them down, U's hand approaches)
U: "Then I would, would do it like that - price level 1 (U pulls left part of upper line (price level 1) further up, while S is still busy) and then
(S now pulls away hand)
(U takes right part of upper line and lays it - indented - under left part) the 7-day ticket here under it." S: "And then - here " / (puts hand in middle again, shoves second line of text down while V is still busy with upper line)
D: (interrupts) " (....) this smaller? " -- (E points to left edge of the upper line and touches it -> now 3 hands visible)
E:" We have this one in big, too, I believe "
3.1.3 Gesture and talk in interaction The session shows a mixture of talk-driven interaction (discussions about requirements analysis and user groups) and instrumental interaction (the concrete physical task of design) (see [24]). Ideas are formed, discussed, and interface elements created during longer phases of discussion with many deictic gestures. When a shared vision is produced, there results a rush of activity of positioning and finishing. Effects on discussion style differ. Sometimes people are busy with preparing and positioning material and almost stop talking, as final design always lags behind discussion. Talk is reduced to a few words (organisational talk), organising the activity (e.g. "Here we have this." "short way ticket", "I'd like this to be here"). During heavy action occurs a kind of fragmented talk, i.e. parallel lines of talk with organisational talk interjected by short ideas or anecdotal stories. While the group is busy with implementing, individuals sometimes seize the opportunity for longer opinion statements. At the end of the session the group manages rapid talk, involving new ideas, and finishing the design in parallel, including instant implementation of new ideas. Whether this is an effect of time running out or getting accustomed to this type of work can only be speculated. Interaction is very quick. When a person states an idea or requirement and there is no objection, another person usually looks around for material and starts cutting and laying scraps into the design space, while discussion proceeds ('ratification qua execution'). Within 10 seconds, no more than 20 seconds, someone reacts (by searching, cutting, laying) to an idea stated. Thus turn-taking consists of verbal as well as action turns which relate to each other (see [24]). Usually there is no explicit discussion of division of labour. This seems extremely quick and effective. Interaction with paper scraps, which is intuitive and direct, allows implementation and fast trial of ideas, shifting inbetween design alternatives, as can be seen in the transcript 1.
82
E. Hornecker / Understanding the Benefits ofGraspable
Interfaces
Some decisions are pragmatic. E.g. orientation of the two sheets serving as 'screen' is decided implicitly and pragmatic by use. At (3.30) S. simulates typing on a virtual keyboard in the lower half of the sheet (seen from her), shortly after she points to the lower edge and says "down here". Three minutes later, when scraps are laid, the screen is oriented to her without any arguments. The second screen receives the opposite orientation because D. (sitting opposite S.) makes the first manual suggestion, although not carrying it through. The group pragmatically accepts the first definition of orientation, although two persons will have problems reading text. But only once someone asks what is written on a scrap, the other times this is inferred from conversation. Non-verbal participation is high. Two persons talk markedly less, but contribute almost as many ideas, questions, and objections as the rest. Regarding non-verbal action, quiet persons are active alike talkative. They react on discussion by searching material, creating and laying down new interface elements on their own accord, thus participating and expressing their opinion non-verbally. In one example (at 10.30) one can see how the background activity of preparing "buttons" keeps B. thinking about the fare system and what is needed on the interface, as she asks a very concrete question, interjecting it into an ongoing discussion "Is it the same fare for a bicycle regardless of being adult or student?". Non-verbal participation thus keeps people involved, allows parallel activity and fosters active participation. Gestures, visible representation and talk augment each other. Together they produce a vivid vision of design ideas. When the design space is yet void (at 3.50), gesture and talk together produce a first vision how elements could be arranged. The following transcript exemplifies this. It also shows how an orientation of the "screen" is proposed (.down here"), implicitly established and accepted by the other participants. In the end, the suggestion is immediately put into action by another participant, indicating that there is a feeling of consensus about the idea. Transcript Nr. 2: (lasting from 3.55 to 4.10) (talk translated) Idea of having a keyboard on the touch-screen hasjust come up — S: "But somehow, I do not know where (moves hand across lower part of paper surface) down here, I can imagine " (points to left lower edge, then onto upper part of surface) D: "can I...." / (interrupted)
S: "To cut all of this out one would go nuts. I believe, We should make a box, " (one hand points to sheet with alphabet printed onto it, then makes a two-handed gesture of a square bracket. ) B: "So, one could do that" / (interrupted) S: "where one says - keyboard " (U takes the sheet with the printed alphabet from the table and starts to cut it)
Figure 6.: Bracket-gesture by S.
D: "she just showed it nicely, perhaps with something like this?" (takes a white slap of paper and lays it where S made the bracket gesture) after some arguments about the slap being too small, U directs attention to the printed alphabet she is cutting out and suggests cutting it into 2 pieces to form a keyboard (as alternative to producing one from scratch)
Further on, this area is referenced to as keyboard area, even before anything is laid down. The same happens for other areas, which e.g. will hold a list of buttons. Verbal and non-verbal suggestions come in one, gestures indicate place and space synchronously with talk, producing a vivid image. Deictic references augment talk and ease expression ("these letters here", "grouping these here". "I want to see the output over there").
E. Homecker/ Understanding the Benefits of Graspable Interfaces
83
Simulations of user interaction with the system (enactment) serves either the summary of results or for clarification. In total there are eleven simulations, rising in number after the first 15 minutes and occurring usually when a design is partly or fully implemented. Only one occurs when no scraps have been laid yet, demonstrating how the virtual keyboard will work. Simulations usually mix a simulation of user actions (typing, clicking), the systems reaction (dialogue flow, output) and referencing objects and areas of the screen with gestures. Other simulations serve as clarification of results (indicated by words like: "Do we agree that...", "Did I understand correctly, that..."), while summarising and delivering a vivid image. In simulatory use, misunderstandings quickly get visible (e.g. taking a ticket from the screen itself). Because there was no time to simulate the final flow of dialogue along several 'dialogue screens', simulations were restricted to referencing gestures. In final enactment of dialogue flow they would need to include replacing sheets or writing on post-its to indicate change of screens. In other domains one can imagine earlier manipulative gestures during simulation. Instantly seeing designs results leads to concrete questions (at 11.10: "But the seven-day ticket costs differently for adults and kids?") caused by irritation about the mismatch of knowledge and visible design. Visibility and concreteness of design offer irritations, evoking questions and objections (at 16.50, very decisive: "When I already said I'm adult, I do not want to see these parts on pupils!") and stimulate imagination about the use situation. 3.1.4 Further observations: role of artefacts, nested activity, process and product quality There seems to be shared ownership, as all participants are allowed to change the design alike and do touch and change scraps. Distribution of labour develops only insofar as the fare information exists only once and by size is accessible for two or three people only. It thus acts as a restricted display (compare [22]). Only the two people sitting next to it use it as resource, giving information to the rest of the group and checking hypotheses and design ideas. They often cooperatively analyse it. The fare information and prepared other materials serve as reminders (of what is not yet included in the design) and as stimulation. Prepared material both opens and closes a space of design ideas, stimulating some ideas, suggesting one trajectory while occluding another. The session can be interpreted as consisting of two nested activity systems (in Activity Theory terms). The outer activity is the workshop on PD methods, the inner activity is the design of an interface with one PD method. In the course of the session the design issue moves to the foreground, as participants take it seriously. Nevertheless there are several focus shifts, when participants ask about the sessions goal, suggest skipping detailed design, or announce the closing of the session. These focus shifts are handled very fluently. Quality of both process and product of the session are good. While still in need of discussing requirements, a quite reasonable proposal was developed within 50 minutes. Discussion covered possible user groups like foreigners, inhabitants and regular commuters, plus different kinds of knowledge (station-name vs. map area). A great deal of controversy covered how to reconcile these S:" I thought, when I type here (types)
. . .. .... . „. then I want to see what type in this output window here, just above it." Figure 7: Simulation of usage as explanation (at 19.30)
K: "Do we agree that when I dick here (taps)
then this screen is replaced with the other menu? " Figure 8: Simulation as clarification and summary (28.30)
84
E. Hornecker / Understanding the Benefits of Graspable Interfaces
competing requirements. The first design trial was given up when the group decided to distribute everything on two screens, for better overview and minimising button-pushing. The virtual keyboard is always visible, allowing to type stop names. Touching a map sign opens a map on which to select the area wanted. Help is available for all ticket types. The resulting design proposal is superior in comparison to most ticket automata in German public transportation and would make a good start for detailed design. The video enables recollecting arguments as input for further requirements analysis. As in participatory design prolonged design sessions may discourage prospective users from participating, methods which enable quick idea development and good discussions are very valuable. 3.2 Relating results to effects of graspable interfaces Quality of both process and result indicate in favour of the effectiveness of modelling with graspable material. The medium of paper was used intuitively. Experience-orientation shows in how participants easily relate scraps of paper to prior experience and adopt complex meaning and usage behaviour, e.g. the possibility of having a virtual keyboard. The importance of parallel manipulation has been discussed in length. The non-verbal activity of searching, cutting and scribbling together fosters involvement and a feeling of shared activity, it allows parallel action and thus supports active participation. In the session focus never got lost, the participants always tried to find new alternatives and played with them. The available materials and information sheets served as reminders and as stimulation, while constraining in some respects, as e.g. the existing fare system couldn't be changed. General discussions always get interrupted by pragmatic solution ideas and efforts towards compromise. The performative meaning of actions can be discerned e.g. in how laying down scraps signifies acceptance of design ideas or is itself part of a new suggestion. Gestures, talk and visible artefacts interact in producing meaning and shared understanding. Awareness can be seen in the sequences of highly interactive manipulation which relies on synchronisation of actions. In addition the group almost always interprets activity on the rim correctly and notices what it contributes to. The positive effects of externalisations can be seen in the quick trial of ideas, consciously exploiting the back-talk of the representation. It relieves of verbalising everything, e.g. in simulations of usage, and eases listeners understanding. Visibility and concreteness evoke irritation, questions and objections, enforcing focus and clarifying discussion. Over the course of the transcript, the evolution of a new language game in which the scraps play the role of boundary objects can be discerned. Meaning is ascribed to parts of the blank surface, even before any scraps are laid down and intended meaning is shown by simulation of usage and talk. Thus understanding builds up and a common vocabulary ('keyboard', 'buttons', 'expert-mode or button', 'Back-button') develops.
4 Concluding remarks, Conclusions, and Future Work The analysis supports the hypothesis of the relation between positive social effects and properties of graspable interfaces. The empirical observation also led to additional arguments and issues which need to be integrated into the theoretical framework. Specific design recommendations can not be inferred from the reported study, except of the demand that cooperatively used graspable interfaces should support parallel manipulation. Nevertheless the study points to several issues. As background activity keeps participants involved and contributing productively, enabling such background activity might be a valuable design goal. But there are some indications that activity on the rim can distract from ongoing talk, making people miss new ideas (there are two incidents, one needing clarification). How much background activity is positive, at which point do negative effects dominate? The study also indicates that one should design to support non-verbal and non-textual participation, which enhances less talkative peoples chances. Systems should also support the use of gestures (compare [5, 42]). But it is neither techni-
E. Hornecker / Understanding the Benefits of Graspable Interfaces
85
cally possible nor sensible to catch all details, to mirror the physical world and to interpret everything. This maintains opportunity for social synchronisation and spontaneously evolving groupspecific procedures, resp. improvisation (compare [10, 21]). For designing graspable interfaces, we also need to get a better understanding of how the given design space, the representations and materials constrain the design process and the evolving distribution of labour. The transfer of results from this study ought to be carefully reflected, as the paper prototyping setting took place in physical space, not in hybrid space. The other parts of Brauer's key characteristics of graspable interfaces [7] hold for the observed setting. This can be taken as a shortfall of this characterisation, as it doesn't seem to include enough issues particular to hybrid settings. From the other proposal [43], the paper prototyping setting weakly fulfils two out of four characteristics. The physical objects serve as interactive controls, but there is nothing to control except themselves. They also embody key aspects of the underlying system state, as looking at the physical workspace allows some insight. But there is no underlying digital state, only the groups design vision. A systematic overview of which characteristics hold for certain systems will be done in the future. More studies of design with graspable interfaces have been conducted [12], resp. are planned. With more empirical studies of graspable media comparison of results across different configurations of real and digital components and relating effects to characteristics will be possible. This research approach will enable structured evaluation of systems regarding cooperative use and the attribution of effects to different design decisions. It will contribute to closing the knowledge gap between single-user interaction and human-human interaction with graspable media and lead to design recommendations. The results show that video analysis of design sessions with graspable media is a fruitful method and motivates further research using similar methods. Acknowledgements Thanks to all my colleagues at artec, to Volker Wulff for comments on a previous paper, to all participants of the PICTIVE session, to IRIS'24 participants (esp. WG Primula) for comments and advice, the COOP reviewers, the researchers at L3D for hospitality and discussions, Brygg Ullmer for intense discussions on characteristics of tangibles, and the Hans-B6ckler foundation for funding this work.
References [ 1 ] R. Aish, Three-Dimensional Input for CAAD System. Computer-Aided Design 11, No. 2, 1979, 66–70. [2] E. Arias, H. Eden, and G. Fischer, Enhancing Communication, Facilitating Shared Understanding, and Creating Better Artifacts by Integrating Physical and Computational Media for Design. In: Proc. of DIS '97. ACM, NY, 1997, pp. 1-12. [3] E. Arias and G. Fischer, Boundary Objects: Their Role in Articulating the Task at Hand and Making Information Relevant to It. In: Proc. of Int. Symposium on Interactive & Collaborative Computing (ICC'2000). Australia. ICSC Academic Press, Wetaskiwin, Canada, 2000, pp 567–574. [4] V. Bellotti and Y. Rogers, From Web Press to Web Pressure: Multimedia Representations and Multimedia Publishing. In: Proc. of CHI'97. ACM, NY 1997, pp. 279-286. [5] M. Bekker, An analysis of user interface design practice: Towards support for team communication. PhD thesis, TU Delft, NL, 1995. [6] S. Braten, Asymmetric Discourse and Cognitive Autonomy: Resolving Model Monopoly Through Boundary Shifts. In: A. Predretti & G. de Zeeuw (eds.):Problems of Levels and Boundaries. Princelet Editions, London/Zurich, 1983, pp. 7–28. [7] V. Brauer, Gegenstandliche Benutzungsschnittstellen fur die Mensch-Computer-Interaktion. PhD-thesis, University of Bremen, Germany. 1999. [8] F.W. Bruns, Zur Ruckgewinnung von Sinnlichkeit. Eine neue Form des Umgangs mit Rechnern. Technische Rundschau, 29(39), Zurich, 1993, pp. 14–18. [9] F.W. Bruns, Complex Construction Kits for Coupled Real and Virtual Engineering Workspaces. In: N. Streitz, J. Siegel, V. Hartkopf, & S. Konomi (eds.): Proc. of Cooperative Buildings. Springer, 1999, pp. 55–68.
86
E. Hornecker / Understanding the Benefits of Graspable Interfaces
[10] J. Cohen, M. Withgott and P. Piernot, Logjam: a Tangible Multi-Person Interface for Video Logging. In: Proc. of CHI'99. ACM, NY 1999, pp. 128–135. [11] C. Crutzen, Interactie, een wereld van verschillen — Een visie op informatica vanuit genderstudies. PhD thesis, Open University of Netherland, Heerlen, 2000. [12] H. Eden, E. Hornecker and E. Scharff, A semi-realistic field trial of the EDC. technical report (in preparation), research center artec, University of Bremen, Germany & University of Colorado, Boulder, 2002. [13] P. Ehn, Scandinavian Design: On Participation and Skill. In D. Schuler & A. Namioka (eds.): Participatory Design. Principles and Practices. Lawrence Erlbaum, 1993, pp. 41–77. [14] G.W. Fitzmaurice, H. Ishii, and W. Buxton, Bricks: Laying the foundation for Graspable User Interfaces. In: Proc. of CHI'95. ACM, NY, 1995, pp. 422–449. [15] M. Fjeld, et al, Exploring Brick-Based Navigation and Composition in an Augmented Reality. In: Proc. of HUC'99. Springer, 1999, pp. 102–116. [16] M. Gorbet, M. Orth and H. Ishii, Triangles: Tangible Interface for Manipulation and Exploration of Digital Information Topography. In: Proc. of CHI '98. ACM, NY, 1998, pp. 49–56 [17] C. Gutwin and S. Greenberg, Design for Individuals, Design for Groups. Tradeoffs between Power and Workspace Awareness. In: Proc of CSCW'98. ACM, NY, 1998, pp. 207–21.6. [18] C. Heath and P. Luff, Technology in Action. Cambridge University Press, 2000 [19] L.E. Holmquist, J. Redstrom and P.Ljungstrand, Token-Based Access to Digital Information. In: Proc. of HUC'99. Springer, 1999, pp 234–245. [20] E. Hornecker, Graspable Interfaces as Tool for Cooperative Modelling. In: Proc. of IRIS'24 (The 24th Information Systems Research Seminar in Scandinavia). Bergen, Norway, 2001, vol. 3, pp. 215–228. [21] E. Hornecker, B. Robben and F.W. Bruns, Technische Spielraume: Gegenstandliche Computerschnittstellen als Werkzeug fur erfahrungsorientiertes, kooperatives Modellieren. In: I. Matuschek, A. Henninger & F. Kleemann (eds.), Neue Medien im Arbeitsalltag. Wiesbaden: Westdeutscher Verlag. 2001, pp 193–216. [22] E. Hutchins and T. Klausen, Distributed cognition in an airline cockpit. In: Y. Engestrom & D. Middleton (eds.): Cognition and Communication at Work. Cambridge University Press, 1998, pp. 15–34 [23] H. Ishii and B. Ullmer, Tangible Bits: Towards seamless interfaces between people bits and atoms. In: Proc. of CHI'97. ACM, NY, 1997 pp. 234-241. [24] B. Jordan and A. Henderson, Interaction Analysis: Foundations and Practice. The J. of the Learning Sciences. 4(1), 1995,39–103. [25] A. Kendon, An Agenda for Gesture Studies. Semiotic Review of Books. 7(3) 1996, 8–12. [26] D. Kirsh, The intelligent use of space. Artificial Intelligence 73( 1 –2), 1995, 31 –68. [27] T. Koschmann, Toward a Dialogic Theory of Learning: Bakhtins's Contribution to Understanding Learning in Settings of Collaboration. In: Proc. of CSCL'99, 1999, pp. 308–313. [28] M. Kyng, Making Representations Work. Comm. of the ACM 38(9), 1995, 46–55. [29] W. E. Mackay and A.-L. Fayard, Designing Interactive Paper: Lessons from Three Augmented Reality Projects. In: R. Behringer, G. Klinker & D.W. Mizell, (eds.): Augmented Reality - Placing Artificial Objects in Real Scenes. Proc. of IWAR'98. AK Peters, 1999, pp. 81–90. [30] M. Muller, PICTIVE: Democratizing the Dynamics of the Design Session. In: D. Schuler & A. Namioka (eds.): Participatory Design. Principles and Practices. Lawrence Erlbaum, 1993, pp. 211–237. [31] I. Neilsen and J. Lee, Conversation with graphics. Int. J. of Human-Camp.-Stud., 40(1994), 509–541. [32] D. Norman, Things that Make Us Smart. Addison Wesley, 1994. [33] J. Pasch, Dialogical Software Design. In: Proc. of HCI'91. Elsevier, 1991, pp. 556-560. [34] H.W.J. Rittel, Second Generation Design Methods. In: N. Cross (ed.): Developments in Design Methodology. John Wiley & Sons, 1984, pp. 317–327 [35] T. Robertson, Cooperative Work and Lived Cognition. A Taxonomy of Embodied Actions. In: Proc. of ECSCW'97. Kluwer, 1997, pp. 205–220. [36] K. Schafer, V. Brauer and W. Bruns, A new Approach to Human-Computer Interaction - Synchronous Modelling in Real and Virtual Spaces. In: Designing Interactive Systems, DIS'97. ACM, NY, 1997 pp. 335-344 [37] D.A. Schon, The Reflective Practicioner: How Professionals Think in Action. NY, Basic Books, 1983. [38] G. Stahl, Contributions to a Theoretical Framework for CSCL. Accepted Paper for CSCL 2002 (preprint) [39] S.L. Star, Cooperation Without Consensus in Scientific Problem Solving. In: S. Easterbrook (ed.): CSCW: Cooperation of Conflict ?. Springer. London. 1993. pp. 93-105.
E. Hornecker/ Understanding the Benefits of Graspable Interfaces
87
[40] M. Sugimoto, F. Kusunoki and H. Hashizume, A System for Supporting Group Activities with a Sensor-Embedded Board. In: E-CSCW 2001 Conference Supplement. Bonn, Germany, 2001, pp. 25–28. [41] H. Suzuki and H. Kato, Interaction-Level Support for Collaborative Learning: AlgoBlock - An Open Programming Language. In: Proc. of Computer Supported Collaborative Learning (CSCL'95). Indiana, 1995. [42] J.C. Tang, Findings from observational studies of collaborative work. Int. J. of Man-Mach.-Stud., 34(2), 1991, 143-160. [43] B. Ullmer and H. Ishii, Emerging Frameworks for Tangible User Interfaces. IBM Systems Journal. 39(3&4), 2000. pp. 915–931. [44] J. Underkoffler and H. Ishii, Illuminating Light: An Optical Design Tool with a Luminous-Tangible Interface. In: Proc. of CHI '98. ACM, NY, 1998, pp.542–549.
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press. 2002
Defining Task Interdependencies and Coordination Mechanisms for Collaborative Systems Alberto B. RAPOSO abraposo @ tecgraf.puc-rio. br Computer Graphics Group (Tecgraf)
and
Hugo FUKS hugo @ inf.puc-rio. br Software Engineering Laboratory (LES)
Computer Science Department - Catholic University of Rio de Janeiro (PUC-Rio) R. Marques de Sao Vicente. 225 — Rio de Janeiro. RJ, Brazil – 22453–900
Abstract This paper addresses the issue of coordination, which is an essential matter to the specification of activities in collaborative systems. An activity can be described as a set of interdependent tasks. In order to specify task interdependencies, an extensible model encompassing a set of temporal and resource management relations is initially presented. Then, Petri Nets are used to formally model coordination mechanisms for those interdependencies. The separation between tasks and interdependencies/coordination mechanisms allows for the deployment of different coordination policies in the same collaborative system by only changing the coordination mechanisms. Moreover, the coordination mechanisms, like the related interdependencies, are generic and may be reused in several collaborative systems. Keywords. CSCW, Coordination, Petri Nets, Synchronization, System Design.
1 Introduction In order to work collaboratively, people need to share information (communication). Communication, although vital, is not enough; "it takes shared space to create shared understandings" [1]. This notion of shared workspace (including user awareness, shared objects, etc.) is called cooperation. To cooperate, however, people need to work harmoniously, avoiding conflicting or repetitive actions (coordination). These aspects (communication, cooperation and coordination) constitute a threesome frequently associated with collaboration [2], [3]. The present work treats one of those aspects - coordination, which in a broad sense may be defined as the activities responsible to ensure the effectiveness of the collaborative work. In this broad sense, coordination is a synonym of what has been called articulation work, defined as "a set of activities required to manage the distributed nature of cooperative work" [4]. Among the activities of articulation work, the identification of the objectives of the group work, the mapping of these objectives into tasks, the participants' selection, the distribution of tasks among them, and the coordination (in a narrow sense, as it will be seen below) of tasks execution can be mentioned. Coordination, in a narrow definition, is "the act of managing interdependencies between activities performed to achieve a goal" [5]. In this sense, coordination is the most important
A.B. Raposo and H. Fuks / Defining Task Interdependencies
89
part of the articulation work because it represents the dynamic aspect of articulation, demanding renegotiation almost continuously during a collaborative effort. Coordination, in its broad definition, is essential to any kind of collaboration. In spite of that, in its narrowest definition, coordination does not need to appear explicitly in some kinds of collaborative activities - called loosely integrated collaborative activities - such as those realized by means of chats or audio/videoconferences. These activities are deeply associated with social relations and generally are well coordinated by the valid "social protocol," which is characterized by the absence of any explicit coordination mechanism among the activities, trusting users' abilities to mediate interactions (the coordination is culturally established and strongly dependent on mutual awareness). On the other hand, there is a large group of activities (tightly integrated collaborative activities) that require sophisticated coordination mechanisms in order to be efficiently supported by computer systems. In this kind of activity, tasks depend on one another to start, to be performed, and/or to end. Examples of tightly integrated activities may be found in workflow procedures, learningware, collaborative authoring, multi-user computer games, among others. In this context, this paper introduces a model for the definition of generic interdependencies that occur between tasks in tightly integrated collaborative activities and proposes Petri Net-based coordination mechanisms to handle such interdependencies. The coordination model encompasses temporal and resource-related interdependencies, which have a direct mapping to the Petri Net-based coordination mechanisms. In the sequence, several aspects related to tasks interdependencies are discussed. Section 3 introduces the coordination mechanisms for those interdependencies, also showing some examples of use. Then, in Section 4, a brief comparison with related work is presented. Finally there are conclusions and suggestions for future research.
2 The Coordination Model: Task Interdependencies In the context of this work, a collaborative activity is defined as a coordinated set of tasks realized by multiple actors in order to achieve a common goal. Thus, a task, either atomic or expressed as a group of subtasks, is one of the building blocks of any collaborative activity. A group of subtasks could be considered to be a task when it presents no external interdependencies, that is, no interdependencies with another task that does not belong to the group. This definition of task enables the modeling of collaborative activities using several abstraction levels (see Figure 1), which facilitates the coordination specification and management. Interdependency is a key concept in the coordination theory - if there are no dependencies between tasks to be performed in a collaborative effort, there is nothing to coordinate [6]. The approach task/interdependency or, more specifically, the clear separation between "articulation work, i.e., the work devoted to activity coordination and coordinated work, i.e., the work devoted to their articulated execution in the target domain" [7] is a step toward giving flexibility to coordination mechanisms, which is crucial to further use of this kind of mechanism. One of the advantages of the separation task/interdependency is the possibility of altering coordination policies by simply altering the coordination mechanisms for the interdependencies, without the necessity of altering the core of the collaborative system. Additionally, interdependencies and their coordination mechanisms may be reused. It is possible to characterize different kinds of interdependencies and identify the coordination mechanisms to manage them, creating a set of interdependencies and respective
90
A.B. RaposoandH, Fuks / Defining Task Interdependencies
Figure 1: Hierarchical model of tasks and collaborative activities.
coordination mechanisms capable of encompassing a wide range of collaborative applications [6]. In this section, a generic set of interdependencies that occur between tasks in collaborative activities is defined. Then, in Section 3, coordination mechanisms to control those dependencies are proposed. 2.1 Basic Temporal Interdependencies Temporal interdependencies establish the relative order of execution between a pair of tasks. The set of temporal interdependencies of the proposed model is based on temporal relations defined by J. F. Allen [8]. He proved that there is a set of primitive and mutually exclusive relations that could be applied over time intervals (i.e., any pair of time intervals are necessarily related by one and only one of Allen's relations). A time interval is characterized by two events, which in turn are associated to time instants. The first event is the starting (initial) time of an interval A, denoted here ia. The other event is the ending (final) time of the same interval, denoted fa, always with ia < fa. According to Allen, the set of seven primitive relations shown in Figure 2 may maintain temporal information considering any pair of time intervals A and B (if one considers the inverse relations, then 13 relations can be defined, because the inverse of equals is the equals relation itself)Based on the relations of Figure 2, a set of axioms is defined to create a temporal logic. For example, there are axioms to prove the mutual exclusion and the exhaustivity of the basic relations and others to define transitivity relations, e.g., if A during B and B before C, then it is inferred that A before C [9]. The fact of being applied over time intervals (and not over time instants) made the above relations suited for task coordination purposes, because tasks are generally noninstantaneous operations. The adaptation of Allen's primitives to the context of collaborative activities takes into account that any task T will take some time (from i to f) to be performed. Nevertheless, Allen's temporal logic is defined in a context where it is essential to have properties such as the definition of a minimal set of basic relations, the mutual exclusion
91
A.B. Raposo and H. Fuks / Defining Task Interdependencies
time A equals B A starts B A finishes B A meets B A overlaps B A B
A
A
B ia =
and
A
A
B ib
ia
>ib
and
and
fa < fb
fa = fb
B fa
:
= ib
A during B A
B
ia< ib and ia
A before B A
B
ia > ib and
B
fa
Figure 2: Allen's primitive relations between time intervals A and B.
among these relations and the possibility to make inferences over them. Temporal interdependencies between collaborative tasks, on the other hand, are inserted in a different context. What really matters here is the management of the interdependencies and the proper understanding by groupware designers. Another drawback of Allen's relation is that they are merely descriptive, not expressing causal or functional relations between intervals [10]. For example, if tasks A and B are related by the equals temporal interdependency, what should the coordination mechanism do when task A is ready to begin, but not task B? Should it block the execution of task A until task B is ready, or should it force the start of task B to guarantee that the interdependency will be respected? In a different situation, if it is said that task A occurs before task B, what should be done when task B is ready but not task A? Should the coordination mechanism block task B until the end of task A, or should it allow the execution of task B, blocking future executions of task A (which would violate the relation)? For all of these reasons, it was necessary to make some adaptations to Allen's basic relations. More than answering questions such as the ones stated above, the goal of the proposed extensions is to offer a larger set of possibilities to create coordination mechanisms that could handle many different situations. The idea is to accept Allen's seven primitives as the basic interdependencies and to provide a list of extensions, which may be viewed as specializations of them.
2.2 Active and Passive Interdependencies The merely descriptive characteristic of Allen's temporal relations allows for different interpretations of a single interdependency. Consider, for instance, that two tasks, Ta and Tb are related by the interdependency Ta equals Tb. This construct establishes that the two specified tasks must be executed simultaneously. In the coordination context, this can be interpreted in two different ways. In the first sense, denoted as the active interpretation, this relation expresses that the beginning of one task should start another task; similarly, the end of one of the tasks should conclude the other task. Consider a situation in which one task is "to start a discussion session" and the other task is "to record an ongoing discussion session." From the coordination point of view, this "active equals" relationship between these two tasks would simply indicate that the second task (record the session) should follow the execution of the first task. However, a problem to proceed with the session recording would not invalidate the discussion session itself. The second possible interpretation for any coordination mechanism is denoted as the passive interpretation. In this case, the coordination mechanism expresses a set of
92
A.B. Raposo and H. Fuks / Defining Task Interdependencies
conditions that should be obeyed in order to carry out the activity. Considering the same example as above, this would be the case whenever the session recording must be ready before the start of the discussion. Thus, a problem to record the session would delay the beginning of the discussion session until the problem is solved. In order to deal both with active and passive interpretations, two operators were defined: enables and forces. The enables operator represents the passive interpretation, while forces represents the active one. These operations may be applied on the initial and final instants of each interdependent task. Additionally, these extreme points have two states, ready and concluded, indicating, respectively, that the task is ready to start (or finish) and that it has already started (or finished). These states are used in the first operand, indicating that it will enable or force the second operand before (ready) or after (concluded) its own execution. Consider, for example, two tasks Ta and Tb, with initial and final points ia, ib, fa and fb. The interdependency Ta equals Tb may be extended into several interpretations. For the simultaneous beginning: ia (ready) enables ib, AND ib (ready) enables ia – this statement indicates the passive situation, in which the tasks will start their execution only when both are ready (i.e., Tb will be enabled to start only when Ta is ready to start, and vice-versa), but neither will force the execution of the other. ia (ready) forces ib, – in this situation, when Ta is ready to begin, Tb is forced to start, indicating a master/slave active interdependency (similarly, Tb could be considered the master if ib (ready) forces ia). ia (ready) forces ib,AND ib (ready) forces ia - active interdependency with no master (the beginning of each task will force the beginning of the other). ia (ready) forces ib AND ib (ready) enables ia – Ta is the master, forcing the beginning of Tb, but Ta will only be started when Tb is ready (returning to the discussion session example, this situation indicates that the beginning of the section – Ta – would force the recording task, but if there is a problem with the recorder, the session will not start). Similar interpretations are applied to the simultaneous end of both tasks: fa (ready) enables fb AND fb (ready) enables fa – passive situation. Tasks will finish their execution only when both are ready to finish, but neither will force the end of the other. fa (ready) forces fb – when Ta is ready to finish, Tb is also forced to finish (master/slave). fa (ready) forces fb AND fb (ready) forces fa – active interdependency with no master. fa (ready) forces fb AND fb (ready) enables fa – Ta is the master, but has to wait until Tb is ready to finish. Thus, interdependency Ta equals Tb may be composed of any combination of the above simultaneous beginning and end situations (for example, it could have a master/slave beginning and a passive end). The interdependency Ta starts Tb has the same interpretations than equals for the simultaneous beginning of both tasks. Regarding the end of the tasks, the original relation imposes that Ta finishes before Tb, which is represented by fa (concluded) enables fb (the end of Tb will be enabled only after the end of Ta, as indicated by the concluded state of fa). However, it is not always interesting to impose any restriction to the end of tasks that have started together. In this case, it is possible to relax the original relation simply not including any relation for the ends of the tasks. For the simultaneous end of the tasks, interdependency Ta finishes Tb has the same interpretations as equals. For the beginning of the tasks, it is necessary to impose at least
A.B. Raposo and H. Fuks / Defining Task Interdependencies
93
the following restrictions, ia (concluded) enables fb AND ib (concluded) enables fa. These restrictions are necessary to guarantee that a task will only be ready to finish when the other has already started (otherwise, in the active interpretation, a task could force the end of a task that has not started). This situation does not make any assumption about which task will start before (relaxing Allen's original relation). To follow the restriction of the original relation, imposing that Ta starts after Tb, it is necessary to add the statement ib (concluded) enables ia. The interdependency Ta meets Tb may have the following interpretations: ib (ready) enables fa AND fa (concluded) forces ib – passive situation. Ta will only be able to finish when Tb is ready to start, and the end of Ta forces the beginning of Tb, in order to respect the interdependency. fa (concluded) forces ib- active situation where Ta is the master. The difference with the previous situation is that the end of Ta does not have to wait until Tb is ready to begin. ib (ready) forces fa – active situation where Tb is the master. When it is ready to begin, Ta is forced to finish. fa (concluded) forces ib AND ib (ready) forces fa – active situation with no specific master. The interdependency Ta overlaps Tb has the following statement for its passive interpretation, ia (concluded) enables ib AND ib (concluded) enables fa AND fa (concluded) enables fb. It is possible to relax the original relation removing, for example, the last part of the previous statement (in this case, it would not matter which task finishes first). An active interpretation of this interdependency could be fa (ready) forces ib, indicating that Tb will be forced to start when Ta wants to finish. Ta during Tb may have the following interpretations: ib (concluded) enables ia AND fa (concluded) enables fb – passive situation. fb (ready) forces ia AND fa (concluded) enables fb – in this case, Tb is the master, forcing the execution of Ta before the master's end. ia (ready) forces ib AND ib (concluded) enables ia AND fa (concluded) enables fb – Ta is the master, forcing the start of Tb when the master is ready to execute. Finally, Ta before Tb allows for a single passive interpretation, fa (concluded) enables ib. An active interpretation is not possible considering just the initial and final instants of both tasks, because that interdependency imposes an interval between both tasks (otherwise, it becomes the meets interdependency). The way to create an active interpretation for this interdependency is by defining a delay parameter for the forces operator, such as in the statement fa (concluded) forces[5s] ib, indicating that Tb will be forced to start 5 seconds after Ta. The interpretations presented in this section do not claim to be exhaustive, but show the large number of possibilities that arise when considering the passive and active interpretations of temporal interdependencies.
2.3 Interdependencies Allowing a Variable Number of Executions In spite of the operators enables and forces created to adapt Allen's relations for active and passive coordination interpretations, there are undefined situations remaining. Such a situation occurs, for example, in Ta before Tb. After Ta and Tb have been finished, how should the coordination mechanism proceed if Tb wants to start again? Should it allows its execution, since Ta has already been executed (one to many relationship), or should it makes Tb wait until Ta is executed again (one to one relationship)? A similar doubt arises
94
A.B.
Raposo and H. Fuks / Defining Task Interdependencies
for Ta during Tb, i.e., how many times Ta is allowed to execute during a single execution In order to deal with such situations, it was necessary to include an optional parameter for the enables operator. This parameter indicates the number of times a condition (first operand) enables the event (second operand). For example, to define that Ta during Tb allows the maximum of two executions of Ta for each execution of Tb, the following statements are used ib (concluded) enables[2] ia AND fa (concluded) enables fb. A similar situation occurs for Ta before Tb, as illustrated by the statement fa (concluded) enables[3] ib, indicating that, after each execution of Ta, Tb is allowed to execute up to three times. It is also possible to define that there is no restriction on the number of times a task may be executed after or during another (equivalent to define the parameter as infinite).
2.4 Blocking Interdependencies In order to enhance the flexibility of the model, it is also necessary to create the blocks and unblocks operators that, respectively disable and re-enable the execution of an event (second operand) when the state of the first operand is reached. The use of these operators, for example, allows for a new interpretation of Ta before Tb: ib (concluded) blocks ia – in this case, there is a restriction in the execution of Ta, which may not be executed anymore if Tb has already started its execution. There is no restriction on the execution of Tb (Tb does not have to wait for the execution of Ta, as would happen with the situation given by fa (concluded) enables ib). The blocking situations should be carefully used, since they could create deadlocks. 2.5 Resource Management Interdependencies According to the coordination model by Ellis and Wainer [11], there are two levels of coordination, one related to the activity level (temporal - the sequencing of tasks that make up an activity) and the other related to object level (resource - "how the system deals with multiple participants' sequential or simultaneous access to some set of objects"). Resource-related interdependencies may be represented by combinations of temporal relations. For example, if two tasks, Ta and Tb, may not use the same resource simultaneously, it is possible to define a "not parallel" dependency as the following statement, ia (ready) blocks ib AND fa (concluded) unblocks ib AND ib (ready) blocks ia AND fb (concluded) unblocks ia. However, besides being prone to deadlocks, this possibility ignores the notion of resource, which is quite important in the context of collaborative activities. Therefore, it is not sufficient to treat the problem of task interdependencies as a temporal logic problem. Moreover, considering resource management dependencies independently of temporal ones, a more flexible model is created, allowing the designer to deal with each kind of dependency separately. Resource management interdependencies in the proposed model are complementary to temporal ones and may be used in parallel to them. This kind of interdependency deals with the distribution of resources among the tasks. Three basic resource management dependencies were defined elsewhere [12]. Sharing - a limited number of resources must be shared among several tasks. Simultaneity - a resource is available only if a certain number of tasks request it simultaneously. It represents, for instance, a machine that may only be used with more
A.B. Raposo and H. Fuks /Defining Task Interdependencies
95
than one operator. Volatility - indicates whether, after the use, the resource is available again. For example, a printer is a non-volatile resource, while a sheet of paper is volatile. Each of the above interdependencies requires parameters indicating the number of resources to be shared, the number of tasks that must request a resource simultaneously and/or the number of times a resource may be used (volatility). 3 Coordination Mechanisms Some conclusions that can be drawn from what has been presented in the previous section are i) a simple group of seven temporal relations is exploded into numerous coordination situations that should be correctly treated by the coordination mechanisms of a collaborative system; ii) although encompassed by temporal interdependencies, resource management ones also are necessary and in) instead of defining coordination mechanisms for each interpretation of interdependencies, it is necessary to propose a direct mapping to construct the adequate mechanisms from the directives presented in the previous section. In this section the last issue cited above is addressed by means of the definition of a model for the construction of coordination mechanisms based on Petri Nets (PNs). The choice for PNs as the modeling tool is justified because they are a well-established theory (there are numerous applications and techniques available) and can capture some of the main features of a collaborative environment, such as non-determinism, concurrency and synchronization of asynchronous processes. Moreover, PNs accommodate models at different abstraction levels and are amenable both to simulation and formal verification. In the following some PN fundamentals are briefly overviewed, then the coordination mechanisms are presented. 3.1 Petri Nets Fundamentals PNs [13], [14] are a modeling tool applicable to a variety of fields and systems, specially suited for systems with concurrency, synchronization and event conflicts. Formally, a PN can be defined as a 5-tuple (P, T, F, w, MO), where: P = {Pl, .... Pm} is a finite set of places; T = {tl, ..., tn} is a finite set of transitions; F (P x T) U (T x P) is a set of arcs; w: F {1, 2, ...} is a weight function; M0: P {0, 1, 2,...} is the initial marking; with (P n T) = 0 and (P u T) 0. In a PN model, states are associated with places and tokens, and events with transitions. A transition t is said to be enabled if each input place Pi e t is marked with at least w(P i , t), which is the weight of the arc between Pi and t. It is also possible to define inhibitor arcs connecting places to transitions. In this case, the transition is enabled only if the places of origin are empty. Once enabled, a transition will fire when its associated event occurs. Firing transition t, w(P i , t) tokens are removed from each input place Pi and w(t, Po) tokens are added to each output place Po e t. Here, t and t means, respectively, the set of input and output places of transition t. A useful notation for PNs is the graphical notation (Figure 3) which is going to be used in the examples throughout this paper. In this notation, circles represent places, rectangles represent transitions, dots represent tokens and arrows represent the arcs (inhibitor arcs are represented with a circle on the edge), with weights above. By definition, an unlabeled arc has weight 1. In the PN of Figure 3, only transition t2 is enabled; tl is not enabled because it would require two tokens in P1 to fire, since w(P1, t1) = 2; t3 is not enabled because of the
A.B. Raposo and H. Fuks / Defining Task Interdependencies
Figure 3: Example of the graphical notation of PNs.
inhibitor arc from P3 (this place would have to be empty to enable t3). When t2 is fired, the tokens in P2 and P3 are removed and P4 receives one token. Note that the number of tokens in a PN is not necessarily conserved.
3.2 Petri Nets-Based Coordination Mechanisms In the proposed scheme, the design of a collaborative environment is divided into three distinct hierarchical levels, workflow, coordination and execution (Figure 4). In the workflow level, each participant's behavior is modeled separately, establishing the interdependencies between tasks of the same participant or those of different ones. The coordination level is built under the workflow level by the expansion of interdependent tasks according to a PN-based model and the insertion of correspondent coordination mechanisms between them. The environment model is simulated and analyzed at this level. The execution level deals with the actual execution of tasks in the system. During the passage from the workflow to the coordination level, each task that has an interdependency with another is expanded in the subnet presented in Figure 5. In this model, events i and f (start and end of the task) are represented as transitions, while states ready and concluded are represented as places connected to the respective transitions. After the firing of i, the flow is divided into two parallel paths, one indicating that the task is in execution – i(concluded) - and another representing the interaction with the task's execution in the system. The task execution is modeled by means of a transition with token reservation (represented with the letter "R"), which is a non-instantaneous transition tokens are removed from its input places when it fires and only some time later are added to its output places, representing the duration of the task. When considering two tasks related by interdependencies, it is necessary to correctly interconnect places and transitions of both models for creating the respective coordination mechanisms. In order to do that, it is necessary to define how to map the operators and parameters previously defined to the PN models. The model of the enables operator is quite simple. It is modeled by an arc from the place representing the first operand to the transition representing the second one. For example, ia (concluded) enables ib is represented by an arc from place ia (concluded) to transition ib. It is also necessary to add 1 to the weight of the arc arriving at the first operand, because this allows that place to enable both the normal flow of the task and the event given by the second operand. To illustrate that, consider the interdependency Ta equals Tb in the passive interpretation, i.e., ia (ready) enables ib AND ib (ready) enables ia AND fa (ready) enables fb
A.B. Raposo and H. Fuks / Defining Task Interdependencies
97
Figure 4: Coordination mechanisms design levels.
Figure 5: PN representation of an interdependent task at the coordination level.
Figure 6: Coordination mechanism for Ta equals Tb, passive interpretation.
AND fb (ready) enables fa. The coordination mechanism for this interdependency is illustrated in Figure 6. The enables operator also requires a parameter when it is necessary to enable a variable number of executions of the second task (as presented in Section 2.3). In this case, instead of adding 1 to the weight of the arc arriving at the first operand, it should be added the number given by the parameter. If that number is infinite, it is necessary to include a return
98
A.B. Raposo and H. Fuks / Defining Task Interdependencies
arc from the transition representing the second operand to the place representing the first operand. The forces operator requires an additional transition in the coordination mechanism. If the forced event is ready, then forces functions exactly like enables. However, if the forced event is not ready, it is necessary to include an alternative transition connecting the forcing state (first operand) to the output places of the second operand, indicating, in the PN model, that it has been forced to happen. Figure 7 illustrates the interdependency Ta meets Tb in the active situation when Ta is the master, given by fa (concluded) forces ib. In this mechanism, similar to that of enables, there is an arc from fa (concluded) to ib and an additional weight on the arc arriving at fa (concluded). Additionally, the alternative transition appears between fa (concluded) and the output places of ib. The alternative transition must have lower priority than the others in order to avoid that it be fired when the forced event is ready. In certain situations, it is necessary to include a delay parameter in the forces operator (Section 2.2). In these cases, it is possible to use timed transitions, which are fired only a determined time after enabled. The blocks operator is constructed exactly like enables, but using an inhibitor arc instead of a normal one. In this case, the block is permanent, because the extra token added to the blocking place is not removed. The unblocks operator is modeled by a transition that removes that extra token. Regarding resource management interdependencies, it is necessary to include a place whose tokens indicate the available resources. For example, Figure 8 represents the situation where two tasks share a single resource, represented as a token in place R. The start of each task depends on the availability of that resource. The end of the task releases the resource, if it is not volatile. 3.3 Example In order to get a general idea of the deployment of the proposed coordination model, a collaborative authoring activity is going to be modeled. Suppose that in this scenario there are two participants, the author (user A) and the reviewer (user B). The revision process should start after the beginning of the writing and may occur partially in parallel with it. Moreover, user B may alter the document to make corrections during the revision, but this may not conflict with the writing task of user A. The workflow level of the coordination model (see Figure 4) is directly obtained from the collaborative activity description. At this level, it is necessary to describe each participant's tasks and the interdependencies between them. In the scenario described, user A has the task of writing the document (writeA). User B has two tasks, the revision (reviewB) and the alterations (alterB). The interdependencies between these tasks are: writeA overlaps reviewB alterB during reviewB writeA and alterB sharing 1 The third interdependency above guarantees that the document will not be edited simultaneously by both users. In order to pass to the coordination level, it is initially necessary to choose which interpretations of the interdependencies are more appropriate in this situation. For the overlaps interdependency, it is not necessary that the writing task forces the revision, although it may not finish before starting the revision. In this situation, the passive
A.B. Raposo and H. Fuks / Defining Task Interdependencies
99
Figure 7: Coordination mechanism for Ta meets Tb, active interpretation – Ta master.
Figure 8: Coordination mechanism for two tasks sharing a single resource.
interpretation is adequate: iwriteA (concluded) enables ireviewB AND ireviewB (concluded) enables fwriteA AND fwriteA (concluded) enables freviewB. For the during interdependency, in this case, it is enough to say that ireviewB (concluded) enables ialterB AND freviewB (concluded) blocks ialterB. The definitions above allow the construction of the PN-based coordination mechanisms for this example (shown in Figure 9). This kind of model, in more complex scenarios, may be very important to detect unexpected situations in the collaborative activity. Furthermore, it could constitute the basis for the development of coordination mechanisms at the
100
A.B. Raposo and H. Fuks / Defining Task Interdependencies
Figure 9: PN coordination mechanism for the collaborative authoring example.
specification level, which interacts directly with the "real tasks" in the system. An implementation of coordination mechanisms at this level using software components that follows a PN model was presented elsewhere [15]. 4 Related Work In order to give support to tightly integrated collaborative activities, many coordination mechanisms have been proposed in the context of some collaborative systems. The first generation of coordination models, proposed in the mid-1980s, was restricted to specific scenarios, with rigidly defined protocols (e.g., [16], [17]). Eventually, there would be situations not predicted by the specified protocols, restraining the application of the defined mechanisms. Therefore, more recent models strive for flexibility, with coordination mechanisms that can be adapted for each application needs. The so-called second generation of coordination models looks for the development of systems with at least one of the following three characteristics, which are accessibility, interoperability, and flexibility. Accessibility is related to exposing the coordination mechanisms to system users rather than having them deeply embedded in the system implementation. In this case, users may be either system designers or end users. Examples of such a line of development include the Oval tool [18] and the ABACO/Ariadne [19]. Interoperability is an essential concept in the field of information systems integration, spanning from infrastructure services to business processes [20]. Originally concerned with the distributed and heterogeneous nature of modern systems, interoperability brings new concerns to coordination. Besides the obvious need to control activities defined within a
A.B. Raposo and H. Fuks /Defining Task Interdependencies
101
group (or intra-group coordination), the need for inter-group coordination may arise, a problem similar in nature to the integration of inter-organizational processes. Interoperability has been the focus of systems such as Reconciler [7], which aims to manage groups at the semantic level through the conciliation of conflicts. Flexibility focuses on the possibility of dynamically allowing redefinition and temporary modifications in the coordination scheme. An example of a system that incorporates the notion of flexible coordination mechanisms is Intermezzo [21]. In this system, data objects with controlled access are assigned to groups of users with a certain role, and these roles may change during the collaboration. The coordination model presented in this work may be fitted within the second generation because it offers a degree of flexibility through separation of tasks and interdependencies (facilitating the changes of coordination policies) and is adequate for dealing with some interoperability aspects. This is so in the sense that the interdependencies are generic (i.e., may be applied to a wide range of collaborative applications) and the implementation of coordination mechanisms may be realized by any tool. Although the use of PN-based coordination mechanisms has been stressed, the model clearly separates interdependencies from their coordination mechanisms, enabling the use of different implementation tools for the coordination mechanisms. Regarding accessibility, the proposed model exposes the coordination mechanisms to system designers, but not to the users. The idea of creating a set of task interdependencies and respective coordination mechanisms was proposed in the coordination theory of T. Malone and K. Crowston. They defined three types of elementary resource-based dependencies (flow, fit and sharing) and worked with the hypothesis that all other dependencies could be defined as combinations or specializations of these basic types [22]. The coordination theory was the inspiration of the genres coordination proposal, which stresses the coordination in relation to resources, place and time [23]. Another work that should be mentioned uses the interdependencies among activities to workflow management [24]. In this case, interdependencies are defined as "constraints on the occurrence and temporal order of events", and are controlled by coordination mechanisms defined as finite state automata, which guarantee that they are not violated. Some of these ideas originated the work presented here, which is refined by defining a larger set of basic interdependencies and modeling their respective coordination mechanisms.
5 Conclusion The necessity of coordination mechanisms to regulate interactions in collaborative systems has been the center of a heated discussion (e.g., [25], [26]). In spite of that, there is a recent trend to conciliate these ideas, trying to bridge the gap between coordination approaches for loosely and tightly integrated collaborative activities, arguing that both approaches are "seamlessly meshed and blended in the course of real world cooperative activities" [27]. In this context, this paper introduced an approach for the coordination of tightly integrated collaborative activities. In the proposed approach, a basic set of interdependencies was defined to encompass a large number of situations, including both temporal and resource management dependencies. The mapping from these interdependencies to PN-based coordination mechanisms was also shown. By means of the PN model, the collaborative environment may be simulated and analyzed, enabling the anticipation of possible problems. Moreover, considering that tightly coupled collaborative activities may be decomposed into a set of
102
A.B. Raposo and H. Fuks / Defining Task Interdependencies
interdependent tasks, such coordination mechanisms provide a framework for the design of collaborative systems. A continuation of this work is the further development of software components to implement the coordination mechanisms - an initial approach has been shown elsewhere [15]. The component model will standardize an event-based interaction between tasks and associated coordination mechanisms in an implementation independent manner. Another possibility to be explored is the use of fuzzy coordination mechanisms, which may bring a higher degree of flexibility and manageability to collaborative systems. Conventional temporal interdependencies do not accept scenarios such as starting a task execution when another one is "almost finishing," because tasks may only be synchronized by their starting or finishing instants. Such modeling imprecision is important because it offers application designers a higher degree of flexibility to focus on their customized version of the interdependency in a manner more closely related to subjective human reasoning. The fuzzy sets theory offers adequate resources to implement coordination mechanisms with such degree of imprecision [28].
Acknowledgements. Part of this work was realized while the first author was sponsored by FAPESP (Foundation for Research Support of the State of Sao Paulo), working as a postdoctoral fellow at the Dept. of Computer Engineering and Industrial Automation, School of Electrical and Computer Engineering, State University of Campinas, Brazil. The second author has a researcher productivity allowance from CNPq (Brazilian National Research Council), grant n. 524557/96-9. Thanks also to Profs. Leo Magalhaes and Ivan Ricarte for their helpful insights.
References [I] Schrage, M. (1995): No more Teams! Mastering the Dynamics of Creative Collaboration. New York: Currency Doubleday [2] Ellis, C. A., S. J. Gibbs and G. L. Rein (1991): Groupware: Some Issues and Experiences. Communications of the ACM, vol. 34, no. 1, pp. 38-58 [3] Fuks, H., C. Laufer, R. Cohen and M. Blois (1999): Communication, Coordination and Cooperation in Distance Education. In AMCIS'99. Proceedings of V Americas Conference on Information Systems, Milwaukee, USA, August 13-15, 1999. Association for Information Systems (AIS), pp. 130-132 [4] Schmidt, K. and L. J. Bannon (1992): Taking CSCW Seriously - Supporting Articulation Work. Computer Supported Cooperative Work (CSCW) – An International Journal, vol. 1, nos. 1-2, pp. 7-40 [5] Malone, T. W. and K. Crowston (1990): What is Coordination Theory and How Can It Help Design Cooperative Work Systems? In CSCW'90. Proceedings of the Conference on Computer Supported Cooperative Work, Los Angeles, USA, October 7-10, 1990. New York: ACM Press, pp. 357-370 [6] Malone, T. W. and K. Crowston (1994): The Interdisciplinary Study of Coordination. ACM Computing Surveys, vol. 26, no. 1, pp. 87-119 [7] Simone, C., G. Mark and D. Giubbilei (1999): Interoperability as a Means of Articulation Work. In WACC'99. Proceedings of the International Joint Conference on Work Activities Coordination and Collaboration, San Francisco, California, February 22-25, 1999. New York: ACM Press, pp. 39-48 [8] Allen, J. F. (1984): Towards a General Theory of Action and Time. Artificial Intelligence, vol. 23, pp. 123-154 [9] Allen, J. F. (1983): Maintaining Knowledge about Temporal Intervals. Communications of the ACM, vol. 26, no. 11, pp. 832-843 [10] Duda, A. and C. Keramane (1995): Structured Temporal Composition of Multimedia Data. In 1WMMDBMS. Proceedings of the International Workshop on Multi-Media Database Management Systems, Blue Mountain Lake, USA, August 28-30, 1995. Los Alamitos, CA: IEEE Computer Society Press, pp. 136-142 [11] Ellis, C. A. and J. Wainer (1994): A Conceptual Model of Groupware. In CSCW'94. Proceedings of the Conference on Computer Supported Cooperative Work, Chapel Hill, USA, October 22-26. 1994. New York: ACM Press, pp. 79-88
A.B. Raposo and H. Fuks / Defining Task Interdependencies
\ 03
[12] Raposo, A. B., L. P. Magalhaes and I. L. M. Ricarte (2000): Petri Nets Based Coordination Mechanisms for Multi-Workflow Environments. International Journal of Computer Systems Science & Engineering, vol. 15, no. 5, pp. 315-326 [13] Petri, C. A. (1962): Kommunikation mit Automaten. Schriften des IIM Nr. 3. Bonn: Institute fur Instrumentelle Mathematik [14] Murata, T. (1989): Petri Nets: Properties, Analysis and Applications. Proceedings of the IEEE , vol. 77, no. 4, pp.541–580 [15] Raposo, A. B., A. J. A. da Cruz, C. M. Adriano and L. P. Magalhaes (2001a): Coordination Components for Collaborative Virtual Environments. Computers & Graphics, vol. 25, no. 6, pp. 1025-1039 [16] Flores, F., M. Graves, B. Hartfield and T. Winograd (1988): Computer Systems and the Design of Organizational Interaction. ACM Transactions on Office Information Systems, vol. 6, no. 2, pp. 153-172 [17] Laufer, C. C. and H. Fuks (1995): ACCORD: Conversation Cliches for Cooperation. In COOP'95. Proceedings of the First International Workshop on the Design of Cooperative Systems, Antibes-Juan-lesPins, France, January 25-27, 1995. Rocquencourt: INRIA Press, pp. 351-369 [18] Malone, T. W., K.-W. Lai and C. Fry (1995): Experiments with Oval: A Radically Tailorable Tool for Cooperative Work. ACM Transactions on Information Systems, vol. 13, no. 2, pp. 177-205 [19] Schmidt, K. and C. Simone (1996): Coordination mechanisms: Towards a conceptual foundation of CSCW systems design. Computer Supported Cooperative Work (CSCW) – The Journal of Collaborative Computing, vol. 5, nos. 2-3, pp. 155-200 [20] Hasselbring W. (2000): Information System Integration. Communications of the ACM, vol. 43, no. 6, pp. 33-38 [21] Edwards, W. K. (1996): Policies and Roles in Collaborative Applications. In CSCW'96. Proceedings of the Conference on Computer Supported Cooperative Work, Boston, USA, November 16-20, 1996. New York: ACM Press, pp. 11-20 [22] Malone, T. W. et al. (1999): Tools for inventing organizations: Toward a handbook of organizational process. Management Science, vol. 45, pp. 425-443. [23] Yoshioka, T. and G. Herman (2000). Coordinating Information Using Genres. Center for Coordination Science, Sloan School of Management, MIT, Working Paper CCS WP#214. [24] Attie, P. C., M. P. Singh, E. Emerson, A. Sheth and M. Rusinkiewicz (1996): Scheduling workflows by enforcing intertask dependencies. Distributed Systems Engineering Journal, vol. 3, no. 4, pp. 222-238. [25] Suchman, L. A. (1994): Do Categories Have Politics? Computer Supported Cooperative Work (CSCW) An International Journal, vol. 2, no. 3, pp. 177-190 [26] Winograd, T. (1994): Categories, Disciplines, and Social Coordination. Computer Supported Cooperative Work (CSCW) – An International Journal, vol. 2, no. 3, pp. 191-197 [27] Schmidt, K. and C. Simone (2000): Mind the gap! Towards a unified view of CSCW. In COOP 2000. Proceedings of the 4th International Conference on the Design of Cooperative Systems, Sophia Antipolis, France, May 23-26, 2000 [28] Raposo, A. B., A. L. V. Coelho, L. P. Magalhaes and I. L. M. Ricarte (2001b): Using Fuzzy Petri Nets to Coordinate Collaborative Activities. In: Proceedings of the Joint 9th IFSA (International Fuzzy Systems Association) World Congress and 20th NAFIPS (North American Fuzzy Information Processing Society) International Conference, Vancouver, Canada, July 25-28, 2001. Piscataway, NJ: IEEE, pp. 1494-1499
This page intentionally left blank
Cooperative Engineering
This page intentionally left blank
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press, 2002
107
Engineering CSCW Tony LAMBIE & John LONG Ergonomics & HCI Unit University College London 26 Bedford Way, LONDON WC1H 0AP
[email protected] [email protected] Abstract There is a troublesome, and unresolved, tension at the heart of HCI, and Cognitive Ergonomics generally. How might we reduce the problem of complex HCI or socio-technical system design to manageable components, without dehumanising the agents and producing analyses which lack cogency and/or faithful representation? We take a proposal from the literature as a starting point in order to introduce a novel idea which might allow for such an analysis of a socio-technical system - an idea which permits analysis as deeply as required without making assumptions of a reductive nature, but an approach consistent with the structured methods of software engineering. The paper exposes a theoretical framework element of a cognitive nature, and argues that such an element might allow the designer to reason about collaborative work in a manner sensitive to social factors but in a way which might still allow integration with a cognitive engineering approach. Keywords Theoretical Study; Cognitive Engineering; Coordination Mechanism; Distributed Cognition; Ethnography; Task Analysis
1 Setting the Scene Cooperative systems of work are not exclusively based on a mutual altruism (the word 'cooperative' is suggestive). That is to say, individuals within such a system may compete and come into conflict as an integral part of working together. In situations where people carry out collaborative work, the question is: What are the most effective interactions, in order to meet the system requirements, involving both willing cooperation and some competition? If systems such as urban traffic are considered, it is clear that people are competing, as well as cooperating in an accommodating way, with one another, not only within but between classes of road-user - cars, bikes, pedestrians etc. Optimising these interactions, i.e., making them as effective as possible given agreed requirements of such urban traffic systems, will sometimes involve recognising that there may be contradictory perspectives on what action to take in a given situation. These perspectives differ depending on such things as the different roles (car driver, cyclist, pedestrian): whatever qualifies as different points of view, as it were, in the traffic system. To take an example from this domain, consider quite a usual scene in the UK where one drives on the left: at a junction controlled by traffic lights, and after which the road narrows to cope with only one lane of traffic (and where there is no right turn), there are no indications on the road before the junction that road-users should occupy any particular lane, i.e., there is no specific instruction that the user of the left hand lane must only turn left. What normally happens is that, in the right hand lane, a 'queue' forms which is made up of people planning to go straight ahead (aware that the road narrows after the traffic
108
T. Lambie and J. Long / Engineering CSCW
lights), and the left lane is used exclusively by those turning left. Normally, that is, until newcomers to the district take up position in the left hand lane but intend to go straight ahead. The result is often that the drivers in the outside (right hand) lane speed away (angrily) to 'beat' the incomers; or if the incomers get ahead, the local drivers use their horn, and often drive aggressively close, believing that the object of their anger has 'jumped the queue', which the incomer was not aware of doing. This simple situation is complicated by drivers (residents, this time) who, being in a hurry, now assert their 'right' to occupy both lanes, since there is in fact no indication to the contrary, and consequently cause the same situation of potentially dangerous conflict. This state of affairs, i.e., the junction as it is currently configured, could be changed to resolve this problem of apparently ineffective interactions. In order to put the road-users in some state of consonance, changes could be made to the road signing: one clear way of solving the problem might be to forbid the use of the left hand lane to those going straight ahead, painting it with a series of left oriented arrows starting well ahead of the junction; another might be to place a very visible sign announcing that the lanes converge at the junction and that cars should merge alternately. These are two sets of behaviour which could be specified and which might resolve an ineffective section of the road-users' system. However, there is one other aspect of the situation to be borne continually in mind: the extent to which the apparent ineffectiveness is real with respect to the whole system and the agreed requirements of that system. In the case outlined, which optional re-design might be adopted would depend on some measure of performance. For example, whether or not the 'solution' were a road sign indicating that cars should merge alternately from two lanes to one might depend on the effect this new behaviour would have on slowing down the rate at which the traffic passed the junction relative to the old system or, indeed, relative to any other optional re-design. This measure is, in its turn, relative to other requirements as one's view of the system is broadened to take in the most general considerations. 1.1 Features of Approach What we have portrayed in a graphic way is a multi-agent problem where a global view of cooperation is adopted which involves the intentions (not only conscious) of group behaviour (of co-workers in the traffic system) as understood from their overt behaviour as they carry out road-user tasks. It is their potentially inconsistent view of these tasks which has to be resolved, and it need not be seen as peculiar to such a system but as an essential feature of any reasonably open (i.e., one not hobbled by rules of procedure) collaborative enterprise. What is also portrayed is an approach which attempts to determine a design solution based on what leads to more effective behaviour by users relative to some threshold of performance. That is to say, it is an approach which takes into account scarce resources and measurable performance, increasing the latter and reducing the call on the former. If the approach can be systematic in the manner of its analysis of the problem and its synthesis of a solution, then such an approach would be an engineering one. As well as being systematic, the approach outlined also has the potential to offer solutions which overcome the limitations of more piecemeal analyses: to offer solutions with a more general application - also an attribute of engineering. It is well recognised [1] that some phenomena result in unsatisfactory road improvement policy: the phenomenon of 'risk homoeostasis', where the road-user will compensate for improvements made to the
T. Lambie and J. Long / Engineering CSCW
109
road or to car safety features by driving or riding faster, therefore maintaining the same overall risk of danger; and the phenomenon of accident 'black spot' migration, where reducing accidents in one place appears to lead to an increase elsewhere. If the recommended changes derive from an understanding of what the road-users want or need relative to one another and to other features of the traffic system, the re-design can be made to respond to more fundamental features of road-user behaviour, which, depending on the granularity of analysis, may apply more generally across the system. As implied above, traffic systems have too often been treated as not only microergonomic as opposed to macro-ergonomic [2], i.e., in terms of the individual driver rather than the aggregate effect of individual drivers, but also that the human-human interface as well as its aggregate effect has been largely ignored. In the account which follows, this aspect of collaborative work - the requirements of the group, as distinct from that of individuals which make up the group and their relationships - will be explored; and a structure will be exposed which might help designers to reason rigorously about group work without sacrificing subtle human interaction. 2 Introduction The intention is to introduce a recent conceptual development [3] which should support the representation of cooperative work in the manner hinted at in the road-user system example above; and which, together with the appropriate method or methods, should also result in the definition of design problems with such a social aspect, the elicitation of the relative user requirements, and ultimately the specification of their design solutions. The paper derives from an application of this conceptual framework in conjunction with an established design method [4] to the problem of eliciting teamwork requirements . In the early '90s, a paper [5] by Sommerville & Rodden was published which highlighted significant deficiencies in the (then) current processes to identify user requirements for cooperative work systems. In general, it was claimed that these deficiencies were rooted in structural features of software engineering, and endemic in the professional/academic organisation which might undertake such investigations: in particular, that "the current procurement model for software systems is deeply embedded in our organizational structure.... It is unlikely that existing approaches to requirements engineering will be discarded in favour of 'user-centred design', ethnography or any other 'human-centred' approach. CSCW systems will have to be developed within this framework." (our italics) It was the aim of Sommerville & Rodden's paper to propose that software engineering could be enhanced in its quest to identify user requirements with the aid of ethnography because "ethnographers and other behavioural scientists are trained in analysis and evaluation. They try to avoid making judgements during the analysis process. By contrast, engineers are trained in design and synthesis - making judgements and formulating abstractions are fundamental aspects of design." Like the authors in [6], Sommerville & Rodden are aiming for a multidisciplinary solution to a complex design context: a pragmatic solution which "prizes a concern with human 'real world' activity higher than the purity of any particular model or discipline"; and they go on emphatically, "and so it should be in any applied science, such as humancomputer interaction". The principal issue which arises is whether such a pragmatic solution offers a satisfactory continuity of representational means which will allow an effective design dialogue to take place between the ethnographers' work and that of the software designers;
I 10
T. Lambie and J. Long / Engineering CSCW
and, in any event, the praiseworthy concern with "human 'real world' activity" rests on the charitable cooperation of groups who are often at odds, and may choose to be again. A secondary issue, which we cannot deal with in depth here, is whether part of the problem of the incompatibility of approaches such as ethnography and software engineering evident in much of CSCW lies precisely in treating HCI and CSCW as applied sciences, as we believe Sommerville & Rodden do. It is implicit in our approach that Cognitive Engineering design knowledge is distinctive and its terms of reference do not bring it into conflict with descriptive or explanatory approaches to social behaviour which acknowledge human intentionality. What is needed is an analysis which provides a means of characterising a systematic view of CSCW design: one which does not diminish or limit input of an ecological nature; and one which is principled rather than pragmatic. 3 Essentially Cooperative Work The entity which best exemplifies the emergent quality of group behaviour is the team - the epitome of a group cooperating in a largely unplanned and reactive manner (because of the speed and/or unpredictability of the changing situation), but working to some welldefined end. Alternatively, it is exemplified by a group working towards some end which is not clear and where the means are consequently not clear either. Football teams, on the one hand, and research and design teams, on the other, instantiate the two kinds of teamwork. Research and design teams are characterised by having ill-defined goals which have to be arrived at in a manner which is, in the nature of the case, unclear, and it is this process which is supported by effective teamworking; and, although in the case of the former - the football team - the aims are clear, the speed (and relative unpredictability) with which the joint action must be executed prevents the cooperation from being other than tacit. The common feature, therefore, of these 'essentially' cooperative groups is that the work process is complex and emergent, and the problem of 'managing' (by explicit means) their cooperative behaviour appears intractable. If we can offer some conceptual and methodological support which might address such rich settings then we may have something to offer the CSCW community more generally, something which, moreover, integrates the systematic with the social in order to design more effective socio-technical systems. As we have suggested, if a work goal is well defined, at least one thing that might mean is that the link between the current state of the work domain and the desired state is understood, and that the means to achieving it are clear. Working with others to reach that goal might then be described as merely coordinating the work tasks of the participants to that end. However, the goal of design work is (by its nature) not defined at the outset. The expression of such work is problematic and is further confounded when one is faced with the challenge of representing the collective behaviour of a number engaged in some such design work. Defining the design problem, it is often agreed, is moving a good way towards solving it. Therefore, at least a significant part of the difficulty in conceptualising the cooperative work involved in design resides in the link between any cooperative behaviour and that goal of design, which is ill-defined to begin with. A conceptualisation ("the recent conceptual development" mentioned above) which makes this link more explicit is one way of supporting the business of cooperative design.
T. Lambie and J. Long / Engineering CSCW
111
One might say, in sum, that there is an inverse relationship between the emphasis on close and intense cooperation and the explicitness of the goals and sub-goals of the associated work. 4 Views of Coordination and Work CSCW has been dominated by views which have an anthropological or sociological origin. These views are appealing because they address a less constrained, but arguably more realistic, set of problems than those traditionally tackled by human factors specialists: respectively, in vivo and in vitro. And the single-mindedness with which these sociallyoriented practitioners approach the dynamic complex process of group behaviour means that they do not feel the paradigm of the model or framework appropriate for such a protean phenomenon. It is this reluctance, we believe, which is largely responsible for the difficulty faced when attempting to integrate the conventional systems development work with an understanding of the part played by the joint activity of those carrying out the work, since the notion of the model brings with it the possibility of exploring a systematic relationship between group behaviour and the work to be done. The challenge is to promote the idea of a framework or model to aid design reasoning, without creating a new, different and more constrained problem: in short, to manage the design problem without reducing it; and at the same time, to maintain a consistent and systematic overall view so that a common basis might be established between the different stages from the understanding of the problem to the setting out of a solution. In addition to a uniform approach, this common framework or model might support a means of communication between different experts to enhance mutual understanding throughout the different stages of the problem formulation and solution. 4.1 Some Frameworks or Models Schmidt's conceptual framework of cooperative work [7] is a comprehensive review of the aspects of collective behaviour at work, and covers its gross and obvious as well as its fine and subtle features, and Schmidt, indeed, has something interesting to say about the core notion of cooperation which we believe is associated with the idea of teamwork. However, just as we shall not address classes of coordination, such as 'augmentative', 'integrative' etc., nor the work organisation nor the modes of cooperation (all dealt with in Schmidt's paper), except incidentally, so we are not concerned with whether the cooperation is between experts and experts or experts and non-experts, as [8] Falzon's paper is. Likewise, whether design is not contingently a cooperative activity but essentially and necessarily a collective one, as [9] B0dker asserts ("Design is a collective activity...."), need not detain us here. All these writers recognise the collective nature of work, but none offers a framework which underpins the workers' interaction, nor do they try to forge a connection between the systematic (and established) activity of the software engineer and this multiple and mutual interaction. Falzon, for example, deals with it as an instance of dialogue governed by the general principles of dialogue, and his treatment is of course nonetheless useful for that. The mechanics of the dialogue of any cooperative activity must be understood and their representation is going to be a key component in any specification or diagnosis of group work effectiveness. However, as [10] Brehmer points out, "cooperation can be achieved without communication". That is to say, we can look below the dialogues for the principles of cooperative behaviour, and, arguably, we can go further as [7] Schmidt does when he writes, "conceptualisation is in principle a coordination of viewpoints": i.e., that
112
T. Lambie and J. Long / Engineering CSCW
individual and collaborative work behaviour is fundamentally inseparable in group work. It is in this region that we shall look for a 'device' which will help us to reason about cooperative work design problems. [11] Storrs clearly deals with some sort of group work scenario but his paper constitutes, as he himself describes it, an ontology and, thus, is more descriptive and classificatory than ours, which will attempt to account for, or provide a rationale for, the multiparty interaction - in the interests of the systematic specification of teamwork design, i.e., it comprises a prescriptive view. Others, such as [12] Rasmussen and [13] Vicente, who aspire like us to a discipline of engineering, deal with group interactions to carry out work, and provide frameworks of a kind. However, we shall show how our conceptualisation (not to be confused with Schmidt's use of the word) is different from theirs, and while stressing the importance of the notion of the task, will extend and re-interpret the idea. Before we proceed with the introduction to the conceptualisation of cognitive coordination, we should point out that it is an addition to, or an elaboration of, a conception of design peculiar to the Ergonomics & HCI Unit at the University College London (UCL) (see [14] & [15]) - a version of cognitive engineering. We shall, therefore, start out with a brief description of the background against which the conceptualisation has taken place. 5 Background to the Conceptualisation 5.1 Performance and Engineering There are two main issues to be addressed when one considers the part the agents play in any work domain - fundamental, but not tantamount, to cognitive engineering: firstly, how they relate to that work domain, i.e., how the interface should be represented, at least for the designer; and, secondly, how they relate to other agents in the work environment. Perhaps the essential feature of cognitive engineering, as we understand it, is that it is a cognitive design activity which takes place against a background of scarce resources, as other kinds of engineering do. Thus, in order that a design problem can be defined and solved, a clear understanding is required of what constitutes performance. For this to be possible, agents doing work are, to one degree or another, conceived as set over and against the work domain (from the point of view of the designer, at least), in order to facilitate reasoning about the design problem, e.g., to consider the merits of different specifications of agents' behaviour to achieve the same work result, thereby answering the requirements. It is true that there is a dynamic and reflexive relationship which develops between the work being done and the work to be done, and that the agent contributes to this developing situation, i.e., the workers transform the object of their work, and consequently have to deal with a new object, sometimes with unanticipated properties. However, in order to get something done to advance the problem, i.e., to facilitate design reasoning, a decision has to be taken as to where the dividing line is between the worker/s and the domain of work. Thus a distinction is made between the work to be done and the means of getting that work done; the design problem being how to improve the means in order to get the work done more effectively. Enhancements to both must be relative to some criteria. These criteria, pertinent to each, are the resource costs of implementing the means, in the case of the agents' behaviour, and the quality of what is done, in the case of the domain. That is to say, the doing of the work implies commitment of time and effort and, for the agents, the aim is obviously to make these acceptable in the context of the requirements and the constraints of the setting. On the other hand, the work has to be performed as desired. The
T. Lambie and J. Long / Engineering CSCW
113
simplest way of putting this is to say that the costs of the work and its quality have to be weighed in the balance; and it is that balance which constrains the process of finding a design solution. To put the argument conversely, unless we have a clear idea of the agents' behaviour and the domain of work, we cannot accurately and consistently measure the costs of their behaviour relative to the quality of work and, therefore, our design solution, or solutions, would be baseless, i.e., we would have no way of measuring performance. As Dowell & Long [14] [15] have it, the conception of the design problem involves the fundamental duality of agents' behaviour and the transformation of the domain contents, its objects and attributes. These are, respectively, the Interactive Work System (IWS) and the Domain of Work (the Domain) (Figure 1). This duality is the basis for what issues as the measure of performance; and is why we can consider such a design problem specification to be, potentially, an engineering one. It is important, however, not to conclude that this separation, which is well-defined in order to permit such a calculation, leads to a radical separation of the IWS from the Domain. 5.2 Design, Human Factors, & the User's View of the Work As [15] Dowell & Long have also indicated, referring to [16] Simon and [17] Neisser, it makes no sense to consider one in isolation of the other. They write, "If the worksystem is well adapted to its domain, it will reflect the goals, regularities and complexities in the domain". Thus, what we are considering is a conception which not only permits a clear specification of a design solution determined by the calculation alluded to above but also, in spite of that rational basis, is arrived at through an understanding of the values of the work to be done. Which is to say that it is not, in any sense, a solution dictated by factors, or knowledge, external to the design problem, such as that, for example, drawn directly from the psychology of cognition or perception, though such knowledge might be the inspiration for a conjectured design model to be determined as engineering knowledge by systematic design practices. The solution is, in short, at once domain-driven and IWS constrained (in a way which awaits a thorough knowledge of the domain and user behaviour). In order that there is a means of connecting these limiting forces during the design problem-solving (i.e., the forces of motivation, or goal-seeking, on the one hand, and the constraints, representing the costs of so doing, on the other), there has to be a semantic connection between the IWS and the domain. This semantic connection is that of the intentions of the agents and the goals in the domain. These are linked, in the sense that the intentions result most of the time in the transformation of states of the domain of work. In general terms, intentions in the work system are continuous with the goals in the domain.
114
T. Lambie and J. Long / Engineering CSCW
Figure 1: World of Work (based on Dowell & Long [14])
That is to say, an intention is not an intermediate goal to be fulfilled before the domain goal is specified or fulfilled. However, though they are one in this sense, they are not one and the same, because the structures which support the behaviour of the worksystem and those which underlie the possible transformations in the domain are different. The 'reflections' which [15] Dowell & Long allude to, in the well-designed system, amount to a congruence between these distinct structures. It is this double aspect of semantic continuity and syntactic disjunction which allows us to maintain the intimate connection between the users and the work, while rigorously separating them so as to define, and tackle, the design problem in a systematic manner. The relationship, then, between the IWS and the Domain can be articulated (a) so that performance may be measured, and (b) so that a specification of the semantics of the design problem may be made and the solution may be addressed. 5.3 Human Factors and Joint Cognitive Behaviour However, the relationship between the agents in the IWS is not one of means to ends. If this were the case, the agents would be either objects in the Domain or form part of the interface; and, in general, when we refer to agents and the work they do, we mean that they are essentially all active and autonomous agents. We need, thus, to know in what such cooperation consists in order to provide an explicit representation of it. It is the intention of this paper to introduce the means to this representation of cooperation, consistent with a cognitive engineering framework. In other words, the paper will introduce a structure which supports joint behaviour in the IWS, permitting the inclusion of social aspects into both the definition of the design problem and the specification of the design solution. Although it is intended to fulfil a cognitive and a social function at one and the same time, our approach is still able to aspire to the title of engineering because of its emphasis on performance as the accomplishment of effective work of a desired quality at the expense of acceptable costs. It is often assumed that such an approach is reductive and mechanistic, and is unable to accommodate adequately the subtler concepts of cognition. It might be
T. Lambie and J. Long / Engineering CSCW
115
thought, for example, that such an approach might work if it is limited to the simplest case, but that CSCW would be beyond it, because in the class of socio-technical systems, for example, teamwork tends very much to the social end of the spectrum, exhibiting strong social influences less constrained by technical knowledge and composed of the various coworkers' diverse sets of values or perspectives. Vicente [13] points this out but fails finally to deal with just those values and perspectives. The development of our conceptual framework for cooperative work should offer just this possibility. 6 Other Engineering Approaches There are other workers in the field of human factors who also espouse an engineering approach, notably, when one considers the question of the behaviour of collective systems: researchers in cognitive systems engineering (CSE), and others comprising Rasmussen, Vicente, Woods, Hollnagel, etc. Their work is very rich and provides many means of representation for cognitive design purposes which assist design problem expression. They, for example, offer levels of description of agents' behaviour in terms of skills, rules and knowedge, and as such their means of representation of this behaviour is consistent with another group of human factors workers, those espousing activity theory (AT) (see [18] Draper). However, as we have pointed out above, these distinctions, in the first place - and with particular respect to, CSE & CWA - concern individual agents, and do not relate those representations in social terms; in the second place, and with respect to AT, although the group shares the cognition, the AT theorists can only advance design knowledge to a limited and uncertain extent without an explicit framework to support this collective cognition; and in the third place, none of the above groups has an explicit means of measuring performance, the pre-requisite of effective design products, in the sense pertaining to our cognitive engineering approach, outlined above. Vicente and the CSE workers, in addition, have a concept of task which is left less well specified apparently in order to leave room for evolutionary design development, i.e., not to limit the designer's freedom nor that of the worker implicated in the designed system. However, we believe that it is possible to specify the task clearly as long as one is not confined to the conventional level of work (or technical) tasks; that is to say, if the notion of task is deepened to comprise the cognitive process, and is understood as shared in a more fundamental way (this should become clearer in our conceptualisation of cognitive coordination). Vicente [13] argues for a review of the concept of task analysis, and we do take account of this difficult concept. However, we feel that there is no need to make too much of Vicente's distinction between "instruction-based" task analysis and the "constraintbased" form, but simply to recognise that task analysis must always be qualified by the limitations of domain knowledge*. In other words, we hold to the position that a normative (i.e., a systematic and prescriptive) approach to design is required, but do not conclude, as Vicente does, that this implies dictated or rigid design solutions. The reason we are in a position to do this is that we take the view that such rigidity, which we agree is a danger, follows from applying, for example, psychological knowledge and failing to recognise the * 'Instruction-based' task analysis (TA), for Vicente, leads to a prescription of what tasks should be carried out and in what order, e.g., sequential flow and timeline TA techniques; a 'constraint-based' version indicates what tasks should not be performed, e.g., input/output TA techniques
116
T. Lambie and J. Long / Engineering CSCW
particularity of the design problem; as well as from assuming that the way things are done determines the way things should be done. We make no such assumptions and pursue neither scientific knowledge of workers' behaviour nor descriptive/explanatory accounts of the work under investigation, but engineering. Thus, though our output is prescriptive it is not a prescription determined either by the way things are done in some domain of work or by the scientific findings of psychology, which may or may not be appropriately scoped for the design problem under investigation. This tendency to lean heavily on psychology, we believe, stems from the perceived need for science to underwrite the guarantee of good design knowledge. It is taken for granted that secure knowledge is warranted by science, but as Popper has pointed out, even science needs to establish its knowledge through its practices: the hypotheses are bold speculations [19]. As we are trying to secure engineering knowledge we must, analogously, establish it through engineering practices, which are distinct from those of science and have different goals: diagnosis and prescription in the case of engineering, and explanation and prediction in that of science. To begin with, then, and analogous with Popper's description of the acquisition of scientific knowledge, we start with a 'conjecture' (or 'bold speculation') - the conceptualisation - but we do so with the intention of integrating it within a larger framework of cognitive engineering and, crucially, with the intention of operationalising, and ultimately testing and generalising some solution to the design problem. After the conceptualisation - the exposition of the rationale for the conceptual framework - there follows what counts as the next step in the acquisition of early design knowledge: seeing how it assists the analysis of an appropriate design problem. We might call this an intermediate stage between that of conceptualisation and that of operationalisation properly speaking: a stage where the concept benefits from feedback provided by its employment in the diagnosis of design issues - in this case, coordination issues. The operationalisation is double-sided: the framework, for its part, needs operationalising (through its application) as part of the process of validation; and, conversely, the properties, in this case of teamwork, need the framework so that the design problem can be made comprehensible and tractable, such that that reasoning (and dialogue between the designers) about a design solution can be facilitated by the models being operationalised. 7 Conceptualisation of Cognitive and Social Coordination Shaw & Fox [20] consider three types of system which might be thought of as sharing the task of problem-solving: collaborative reasoning systems; distributed problem-solving systems; and connectionist systems. Among the considerations which these types of group reasoning share (i. e., a component of a common framework for such systems) is what the authors call the 'coordination mechanism': "since each problem-solving agent only possesses a local view and incomplete information, it must coordinate with other agents to achieve globally coherent and efficient solutions". They believe that "the design of coordination can be viewed from three different perspectives: the information content, the exercise of control, and the coordination mechanisms". They go on to state that "coordination can be achieved through passing different types of information among the agents, such as data, new facts just generated, partial solutions/plans, preferences and constraints". We may think, then, of these as the objects or contents of the coordination process. The authors go on to suggest that "the initiative to coordinate may result from a
T, Lambie and J. Long / Engineering CSCW
117
variety of means of control: it may be self-directed, externally directed, mutually directed, or a combination of them, e.g., coordination by synchronisation, coordination by negotiation etc. Whereas, therefore, the data etc. which is passed could be thought of as the content or object of the process, the "means of control" might be understood as the manner of passing and its potential impact. If we look at Schmidt's conceptual framework [7] we find him asserting, as we have seen, that "conceptualisation is, in principle, a coordination of viewpoints", that "conceptual thinking is a cooperative effort". We come then, in his view, and from the start, enabled for coordination; and there is no inherent difficulty for us in coordinating: we start with a primitive basis for mutual knowledge, without which we could not communicate, i.e., it allows us to coordinate our attempts to communicate. It is part of what it means to have a conceptualisation of the world that it is shared thus (cf. Schmidt's earlier reference), and it is quite unlike that of the artificially intelligent agents under discussion by Shaw & Fox, which require an explicit input under something like the categories of 'object' and 'means of control' - what they call the "coordination mechanism" in order for them to be coordinated (Figure 2). Cooperative behaviour presupposes a common or shared view - a fundamental coordination, such as Schmidt writes about; and, likewise, [21] Hutchins writes, "the distribution of labour can only be negotiated if the distribution of knowledge and ability is at least partially redundant". What we may now consider is how this natural, and fundamental, coordination which Schmidt writes about can be characterised in a more formal and explicit manner by reference to the idea of the 'coordination mechanism'. If this is possible we could be better equipped to articulate problems concerning the design of group behaviour.
Figure 2: CM & the Speech Act
What we are looking for is a means of amalgamating the analysis of Shaw & Fox (concerned with Distributed Artificial Intelligence (DAI)), which exposes the elements of a coordination mechanism without the assumption of mutual knowledge, with something like the idea of Distributed Cognition (DC). There is, in fact, a convenient concept associated with linguistic behaviour which has striking similarities in structure and in function, we shall argue, to the coordination mechanism as exposed by Shaw & Fox: that of the Speech Act [22] & [23]. The Speech Act is a structure with two components, the 'illocutionary'
118
T. Lambie and J. Long / Engineering CSCW
and the 'prepositional1, where the illocutionary might be the act of commanding or asserting something etc., and the prepositional content might be the representation of some state of affairs commanded or asserted. Together, such components might be exemplified by the imperative 'put the cat on the mat' and by the utterance, 'the cat is on the mat1. The first is the imperative, that the cat should be put on the mat (identified by the action word "put"); and the second is the assertion, that the cat is on the mat (signifying 'It is the case that the cat is on the mat, perhaps informing someone of the fact). Even though this assertion seems indistinguishable from an exposition of its content (i. e., the prepositional component) it is, nevertheless, not merely a representation of a possible state of affairs because it is a speech act, and as such may have consequences. Speech Act theory rests on the premiss that speech can be understood as another form of action. It is easy to see, then, that these components could be described in terms of the content or state of affairs, the cat being on the mat, and the exercise of control, some communicative act over the content or state of affairs: in other words, an analogue of the coordination mechanism as described by Shaw & Fox (see Figure 2). In the first case - the statement of a fact - the exercise of control is in the form of an assertion of the state of affairs expressed in the proposition, the cat's being on the mat, and can be true or false; in the second, it is in the form of a command that the state of affairs should exist and that someone should effect it. Two things should be emphasised, and the one leads on to the other. The first is that these utterances are expressed for a purpose in the normal way of things. A statement of facts is made, for example, to satisfy someone's curiosity, perhaps allowing him or her to make or modify some plan of action: to use the vacuum cleaner now or later. A command is, more evidently, also part of a chain of actions. They are, therefore, actions and they are goal-oriented. The second thing which needs to be taken note of is that, as such, they have to be coordinated, like other group actions, for communication to take place at all; and that, as Clark [24] succintly puts it, "language use and joint activity are inseparable". So, what we are suggesting is that it may make sense to suppose a more fundamental structure - the coordination mechanism (hereafter, in its adapted form, the CM) - which is goal-oriented, might underpin models such as distributed cognition, and may support all of what Clark means by 'joint activity' (including the use of language) and comprising coordination in CSCW, i. e., working or behaving together effectively. The difference alluded to above between the artificially intelligent agents and their human counterparts is that in the case of the AI agents "each problem-solving agent only possesses local view", whereas, at the very least, human agents often share knowledge publicly and transparently. It is this sharing which allows them to coordinate - even to communicate - on the basis of recognising objective states of affairs and of employing mutually understood means of controlling them, including the resolution of contradictory perspectives. Thus, to bring together the observations of Hutchins and Schmidt: Cognition is already distributed for conceptualisation to take place, and the CM is the formal structure which can account for this distribution in the familiar terms of goal-oriented acts; and, at the same time, it can bring together the terminology and conceptual apparatus of Distributed Cognition (DC) and Distributed Artificial Intelligence (DAI) (see Figure 3). To distinguish this CM from the speech act structure (and from others who have employed the term) we shall refer to the element which exercises 'control' (in Shaw & Fox terminology) as the 'attitude'; and the object, data or representation of a state of affairs (designated 'information' by Shaw & Fox) will be referred to as the 'content'. It is useful
T. Lambie and J. Long / Engineering CSCW
119
then to begin thinking of the coordination of social behaviour as on the same plane as the coordination of language in the sense that they are both species of communication. In order to characterise the mutual knowledge underlying this fundamental communicative process one can, indeed, adapt the famous account of non-natural meaning (i. e., the understanding of the intentions of others) by Grice [25] as follows: "The speaker meant something by x' is (roughly) equivalent to 'the speaker intended the utterance of x to produce some effect in an audience by means of the recognition of this intention'". If we substitute 'speaker', 'utterance' and 'audience' by 'actor/agent', 'execution', and 'spectator', respectively, we get: '"The actor/agent meant something by x' is (roughly) equivalent to 'the actor/agent intended the execution of x to produce some effect in a spectator by means of the recognition of this intention'". We can see, therefore, that it is possible to embrace indifferently both language and more general social activity with the more fundamental and general concept of the CM as an expression of a mutually understood state of affairs or 'content' governed by a meaningful 'attitude' in language or in behaviour, more generally. We have referred to simple cases, with reference to the speech act, of assertion and command, but all manner of attitudes are circumscribed, such as hope, desire, belief etc. Other 'attitudes' are more difficult to characterise, such as those alluded to in the section, 'Setting the Scene' at the beginning of the paper: that of arrogating a right to oneself, of expecting deference and so on: the kind of 'attitude' associated with social interaction of a "The distribution of labour can only be negotiated if the distribution of knowledge... is at \ least partially \ redundant. "
each agent only possesses local view... " (Socially based Cognition)
Distributed Cognition
(Individually based Cognition) Distributed Artificial InteUigence
(Hutchins, '89)
(Shaw & Fox, '93)
With the help c f the Speech Act...
Coordination Mechanism (CM) Attitude(content) Figure 3: DC+DAI=CM for Language & other Joint Activity
general nature. To return to the traffic system illustration, an instance of the CM's employment as an analytical tool would be to allow the identification of conflicts between road-users in terms of their differing 'attitudes' over a given state of affairs or 'content'. If, for example, the same road users expressed contradictory attitudes when acting out different and conflicting roles in a given situation, then it would provide a rationale for redesign to bring the 'attitudes' associated with the different roles into greater harmony through road re-design, sign design or modified training.
120
T. Lambie and J. Long / Engineering C5CVV
8 Conclusion The aim of the paper has been to bring together a systematic context - Cognitive Engineering - and the possibility of rich data gathering methods, such as ethnography offers, by suggesting a conceptual framework which allows the designer to articulate the cognitive interactions of agents in complex, and essentially emergent, work environments. The crux of the contribution is effectively a re-interpretation of the idea of the task in CSCW analysis. The conventional idea of a task is adequate for describing and analysing a given piece of work, if the agents and their different perspectives are marginalised, and for providing what is often an idealised view of what the work consists in. However, attention has been drawn to its inadequacy when one considers different agents' view of what constitutes that work [26]. When these different perspectives on work are highlighted, as they are in the sharing of tasks implied by the idea of teamwork (see section 3), we need therefore to consider how such divergences should be characterised. The CM is introduced to conceptualise this aspect of the work and render it amenable to task analysis. As described before, Vicente distinguishes between two kinds of task analysis: two conceptions of how work can be analysed into tasks. His distinction might be seen as further evidence of the caution with which task analysis should be approached. However, he does not take further account of the part the agents' or co-workers' perspectives might play in the undermining of even this qualified view of task analysis; and
his examples [13] only reveal an allocation of tasks - what might be called collaborative taskwork - contrasting with the subtler interactions of teamwork. A significant role, however, is in fact played by individuals in connection with this difficulty of specifying what tasks (and their order) make up a given piece of work (see Figure 4). It is one of the contentions of this study that it is just such discrepancies between individuals' viewpoints which have to be resolved as a part of the removal of ineffective coordination. Two things have to be emphasised therefore: one, any application of task analysis to cooperative work is only descriptive of the work which is carried out (acknowledging the cautionary injunctions of such as Vicente); and, two, account has to be taken of the views which co-workers have of the work and, by implication, of others' performance of that
T. Lambie and J. Long / Engineering CSCW
121
work or of tasks which make up that work. The fact that it involves a description of the work, however, does not leave us without a way of prescribing improvements to a work system, since the task analysis which is central takes account of the needs of the coworkers as expressed by those 'attitudes' which are disclosed. Any prescriptive outcome, therefore, is driven by the design goals which consist of the requirements of the domain of work plus some alignment or harmonising of the 'attitudes' with respect to the 'content' of the tasks to be carried out. In other words, the design recommendations do not rest on some supposed objective needs of the work domain but on the requirements of that work plus the needs of the co-workers The employment of the notion of task back to the cognitive level makes it adaptable to conventional methods of analysis in HCI and CSCW but addresses the social aspect. However, it has no implications of a metaphysical nature. This is not an example of reductionism. Because the CM is introduced into a Cognitive Engineering context it is not a vehicle for expressing how things are. It is not explanatory. It functions only as a methodological reduction, and the reduction is undertaken as far only as it is deemed necessary to deal with the design problem, defined by some user requirements. It is thus not antagonistic to an approach such as that of ethnography. Cognitive Engineering's only stipulation is that there is a stopping rule imposed by practical limitations such as resources, and the need to answer the design problem. References [I] J. Adams, Risk and Freedom: the record of road safety regulation. Transport Publishing Projects. 1985. [2] Y. I. Noy, "Human Factors in Modern Traffic Systems", Ergonomics, 40(10), (1997) 1016-1024. [3] T. Lambie, A. Strork, & J. Long, The Coordination Mechanism and Cooperative Work. In T. Green, L. Bannon, C. Warren & J. Buckley (Eds. ), Proc. of the 9th. European Conference on Cognitive Ergonomics, 1998, pp!63–166. [4] T. Lambie & J. Long, Report for DERA on the Development of a Framework/Process for Eliciting Teamworking User Requirements for Defence Procurement IPTs (Integrated Project Teams), for the Director, Centre for Human Sciences Sector, DERA Farnborough [5] I. Sommerville, & T. Rodden, Requirements Engineering for Cooperative Systems, Collaborative Computing, I (1994) 219–235. [6] J. McCarthy & A. Monk, Channels, conversation, cooperation and relevance: all you wanted to know about communication but were afraid to ask. Collaborative Computing, 1, 35-60. [7] K Schmidt, Cooperative Work: a Conceptual Framework. In J. Rasmussen, B. Brehmer & J. Leplat (eds), Distributed Decision Making. England: J Wiley & Sons, 1991, pp75-l 10. [8] P. Falzon, Cooperative Dialogues. In: J. Rasmussen, B. Brehmer & J. Leplat (eds), Distributed Decision Making, J Wiley & Sons, 1991. [9] S. B0dker, Understanding Representation in Design. Human-Computer Interaction, 13 (1998) 107-205. [10] B. Brehmer, Some Notes on the Literature. In: J. Rasmussen, B. Brehmer & J. Leplat (Eds. ) Distributed Decision Making. England: J. Wiley & Sons. 1991, pp. 3–14. [ I I ] G. Storrs, A Conceptualisation of Multiparty Interaction. Interacting with Computers, 6(2) (1994) 173189. [12] J. Rasmussen, Modelling Distributed Decision Making. In J. Rasmussen, B. Brehmer & J. Leplat (eds), Distributed Decision Making. England: J Wiley & Sons, 1991, ppl 11–142 [13] K Vicente, Cognitive Work Analysis: Toward Safe, Productive, Healthy Computer-based Work. Lawrence Erlbaum Associates, 1999. [14] J. Dowell, & J. Long, Towards a Conception for an Engineering Discipline of Human Factors. Ergonomics, 32(11), (1989) 1513-1535. [15] J. Dowell, & J. Long, Conception of the Cognitive Engineering Design Problem. (Peer Review Paper). Ergonomics, 41(2) (1998) 126-139. [16] H. A. Simon, The Sciences of the Artificial. MIT Press, 1969.
122
T. Lambie and J. Long / Engineering CSCW
[17] U. Neisser, From Direct Perception to Conceptual Structure. In U. Netsser. Concepts and Conceptual Developments: Ecological and Intellectual Factors in Categorisation. England: Cambridge University Press, 1987. [18] S. W. Draper, Review of "Designing Interaction: Psychology at the Human-Computer Interface" in EACE News, Views and Reports, Int. Journal of Man-Machine Studies, 37, (1992) 812-821. [19] K. Popper, Conjectures and Refutations: the Growth of Scientific Knowledge. Routledge, London & New York, 1963. [20] M. J. Shaw, & M. S. Fox, Distributed artificial intelligence for group decision support: integration of problem solving, coordination & learning, Decision Support Systems, 9 (1993) 349-367. [21] E. Hutchins, The technology of team navigation. In Galegher, Kraut & Egido (Eds), Intellectual Teamwork. Hillsdale, N. J., Lawrence Erlbaum, 1990, pp. 191-220. [22] J. L. Austin, How to Do Thing with Words. Oxford University Press. 1962. [23] J. Searle, Speech Acts. Cambridge University Press, 1969. [24] H. H. Clark, Using Language. England: Cambridge University Press, 19%. [25] H. P. Grice, Meaning. Philosophical Review, 66 (1959) 377-388. [26] S. W. Draper, The notion of task in HCI. In S. Ashlund, K. Mullet, A. Henderson, E. E. Hollnagel, & T. White (Eds. ), lnterchi'93 Adjunct proceedings. (ACM) (1993) 207-208.
Cooperative Systems Design M. Blay-Fornarino et al. (Eds. ) IOS Press, 2002
123
Let's work together: supporting two-party collaborations with new forms of shared interactive representations Mike SCAIFE', John HALLORAN & Yvonne ROGERS Interact Lab, University of Sussex, Palmer, Brighton BN1 9QH, UK fjohnhall, yvonnerj @ cogs. susx. ac. uk Abstract Collaboration generally refers to people working together as a team towards a common goal. Here we discuss a different kind of collaboration: salesbased transactions which occur between two parties who have different goals, which are nevertheless mutually interdependent. Findings from an ethnographic field study of the way travel agents and customers build up complex products (e. g. round-theworld trips) showed the collaboration to be asymmetrical, affecting the success of its outcome. We discuss how we developed a collaborative technology, aimed at reducing the cognitive load involved during planning. We designed an interactive trip planner, that dynamically links a number of information resources and visualisations, co-displayed on three adjacent screens embedded in a shared worktable. Preliminary findings showed that using these kinds of graphical constraining enabled the two panics to quickly build up a shared reference and mutual understanding, which in turn facilitated more effective collaboration. Keywords. CSCW, two-party collaboration, sales-based transactions, information visualisations, external cognition, cognitive offloading, dynalinking, shared representations, planning, multiple representations, travel agents
1 Introduction The term collaboration is generally used in CSCW to refer to people working together as part of a team, for a common purpose [3], [5], [6]. Teams may work physically together in the same office or virtually, by being geographically co-located. They may be 'close-knit', where much of the collaboration depends on constant monitoring of what each other is doing when and where [5], [11]; or more distributed, where members come together for meetings and work separately at other times. Whatever the make-up of a team and how they might work, an overriding characteristic is having a common goal. Another form of collaboration, often overlooked in CSCW, is where people work together not as teams but as 'parties', who belong to different groups, but who need to collaborate to achieve different goals that are mutually interdependent. An example of this form of two-party based collaboration is between an agent and a customer2 in a sales transaction, where the agent wants to sell something to the customer and give them a satisfactory service, and the customer wants to get the best product that suits their needs. Each needs to provide information for the other, and to cooperate to enable the transaction to progress. While both parties may try to help each other out to achieve their respective goals, it is often the case that 'obstacles', such as misunderstandings and mistrust, can get 1 Mike Scaife died suddenly and unexpectedly while we were writing this paper. He was the instigator of, and inspiration behind, the research. 2 We use the term customer in the singular but it can refer to more than one person, such as a couple or family.
124
M. Scaife el al. /Let's Work Together: Supporting Two-Party Collaborations
in the way, making the collaboration difficult to manage. Furthermore, the collaboration can often end up being one-sided, where one party takes over the work needed to achieve both party's goals (usually the agent), leaving the other party highly dependent on him/her. Such asymmetry in the sharing of the work can have undesirable effects on the success of the collaboration. There are other differences compared with team-based collaboration. For example, the collaborators are usually complete strangers beforehand. This means that they have to build up a shared understanding of what each other wants and is trying to do for the other in a relatively short period of time. In so doing, they need to establish mutual trust, whereby each needs to find out the other's intentions and to believe what the other is saying is true. Much stumbling in the dark can happen at the beginning of a transaction where sensitive topics (e. g. the customer's age, how much money they want to spend) are often skirted around. In this paper we begin by exploring the problems that can arise in two-party collaborations. On the basis of our analysis we consider how we might overcome some of these through designing collaborative technologies. We then describe how we went about designing a more enriched physical and socio-technical environment intended to support more equitable collaboration and work. The particular domain we are concerned with here is the travel industry, where an agent and customer have to work together to specify a round-the-world trip, typically starting out with a vague and hazy plan. Sales-based transactions involve the creation of qomplex products. Developing an insurance portfolio, configuring a digital TV package, or booking a round-the-world trip, typically take a long time and much work to develop [14]. This is especially so where neither the customer nor the agent have a clear idea of what the customer really wants, and hence what the outcome of the transaction might be. To determine the nature of the product, much discussion, negotiation and 'fleshing out' needs to be carried out early on, and various alternatives have to be weighed up, together with the trade-offs involved in including certain options and not others. Achieving this state requires much accessing and interpretation of a diversity of information sources, e. g., online booking systems, brochures, websites and promotional materials, as well as the knowledge of the agent and the expectations of the customer. In addition, various kinds of representations need to be created, e. g., booking forms and schematic plans. Hence, much distributed and external cognition [12] is involved, where the socio-cognitive processes of planning, remembering, problem solving, conveying knowledge, constraint-matching and decision-making are central [7]. The disparate nature of having to deal with so many different kinds of information and representations at the same time, can cause high cognitive load, sometimes resulting in confusion and misunderstanding (see also [2]). A main goal of our research, therefore, is to design new forms of collaborative technologies that can support two-party collaboration. Specifically, our objective is to provide shared representations and interactive computational tools that can offload some of the cognitive work required to develop complex products during a transaction. They should enable both parties to be able to integrate more effectively the disparate kinds of information and representations needed during planning. The combined effect of providing such computational support should also allow both parties the opportunity to explore more alternatives given what is available, and what the consequences of making certain decisions are at a given point during their planning activities. In so doing, it could lead to a more equitable sharing of the 'work' involved when developing a product, better product specification and enhanced social aspects of the transaction (e. g. more enjoyment, better trust).
M. Scaife et al. /Let's Work Together: Supporting Two-Party Collaborations
125
2 The problem space: asymmetrical collaboration A general observation about two-party collaboration is the extent of the asymmetry that occurs between the two parties. To begin with, the customer makes all the running, starting by reading the literature, searching the web, visiting agencies, and writing out a plan. When they decide they have reached a point of 'readiness' (i.e. they need to turn their ideas into a product) they will approach an agent. It is at this stage that the customer is obliged to hand over responsibility for their planning to the agent, rather like handing over a baton in a race. Having taken over the customer's plan, the agent then takes control of the research and development work. In so doing, the customer's role is significantly reduced to being essentially that of a respondent, answering questions which the agent poses during their interactions. One of the main reasons why the customer becomes essentially excluded at this stage is that the information necessary for progressing the plan into a product is typically not designed to be used or accessed by them. Whereas the literature about a product, in the form of brochures, websites, pamphlets, posters, etc., is designed specifically for the customer (see Figure la), much of the information needed to actually build a product is represented in arcane formats, which are a legacy of the way database and software systems have been designed. While there are also good reasons why some information should only be available to agents (e.g. the amount of commission they are earning), other information remains inaccessible to customers not because it needs to be, but as a result of esoteric information representations. These frequently appear in the form of tables and codes which can only be understood and used by agents, who need to be highly trained to do this (see Figure Ib). Thus, the way the information is presented and accessed in sales-based transactions means it can only be interacted with by agents, with the customer depending on an agent to translate the information into a verbal form they can understand.
Figure la Information designed for the customer
Figure Ib Information designed for the agent
Figure 1 Differences in information resources used by customer and agent
The asymmetrical relationship between the agent and the customer is exacerbated by the particular arrangement of the technology and information displays used. The technology is usually set up in such a way that it makes it difficult for the customer to become engaged in the collaboration, even if both parties are willing {see Figure 2).
126
M. Scaife et at. /Let's Work Together: Supporting Two-Party Collaborations
Figure 2 Arrangement of information and technology at a travel agency
The agent and the customer will often sit on opposite sides of a desk, in the canonical 'office' set-up. The PC is positioned in front of the agent, who uses it to do a variety of tasks, like looking up flight availability or special offers, and filling in forms. Sometimes, the customer will surreptitiously try to peer at the screen (similar to reading someone else's newspaper on the train) and very occasionally the agent will swivel their monitor towards the customer to show them something. This often happens when the agent needs to convince the customer that a choice is not available, e. g. that there are no seats left for the days they want, and where the customer does not want to believe it because these are the only days they can go on. During these kinds of interactions, the customer will provide answers to questions like destinations, dates and budget, while waiting for the agent to tap this into a database and try to come up with suggestions. This means there is much time spent when the customer is waiting doing nothing, and is not being communicated with by the agent. The agent may also need to break away from the interaction with the customer, and leave their desk to access other resources like handbooks or to query other agents [8]. Clearly, these kinds of asymmetries affect the nature of the collaboration, and can result in the customer feeling disempowered and the agent overstretched. Hence, we wanted to find ways of improving two-party collaborations such that they might proceed more efficiently, more equitably and more enjoyably. Before deciding on what collaborative technologies to design, however, we needed to get a better idea of what is actually involved in two-party transactions and how existing resources are used. Thus, we began by carrying out a six month ethnographic study at a London-based travel agency specializing in roundthe-world and intra-continental tours. In particular, we wanted to find out how the arrangement of technologies, and the displays of information available, impact the way the transaction is introduced, followed through and completed. 3 Analysing the intricacies of the planning involved in two-party transactions using the external cognition framework We observed and video-recorded a number of different transactions that took place at the travel company, following through all the different stages involved in building a round-theworld trip. We also interviewed customers and agents about their strategies and the problems they encountered. We spent considerable time observing how various information resources, such as brochures, are used by the customers and agents, alone and when together. To analyse the findings from the study, we used our external cognition framework [9], [12]. Briefly, the framework provides a means of analyzing how different representations are used during collaborative activities such as planning, deciding and problem-solving. The framework also allows us to examine the level of cognitive offloading that takes place.
M. Scaife et al. /Let's Work Together: Supporting Two-Party Collaborations
127
This refers to how different external representations offload the cognition, such that they reduce the mental effort needed to perform a given task. We were interested in initially discovering what kinds of activity required a heavy cognitive load and how the different parties coped with them. In particular, we wanted to see how they re-represented information in order to constrain the problem space to progress with their tasks. We also wanted to see what bottlenecks arose during the transaction process and what role the available external representations played in this. We were also interested in the social and affective aspects that cognitive offloading might produce. For example, shared displays that both customer and agent can understand and operate might mean solutions are arrived at more effectively, and with a different character: more cooperative, and more satisfying for the customer. 3. 1 Key components of a typical transaction that takes place between a customer and agent When a customer first approaches an agent about planning a round-the-world trip, the agent will usually give the customer a set of brochures and ask them to return once they have worked out an itinerary showing where they want to go and what they would like to do. The agent needs to get the customer to do this initial 'pre-planning' work, because the degrees of freedom in what it is possible to do are immense. Hence, the customer needs to constrain this before the agent can start to progress with building the product. One of the agent's worst nightmares is when a customer comes in and says 'I'd like to go somewhere hot and sunny for a month'. 3. 1. 1 Using external representations Brochures are the main resource used by agents to help customers start making their first choices. They are designed to be attractive and pleasant to browse, and importantly provide suggested itineraries. However, when customers are initially given a set of brochures to look at, this can create high cognitive workload for them. This is because the different brochures are not usually cross-referenced, requiring the customer to have to do a lot of switching between them. For example, one brochure might feature only tour and travel information, and another hotels. If the customer wants to develop an itinerary involving a tour and then a week by themselves staying in hotels, they have to move between the two brochures and coordinate the information together, remembering where they last were in each. When deciding on a particular hotel, customers have to think, 'what does this cost'; 'how does it add to existing costs'; 'how long should I stay there'; 'what are the effects on my timescale'; and so on. Carrying all this in the head is difficult, so customers tend to compensate by creating their own plans, using pen and paper. These are usually timelinebased, with annotations, involving the times, destinations and cost of their planned trip. Creating these representations while going through the various brochures is also timeconsuming when trying to find something specific. Moreover, brochure material is difficult to reference without adding post-it notes or other indicators, otherwise the place one previously was will be forgotten. On top of this, once annotations have been made, items are difficult to move about without erasing and starting again. Also, it is difficult to represent different possible alternatives at the same time. Creating itineraries in this manner, therefore, requires a lot of cognitive effort. Added to the cognitive work is manual work: for example, annotations are often re-represented by writing up or wordprocessing for the agent. Having done all this work, a typical plan a customer might bring to an agent for their follow-up visit is as follows: the customer wants to spend six weeks travelling around Australia, arriving and staying in Sydney for one week; then travelling up to Cairns by air. They then wish to do a one week tour of the Great Barrier Reef taking in some of the islands, afterward returning to Cairns and travelling by train to Alice Springs, where they
128
M. Scaife et al. /Let's Work Together: Supporting Two-Party Collaborations
need three nights' accommodation. From there they will join another tour around the Red Centre, seeing Ayers Rock, and eventually returning to Alice Springs. They then wish to fly to Perth, staying for a week. Finally they wish to take a deluxe train journey from Perth back to Sydney via Melbourne. On the return flight out of Sydney they want to stay in Singapore. 5. 7. 2 Re-representing the itinerary When the customer has developed their initial itinerary, they present it to an agent. For the customer itinerary to be useful to the agent, it has to be formatted according to a particular structure, with dates, times, sequenced destinations, and budget. This requirement is always implicit, so customers may, despite writing down a detailed itinerary, produce something that is not easy for the agent to work with. At the same time, the customer's effort in organising their itinerary chronologically is only part of what is needed to quote on it.
Figure 3 Differences in customer- and agent-produced itineraries
Figure 3a shows a chronologically-ordered customer itinerary - an Excel spreadsheet produced after searching through brochures, and involving a lot of work. In contrast, Figure 3b shows an agent's itinerary made in response to a phone enquiry. This is organised quite differently. In order for the agent to produce a quote for the customer, the agent has to rerepresent the customer's chronologically-organised itinerary according to the different 'products' involved. This involves pulling out and ordering the different product types (a product can be a flight, a hotel, car hire, a tour, etc. ). The horizontal lists shown in Figure 3b, e. g. SYD-CHC/AKL-SFO, are abbreviations for sets of flights which the agent has to sort out before anything else can be done. So, using our example, the agent has to work out what flights are available into and out of Sydney, whether these fit with the customer's requirements, and how the internal flights (Sydney to Cairns; Ayers Rock to Perth) fit in. Following this, the agent moves onto accommodation and other 'land sales' products like tours. At this point the agent needs to see what hotels fit in with the itinerary, whether these are those specifically requested by the customer, and whether they are available. The same happens for tours (for example, the guided tour around the Red Centre; the deluxe train from Perth to Sydney). Thus product order follows booking order, where flights frequently have to be booked early to get savings, while the other products can come later.
M. Scaife et al. /Let's Work Together: Supporting Two-Party Collaborations
129
A lot of the booking work is done by the agent when the customer is not there, as it is time-consuming and requires considerable work to completely specify. Once a quote form has been completed by the agent, it is then sent out to the customer as a letter or email. The customer examines this and if satisfied confirms by sending in the required deposit. However, what often happens is the customer wants to make changes or query certain components of the product, or add detail (for example, requesting a room with a view at a hotel) before the deposit is given. The customer translates the quote form into another communication, sending this back to the agency via email, fax or phone. Thus, changes to the itinerary can come from two sources for two different reasons: from the customer for reasons of preference; and from the agent because of booking issues, non-availability of deals, replacement deals, and so on. Hence, the retranslation of chronological itineraries sketched out by the customer into product-ordered ones needed by the agent to make the booking, reflects different customer and agent models of how a product can be specified, as well as different expectations and requirements. Itineraries submitted by customers are often highly specified, with every component having equal priority in the sense that the customer wants simultaneous confirmation of all of them. However, working with booking orders, and issues like not knowing the exact price of a tour until closer to the time of the proposed trip, the agent cannot guarantee this. It may be that the whole product cannot be finalized for several months, and may even have to be changed. Thus, mismatches can arise between customer and agent expectations and priorities. 3. 1. 3 Media translation overheads and the linear planning effect To progress a provisional itinerary written by the customer into a firm booking developed by the agent, usually requires several interactions between the customer and the agent. These interactions can vary in the way they are conducted: face-to-face, via phone, e-mail, letter, or fax. The different modes of communication mean there are different media involved, and this creates media translation overheads, e. g., phonecalls translated into itineraries translated into booking forms translated into emails. In many cases the new medium requires a retranslation of the itinerary, sometimes from scratch. While this has the benefit of both parties being able to overview and recapitulate the itinerary every time it is looked at, the lack of a shared representation which clearly shows its current status creates extra workload for both parties as they do this updating. There are also update issues in that a new quote form has to be done from scratch; and any changed itinerary does not show where or how it has changed. At the same time, itineraries have to change because of issues to do with the booking; again, this involves a lot of redevelopment work where the impact of a change on the whole product has to be recomputed. Media translation overheads, agent workload, and the lack of shared representations, all create what we call a linear planning effect. By this we mean two things: first, before the product is actioned, only one possibility is decided (as we saw in our example), in order to constrain the work that needs to be done, given all the possibilities for a round-the-world trip. Second, the agent limits the space of parameters worked with to major ones like costs, dates, and destinations, which enables him/her to manage the cognitive overload of dealing with multiple concerns in the development of the product. If the agent is able to confirm the itinerary as requested by the customer, linear planning is effective. However, as we saw, the customer may want confirmation of a specific, detailed itinerary. Because of the booking order, this may not be possible; and if not, the agent has to find alternative components like different hotels or flights. These changes can be difficult to integrate into itineraries because of knock-on effects such that other parts also have to be reorganized. To facilitate this, the agent has to concentrate on the major parameters mentioned rather than considering other parameters such as their knowledge of the customer's interests, or what type of person they are. Here, concentration on a single itinerary, plus limitation of the parameters worked with, can create problems: the customer's rejection of changes, and the
130
M. Scaife el ai. / Let's Work Together : Supporting Two-Party Collaborations
need to do more development work. The linear planning effect, then, can create a paradox: the agent's requirement for a single itinerary specified in terms of a limited number of parameters, while it appears to save time and work, may in fact do the opposite. These kinds of product development problems can be exacerbated by the 'ping-pong' nature of the transaction. Communications going to-and-fro, often at a distance, mean it can take a long time for queries to be resolved. Products can sometimes take months to put together in this way. Here, we see how the sorts of information representation used by agents, and the onesided, linear form of transaction they give rise to, can create problems in sales-based transactions. Ways in which such bottlenecks might be reduced include developing different itineraries in parallel; the two parties getting to know each other; and through cooperative rather than serial ('ping-pong') practices. For example, the agent may be able to show a customer difficulties with the itinerary such that the customer chooses to leave certain things open, reducing the linear planning effect. However, agents have to work out the cost of how much time they are spending with a customer and the effort required to work up their product, ensuring that the total amount of time spent is profitable. Spending, say, ten hours with a customer to book a trip costing only five hundred pounds is a very inefficient and loss-earning use of their time. Thus the amount of time agents (and also customers) spend on developing a product is a critical factor that we needed to take into account: one that explains the linear planning effect despite potential difficulties. 3.2 Modelling the transaction process A key finding from our study, therefore, is that mismatches can arise between the information the agent uses to create a product, and the information the customer has and understands. In an attempt to bridge this gap, we decided to explore ways of providing new forms of information visualisations and interactive planning that would provide shared resources that both agent and customer could refer to during the planning and creation of an itinerary. Before building our prototype, however, we needed to consider which stage of the transaction process to focus on; one that would be most beneficial to the agent and the customer. Below is a simplified model of how a transaction, of the kind we have just described, develops, showing the roles of external representations used and created during the process (see Figure 4, below). The model depicts the transaction as a set of phases (as shown by the boxes). At the approach phase, the customer approaches the agency. If this is successful, an engagement phase follows, where the agent is able to act on the approach. This is followed by development and commitment phases, which may go through several iterations as the product is revised and finalized. At closure, the complete sale occurs. For one phase to flow
Figure 4 Transactional model
into another, the model proposes that a transition has to occur involving an interaction between (1) external representations used (e.g. brochures); (2) customer knowledge - the
M. Scaife et al. /Let's Work Together: Supporting Two-Party Collaborations
131
current plan of the trip, the budget, etc.; and (3) agent knowledge - of products, the current plan of the itinerary, how the product can be further developed. For transitions (and hence the whole transaction) to go smoothly, the cognitive work at each phase needs to be supported by appropriate external representations. The model identifies a critical transition between the approach and engagement phases. Here, the customer must have a workable itinerary already developed, in a form the agent can act on. Engagement can only occur once this transition has been achieved. We decided to look to support the transaction at the beginning which impacts the transition from approach to engagement. Here, as identified in our study, is where there is asymmetry in the distributed work done and where the external representations place high cognitive load on agents and/or customers. 4 Designing 'The Trip': a prototype to support better collaboration We decided to focus on how to better support the joint planning that goes on between customer and agent - especially in terms of shared representations used and created when making an itinerary in the early phase of planning. Our analysis generated a set of requirements, including: • reduce stages where there is an exceptionally heavy cognitive load on the customer and agent • provide shared external representations to allow shared planning and decision making • allow ways of exploring and discussing different product possibilities and alternatives, especially where there are awkward issues to be raised (e. g. how much a person is prepared to spend) • provide better links between different external representations to allow for more effective coordination of information Based on these, we decided to design a prototype that would provide: • a new physical set-up of technologies, changing the way the customer and agent sit together and hence collaborate • a way of dynamically linking existing external representations (e. g. brochures, notes) • new information visualisations and an interactive planning tool enabling easy comparison of alternatives and weighing up multiple parameters during the creation of an itinerary The prototype, called 'The Trip', consists of three interlinked components: • theeTable • the interactive planner • the shared visualisations 4. 1 Thee Table (Mark II) We propose that one way of improving collaboration, especially the ability for shared referencing, is to change the ways of displaying and sharing information and its accessibility, particularly for the customer. To do this we have been developing and combining different arrangements of displays and devices. Here we describe one such 'device and display' configuration, known as eTable (Mark II), that essentially provides three integrated large flat 21-inch displays set at 1280 x 1024 resolution, embedded in an oval table 1. 5m long and 1m wide (see Figure 5b). The customer (this can be one or two) and agent sit (or stand) in front of the eTable, where they can see each other and make eye contact, without the table acting as a barrier between them. Sufficient surface space is also provided as part of the table for a wireless mouse and keyboard together with room for placing other materials.
132
M. Scaife et al. /Let '.v Work Together: Supporting Two-Party Collaborations
Figure 5a eTable Mark I
Figure 5b eTable Mark II in use at a travel trade show
Figure 5 eTable Marks 1 and II
eTable Mark II evolved from an earlier version, Mark I [13] shown in Figure 5a. This was originally designed as a horizontal display surface, using back-projection from an xga data projector with a mirror arrangement under the table. The size of the table, 1m x 1. 5m and the embedded screen was designed to give a space that allowed two or three people to view and work with the display in relative comfort. People could sit or lean on/stand at it, and the asymmetric display location allowed an area for placing and working on things such as paper/brochures. Interaction with the display surface was pen-based, using an inexpensive mimio unit. Studies using eTable Mark I showed it improved collaboration between groups of three markedly, compared with the same groups trying to collaborate when huddling around a single workstation [10]. However, it only allowed for one screen of information to be displayed at any one time. To meet our core requirement for "The Trip', of allowing the customer and agent to integrate different representations readily and understand the connections, we decided that it was imperative to have multiple interconnected displays. The use of multiple monitors has also been found to be an effective strategy by users for coordinating information [4]. Hence, we designed eTable Mark II to have two embedded horizontal flat screen displays that are adjacent to each other, together with a third vertical screen behind them. This gave us considerable flexibility to experiment with partitioning of information, visualisations and plans that both parties had shared access to. 4. 2 The interactive planner component One of the ways we propose to reduce cognitive overload for both the agent and customer is to provide shared interactive representations that supported both chronological and product-ordered models, allowing for more effective searching and integration of information found from brochures. The interactive planner was designed to present information on the different screens of the eTable, that are contextually 'dynalinked' [9] with each other (see figure 6, over). For example, an itinerary selector is presented on one screen and the effects of making certain choices in this are shown on another screen. The agent and customer can refer to both and make changes to one which will cause further changes in the other. The interactive route component of The Trip' provides a palette of 'product' icons (planes, trains, hotels) which can be dragged onto an interactive map (such as Australia) which is then iteratively built up. For example, a hotel can be placed on Ayers Rock, which is where the customer is at that stage of planning their itinerary. A pop-up box appears,
M. Scaife et al. /Let's Work Together: Supporting Two-Party Collaborations
133
allowing them to interactively put in the number of days they would like to stay there. This triggers a relevant page from the brochure to pop up on the back screen showing them various options, which they can select. The hotel icon remains fixed at Ayers Rock as a reminder that this part of the planning has been done (it can equally be undone by removing it from the map). Dragging a plane icon next from the palette onto Ayers Rock and then moving it (for example) to Perth, produces an externalization of an internal flight. Hence, the itinerary can be built in terms of product order (e. g. flight, hotel, tour) or chronologically, by adding a hotel, then a flight, then another hotel, and so on. Any of the components of the itinerary can be removed and new ones tried out. Other flights can be added in the same way, and then hotels can be added. The interactive planner is linked to a brochure database and itself represents an itinerary. The itinerary can be downloaded as text to form a quote. The configured planner can also then be saved and printed as another form of externalization that the customer can take away with them. Thus the interactive planner was designed to provide new forms of externalization, making it easy for the customer and agent to try out alternative plans, by adding and removing components, without needing to start from scratch. In turn this supports experimenting with alternatives, something that is much harder to do using existing means. For example, a hotel configured at a certain cost and number of nights, if it is not available can be removed and another inserted, with the effects on the rest of the itinerary immediately shown in the shared visualisations (see next section). 4. 3 The shared visualisations Interlinked to the interactive planner are two visualisations, that show dynamically-updated graphical representations of the amount of budget spent so far and the segments of the itinerary being built up, using a time-line. They appear together on the adjacent display to the interactive planner. Both are updated in relation to the way the itinerary is being developed. For example, the cost of the flight from Sydney to Cairns plus four nights in a hotel selected from the online brochure are shown as a proportion of the budget, using a 'piechart' visualisation, which moves clockwise and anticlockwise depending on what the different selected products cost. At a glance the agent and customer can see how much is being spent, without ever having to explicitly refer to it. The time-line visualises the various segments of the itinerary using colour coding (e. g. a mauve block for a week on a tour) and weekly segments. Again it provides the customer with a way of seeing how parts fit together and also how they might change them, for example so that they spend two weeks in Cairns and one less in Sydney. It is also relatively easy for the agent to show the consequences of making such changes.
134
M. Scaife et al. /Let's Work Together: Supporting Two-Party Collaborations
Infofmcflon on hotels In Avers
Action in Screen brings up Information in Screen 3 and then Screen 2 changes accordingly
Interactive planning tod, | allowing exploration of | different options (e. g hotels in Avers Rock)
Two linked visualisations showing a timeline filling with the planed itinerary and cos adding up as a proportion of the budget
Figure 6 Screen shots of the three displays from The Trip' showing the itinerary unfolding
Post-it notes are also provided to allow for annotation to the itinerary as another form of external cognition. For example, if the customer is not certain whether they should go for a particular hotel, they can add a post-it to this component with a note to that effect. This can be dragged to one side while other possibilities are tried, and reinserted in the itinerary if required. 4. 4 The shared experience The combination of the eTable, the interactive planner and the shared visualisations provide a new experience for the agents and customers. The way it is envisioned to be used in a travel shop is that when a customer is first thinking about going on a round-the-world trip, they would be encouraged to sit down and explore possibilities with an agent. Within a 510 minute time frame, it would be possible for the two to have reached a shared understanding of what the other expects and wants, to allow the transaction to progress to the next phase. Working together like this, which they cannot do at the moment, means that there is less need for translation in the initial phases of building up a product: both parties can work with the same representations.
M. Scaife et al. /Let's Work Together: Supporting Two-Party Collaborations
135
5 Preliminary evaluation of 'The Trip' To evaluate our prototype, we carried out a preliminary experiment in which joint planning tasks were set up. We also placed The Trip' in a large travel trade show in London, where customers explored its potential with the aid of a sales agent from the travel company we are working with. 5. 1 Findings from the experiment The aim of the experiment was to determine whether the new forms of shared interactive visualisation allowed participants to more effectively plan an itinerary. There were two conditions. In the first set-up, two participants acted as customers and were asked to plan a trip with budget, time and destinations constraints. Specifically, the participants were asked to plan a holiday around Australia starting and finishing at Sydney, and taking in Cairns and Ayers Rock. It should also include a tour around the Great Barrier Reef. The 'customers' were given £2, 200, and asked to limit their trip to three weeks maximum. The set-up was designed to be sufficiently complex, to enable us to evaluate the extent to which the visualisations and the integrated representations facilitates customer planning, where many parameters have to be taken into account. In the other set-up, two participants interacted with each other: one acting as a customer who had already made their plan; and the other acting as an agent. This second participant had been 'trained' on what they needed in order to progress the transaction. One of the participant customers from condition one used their saved planner visualisation to 'approach' the agent. The agent had to be able to interpret it in terms of booking orders and to be satisfied that they could progress the plan into a product. Hence, the aim here was to see how well 'The Trip' supports joint planning between customer and agent. In both conditions, sets of brochures were provided as well as pen-and-paper. A control condition was also carried out where the participants carried out the same tasks, but using only paper-based brochures and pen and paper. Briefly, the main findings from the preliminary experiment were: • Planning was found to be much quicker for the conditions when the participants were using 'The Trip' compared with the condition where brochures only were used. Participants were able to put together an itinerary in about ten minutes using 'The Trip', compared to over thirty minutes in the brochure-only condition. This was because they did not need to search and coordinate the brochures. Also, there was no need to make and revise a pen-and-paper plan because the computation was being done for them by the planner. • Although they were not instructed, participants spontaneously tried out and saved different possibilities using 'The Trip'. The post-it notes and the ability to save configured components appear to have contributed to this. Another reason is that the time and cost scales, which update automatically, made it easy for participants to see immediately what the effect of changes were on the whole itinerary. • The agent participant was able to see clearly what the product order was, when provided with the saved itinerary. • The customer drew attention to two alternative itineraries involving how to do a tour, either by starting and finishing at the same point, or by finishing at a different point, and what the timing/cost implications were. The agent was able to demonstrate these alternatives easily and to save these. These findings suggest therefore that 'The Trip' was effective in reducing cognitive load at different stages of the planning of the itinerary. Information, as it unfolds during the planning, is externalized and coordinated such that computations and representations do not need to be made and kept by the customers in their heads or on scraps of paper. The findings also showed the benefit of having shared planning, in terms of speed, efficiency and ease of communication when using 'The Trip'. The participants were able to develop
136
M. Scaife et ai I Let's Work Together: Supporting Two -Party Collaborations
alternative plans relatively easily and the linkage of information between different representations also appears to have been effective. 5. 2 Preliminary findings from the situated use of 'The Trip' at the travel trade show At the trade show, The Trip' was set up as shown (above) in Figure 5b. Three types of video data were gathered: demos to agents by ourselves; demos to customers by ourselves; and use of the system by agents and customers together to support a transaction. Feedback from the demos on the customer side was positive, with many customers interested in, and spontaneously trying out, the prototype. Agents were also enthusiastic, but under time/commission pressure needed to be able to integrate the results more seamlessly with their standard practice of using DOS-based applications on other PCs. This meant that the consultation, after the engagement phase, was continued in a separate area using these other systems. However, despite this issue, successful consultations were held using "The Trip' to support approach and engagement phases, leading to potential bookings (due to the technical restrictions of the trade show setting, actual bookings were not possible). These consultations differed markedly in character to those taking place simultaneously with the agents' standard systems, with shared use of the display, a more informal interaction without any table/PC 'barrier', much more direct customer input, and exploration of different options. In terms of time, customers and agents working together were able to put together reasonably full itineraries in around 20 minutes - faster than the typical consultation, with customers more engaged and active. 6 Discussion The focus of our paper has been on supporting collaboration between two parties that have different goals, but which are mutually dependent. In particular, we looked at how multiple representations are produced, translated and coordinated between different parties when carrying out a complex sales transaction. We found that existing ways of building up a shared product can be hindered by the problems associated with using disparate resources. Our ethnographic study showed, for example, how the information represented in brochures and databases is quite different, sometimes leading to misunderstandings and divergent planning strategies between the agent and customer. Equally, we saw how the discrepancies between the way such representations are structured can constrain the collaboration between the agent and customer to one of linear planning where there may be more optimal means: for example, exploring alternative plans before making a commitment. A key concern arising out of our research was how to provide a collaborative environment that could more effectively support the coordination of the multiple representations that are generated and accessed by the two parties. Currently, customers carry out their planning in a largely ad hoc fashion, not knowing fully what the parameters are that they must specify, the level of detail, or the order the itinerary should be built in. They can also be overwhelmed by the information overload when consulting numerous brochures, the web, etc. The agent, on the other hand, has the knowledge of how to use brochures and what form to write an itinerary in (in order that it can be progressed into a product), but cannot dictate or suggest to the customer how to proceed at the outset. They must instead spend considerable effort re-translating the idiosyncratic plans and ideas that the customer has generated, by themselves, into a workable form. Such a problem space is ripe for being better supported. However, there is little design guidance on what kinds of collaborative tools might be developed to more effectively support the use and generation of multiple representations, that would enable reduction of the current level of cognitive effort and work needed. While there has been some work on
M. Scaife et al. /Let's Work Together: Supporting Two-Party Collaborations
137
how to design effective multi-representational learning environments, the focus has been on supporting the learning process per se (e. g. linking multiple representations to provide deeper understanding of the way abstractions work). One design principle that has proven to be generalisable to other contexts and we used extensively here is dynalinking [9], which proposes co-varying context-relevant information in co-located displays to show the relationships between them more explicitly. Cognitive offloading was also reduced for both parties, and in so doing, supported better collaboration between agent and customer in terms of allowing them to work on the same product using the same set of tools. It also provides them with a shared reference that both can refer to and have a mutual understanding of, allowing them both to talk on more equal terms. More specifically, it was shown to reduce the media translation overheads, divergent planning strategies, and repair work which come from having disparate information representations; and representations which are not accessible by both parties. The Trip' goes some way to showing how better coordination across external representations can be achieved, and how this impacts a two-party collaboration not only cognitively but also in terms of its social character to create a transaction which is more effective as well as more congenial.
Acknowledgements The authors gratefully acknowledge the support of a grant from the UK ESRC/EPSRC 'People At The Centre Of Communication & Information Technology Programme', Grant Number L328253027. Thanks to Bridge the World, especially Jerry Bridge, James Bell and Gillian Hughes, for all their help and encouragement. We also thank our eSpace partners Tom Rodden and Ian Taylor at Nottingham University for many invaluable discussions; and Eric Harris for coordinating the building of eTable Mark II. References [1]
S. E. Ainsworth (1999) Designing effective multi-representational learning environments. Technical Report No. 58, School of Psychology, University of Nottingham. [2] S. E. Ainsworth, P. Bibby and D. J. Wood (1997) Information technology and multiple representation: new opportunities new problems. Journal of Information Technology for Teacher Education, 6, 93–104. [3] C. A. Ellis, SJ. Gibbs, and G. L. Rein (1991) Groupware: some issues and experiences. Proceedings of CSCW'88, 250–256. [4] J. Grudin (2001) Partitioning digital worlds: focal and peripheral awareness in multiple monitor use. Proceedings of CHI' 2001, 458-465. [5] C. Heath and P. Luff (1992) Collaboration and control: crisis management and multimedia technology in London Underground line control rooms. Journal of Computer Supported Cooperative Work 1(1), 24-48. [6] E. Hutchins and T. Klausen (1996) Distributed cognition in an airline cockpit. In Y. Engestrom and D. Middleton, Eds. Cognition and Communication at Work. NY, Cambridge University Press, 15-24. [7] B. Raskutti and I. Zukerman (1997) Generating queries and replies during information-seeking interactions, International Journal of Human-Computer Studies, 47, 689-734. [8] Y. Rogers (1994) Exploring obstacles: Integrating CSCW in evolving organizations. In CSCW'94 Proceedings, ACM, NY. 67-78. [9] Y. Rogers and M. Scaife (1999) How can interactive multimedia facilitate learning? In: CDROMAVWW Proceedings on Multimedia Interfaces, AAAI. Menlo Park, CA. [10] Y. Rogers, M. Scaife, and S. Lindley (2002) The affordances of vertical and horizontal displays: their effects on group collaboration, (in prep). [11] Y. Rogers, H. Sharp, and J. Preece (2002) Interaction Design: Beyond Human-Computer Interaction. Wiley.
138
M. Scaife et al, /Let's Work Together: Supporting Two-Party Collaborations
[12] M. Scaife and Y. Rogers (1996) External cognition: how do graphical representations work? International Journal of Human-Computer Studies, 45, 185–213. [13] M. Scaife, Y. Rogers and J. Halloran (2001) eSpace: integrating novel displays and devices for augmenting collaborative transactions, www. cogs. ac. uk/users/vvonner [14] A. Zalatan, (1996) The determinants of time planning in vacation travel. Tourism Management, 17, 123-131.
Cooperative Systems Design M. Blay-Fornarino et al. (Eds. ) IOS Press, 2002
139
Studies of Computer Supported Collaborative Writing Implications for System Design Teresa CERRATTO' and Henrry RODRIGUEZ Interaction and Presentation Laboratory Department of Numerical Analysis and Computing Science Royal Institute of Technology Stockholm Sweden
Abstract This paper analyses transformations in collaborative activities that a computersupported collaborative writing system introduces into co-authors' practices, and, discusses implications for the design of collaborative tools for writing. The analysis is grounded in user studies of four different groups of co-authors writing an academic report during two different collaborative situations. Two groups collaborate face-to-face using a word processor and the other two groups collaborate at a distance using a synchronous collaborative writing system. The study analyses differences between the activities of the groups and focuses on transformations that the CSCW system introduces into the organization of the co-authors' collaboration. The results show that the use of a CSCW system presents constraints in the collaborative activity; in particular we observed a cleavage that makes co-authors write different parts of a common document independently, i. e. without collaborating. This result opens up important issues related to the design of writing technologies for collaboration: how collaborative can collaborative writing be when it is supported by CSCW systems? How can the design of collaborative tools support organized and integrative collaborative activities such as writing? We discuss these issues and some of the alternatives that collaborative tools should offer for writing. Keywords: CSCW, computer-mediated collaborative writing, field study, design
1 Specificity of collaborative writing activities A fair amount of research has been conducted into the ways groups write together ([1], [2], [8], [10], [18], [20], [37]). Much of this work agrees on the idea that collaborative writing involves phases of writing and communication [1], periods of synchronous activity where the group works together at the same time, and periods of working alone where group members work at different times. The diverse range of activities involved and the different modes of interaction make collaborative writing a particularly interesting domain for CSCW support. Collaborative writing is a very complex and specific collaborative activity that differs from others in that written language is both the group's product and its means for communication between the writers. As Sharpies [37] put it "this may introduce confusion as to the extended purpose of a communication from a co-author. Is the text of a written message intended to form part of a draft document? " (p. 20). The fact that the 1 The empirical study presented in this paper was part of my doctoral dissertation conducted at the University of Paris 8, St-Denis, France.
140
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
collaborative activities are mediated by text, adds an extra twist since comments are communicated in a form which resembles the text-finished document. This tends to add more weight to the written comments, perhaps contrary to the co-author's intention. The following transcript of the text-based dialogue between two co-authors exemplifies the problem of differentiating on text two levels that are intertwined in collaborative writing: "writing for myself and "writing for the others". Transcript 1 Paul asks: "Have you seen the "?????" in part al ? Mark answers: that is none of your business, it is my part, questions marks are just for me How can co-authors distinguish between "comments for me" and "comments for the group" ? How can co-authors distinguish between the final version and a draft version under development? This is just one of the characteristics that makes collaboration complex in writing activities. Each co-author advances in the common text while being based on the representation and the action of the other. The means of coordination thus differ according to whether they are centered on the text, the others, oneself or on the artifact in use. According to Schmidt and Simone [36] what differentiates collaborative activities from other types of interaction is that they are constituted by several actors who are "interdependent in their work and who therefore divide, allocate, coordinate, schedule, mesh, interrelate, in short: articulate their individual activities" (p. 56). People involved in collaborative activities are inexorably interdependent in terms of strategies, heuristics, perspectives, goals, motives, etc. The more distributed the activities of a given cooperative work arrangement, the more complex the articulation of the activities of that arrangement. With low degrees of complexity, the articulation between individual activities can be achieved by means of the modes of interaction of every day life. As Schmidt and Simone [36] put it "under such conditions the required articulation of individual activities in cooperative work is managed so effectively and efficiently by our repertoire of intuitive interactional modalities that the distributed nature of cooperative work is not apparent most of the time". However in situations with high degrees of complexity such as distant collaboration for writing a document, a common understanding about modes of coordination and integration become a must for the collaboration of a group of actors. The relation between coordination and integration has been studied in detail by Beguin [5]. According to him, there is interdependency between coordination and integration that is essential for the organization of collaborative activities. In a collaborative writing situation the inter-playing between coordination and integration organize co-authors collaboration, elaboration of meanings and mutual understanding: The following transcript exemplifies co-authors' modes of coordination and text integration. Transcript 2 Nils says: Don't do anything I'm gonna write the general table of contents right now for the whole document [coordination of individual actions] Paul asks: Who has written that the machine does support decision making processes ? I don't agree [content integration] Collaborative writing is thus doubly mediated by integration of texts and ideas and by coordination of actions involved in the document production. The study of computer mediated collaborative writing activities thus becomes the study of the impact of the
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
141
properties of CSCW tools on the interdependency between co-authors' coordination and integration; mechanisms that are at the core of collaborative writing practices. 2 Framework Although there has been a growth of interest in collaborative writing ([10], [15], [19], [26], [33], [44]) and in computer mediated collaborative writing ([2], [20], [22], [25], [40], [43]) questions about appropriation and development of use of a computer system for writing collaboratively have mainly been overlooked, even though Sharpies & Pemberton [39], Rimmershaw [29] and Haas [14] have argued that writing experiences are transformed by the use of writing tools. The theoretical perspective adopted here approaches the use of tools as a developmental process that can be viewed from two angles: in terms of artifact development and in terms of the development and realization of the needs of individuals and groups ([4], [5], [6], [7], [27]). Special emphasis is placed on the idea that an artifact cannot be confounded with an instrument. An artifact only becomes an instrument through the subject's activity. The distinction between artifact and instrument is crucial to understand the aspects involved in the integration of a new tool into human practices. A technical system does not immediately constitute an instrument for the user. Even explicitly constructed as an instrument, it is not, as such, an instrument for the subject. It becomes an instrument only when the subject has been able to appropriate it for him or, in other words when he has been able to subordinate it as a means to accomplish his ends. Thus, an instrument results from the establishment, by the subject, of an instrumental relation with the artifact. According to the approach, human activity is always mediated and highly organized ([12], [21], [27], [28], [41]). People interact with the external world with a minimum of organization in their intentions and actions. Each human activity presents a specific organization that is both invariant acquired- and highly situated –flexible-. This organization is associated to the different features and characteristics of artifacts that make them fully into means of action for the user. From this perspective collaborative writing is regarded as an organized activity that has been acquired and developed historically and relates to the available writing artifacts. Interacting with a CSCW tool when writing a common document will thus penetrate a particular interactive and subjective organization of the co-authors' activity. 3 Empirical Study The interest in the use of collaborative writing systems and type of transformations that a CSCW tool brings to collaborative writing practices, motivated the study of co-authors' collaborative writing practices. The groups of co-authors were attending a course at a Business School. They had to write a real report for the validation of a course attended. The study was organized together with the responsible for the course who encouraged students to participate in the study and offered two extra points to them taking part in the study. Four groups accepted to participate in the study. Two groups collaborated face-to-face using a word processor and two groups located at a distance collaborated using a CSCW tool. The groups were observed, video-taped and interviewed. The groups had an agreement with us; they proposed to write the report where the video-equipment and the CSCW tool were installed. However, analysis of different drafts reveals that the groups have also worked by other means that those we had agreed on. In that sense, the present study reports on a portion of the collaborative writing activity actually developed by the groups.
142
T. Cerratto and H. Rodriguez- / Computer Supported Collaborative Writing
The focus of the study was the collaborative writing activity of a group of co-authors collaborating in different situations for a real learning purpose. The goal of the study was to account for transformations introduced by the CSCW tool into the co-authors' activity. To study and analyze face-to-face collaborative writing activity when the interest relates to the introduction of a synchronous collaborative writing artifact is a conscious methodological choice. Within the framework of the instrumental approach, a new tool does not carry a new activity inside itself that will "bloom" and be developed when the subject starts to use it. On the contrary, the artifact will be assimilated into a practice, a cognitive a social organization, which predates the artifact. In this sense, the artifact will transform the user activity as well as the artifact itself will be transformed by the user actions. The study of face-to-face collaboration thus worked as a referential framework for the interpretation of the transformations of the collaborative activity observed with the groups that used a computer support for collaborative writing. 3.1 The artifact: a synchronous computer support for collaborative writing Aspects® is a commercial synchronous collaborative editor, a word processor running on a Macintosh platform that allows a group of users to work on a single shared document using separate computers. It provides a common writing space through which the group can communicate and interact. The document that appears on each user's screen is the same document that every other user can see and has access to. Figure 1 : The collaborative writing interface. There are two spaces, the writing space and the communication space (chat-box).
Each person can look at different parts of the text, but the overall document will always be the same. Remote users' selections show up as hollow rectangles, whereas the local user's selections appear in the usual manner, as a reversed rectangle. The tool provides a variety of communication tools, including a chat-box for communication, electronic pointers, i.e. cursor that appears on the other user's screens. Because of the technical problems of distributing a text document across a computer network between several users and still
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
143
maintain consistency between the different copies of the document, a number of control mechanisms have been built into the system. For example, all users are able to enter text at the same time but only one user may edit a given paragraph at a given time. A bar to the left in the text window indicates that someone has the paragraph locked (see figure 1). A black bar represents the local user whereas a gray bar represents a remote user. 3.2 The task: to write together an academic report The task consisted of writing an argumentative document reporting on a study about decision-making processes in French companies. The report comprised of an analysis, an evaluation and a comparison of computer supported decision-making processes (an expert system) in relation to human decision-making processes. The teacher asked the groups to reflect on the limits and benefits offered by a tool such as the expert system in decisionmaking processes and write about them. More specifically, the teacher asked the students to think critically about the limited rationality of artificial intelligence tools. The report was part of the examination of the course. The groups produced texts ranging from fourteen to twenty-one pages. 3.3 The groups The groups consisted of sixteen graduate students distributed into four groups of four people each. The students were familiar with working in groups and in particular with writing group documents; they were furthermore familiar with word processors, electronic mail, spreadsheets and the Internet. They were also interested in new technologies of communication and work. They were however not familiar with synchronous collaborative writing systems. They received training aiming at presenting the collaborative features of the CSCW tool and at exercising to write and communicate in the CSCW environment. The fact that they were students attending a course was of the particular interest for the study. Learning is a favorable context for studying developmental processes in which artifacts are used and transformed into instruments. In such situations, users can express lack of comprehension and make errors without trying to follow normative forms of using the tool. 3.4 Methods The analysis focused on the impact of the CSCW tool on co-authors' verbal and text-based exchanges while they were writing the common document. We conducted qualitative analysis of the groups' activity and the task as well as quantitative comparisons. Three dimensions of the collaborative activity were studied in detail: the co-authors' construction of the writing context from the CSCW shared file; the coordination of the writing rhythms in a synchronous computer environment and the impact of the synchronous text-mediated communication on the collaborative writing practices. In the study reported here, we only account for the latter dimension. Focusing on the transformations that the use of a CSCW tool brings to group communication, we compared the amount, the type and
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
form of the exchanges2 of the groups communicating face-to-face and through the CSCW tool.
Table 1 : Situations, artifacts used by the groups and working time Collaborative task : Writing a report Situation Face-to-face group Remote group PC laptop, word CSCW tool, notes, processor, floppy disks, pen and paper, a Artifacts in use notes, pen and paper floppy disk Time employed per group
Group F 5 h 20 min Group Y : 4 h 13 mim
Group M : 6 h 30 min Group X : 1 1 h 20 min
The total activity studied in the two conditions was in average 4 h 30 minutes for the faceto-face groups and 8h 30 minutes for the CSCW groups. We have systematically compared the first 30 minutes of composition3 and the first 30 minutes of revision of the shared document [6]. Two different raisons made us establish thirty minutes as the time unit for the comparison. First, the CSCW groups decreased the production of exchanges considerably during the sessions. Second, the first minutes of any activity are in general indicative of how the activity will organize itself afterwards [32]. 3.5 Findings Analysis of the CSCW groups' activities using a computer-collaborative writing system can report serious failures in both the writing product and process which indicates serious constraints in the activities of the CSCW groups. First, CSCW groups exchanged very few messages compared to the groups working faceto-face. Second, CSCW groups focused their activities largely on the organization of their actions while the face-to-face groups focus their collaboration on the content of their texts. Third, the collaborative writing activity of the CSCW groups was characterized by the production of minimal exchanges while the face-to-face groups produced complete exchanges. In order to analyze the meaning of these results, we develop on each one below. The following tables show some of the results the impact of the CSCW tool on communication during the composition and revision phases of collaborative writing. 2 The verbal or text-based exchange is the unit of analysis for communication. Exchanges are defined as equivalent to a turn taking unit. An exchange consists of at least two utterances; called initiative and response. An evaluative utterance can appear as a third component in a turn [34] [35]. According to Kerbrat-Orecchioni [17] only when the third utterance appears the exchange can be regarded as complete.
3
When writing with computers the composition and reviewing phases are intertwined and thus difficult to distinguish. The distinction described here was introduced for the purpose of the study.
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
145
Table 2: Comparing amount, type and form of the exchanges in different collaborative situations during composition phase Face-to-face situation CSCW situation Composition phase F and Y groups M and X groups Number 87 exchanges 17 exchanges Type Action 17 20% 13 76% Content 70 80% 0 0% Artifact 4 24% 0 0% Form 11 Minimal 22 25% 65% 75% 65 6 35% Complete Table 3 : Comparing amount, type and form of the exchanges in different collaborative situations during revision phase Revision phase Face-to-face situation CSCW situation M and X FandY Number 117 exchanges 12 exchanges Type 26% 50% Action 30 6 50% Content 82 70% 6 4% 0% Artifact 5 0 Form 37% 9 75% Minimal 43 74 63% 3 25% Complete
Groups using a CSCW tool exchange very few messages compared to the groups collaborating face-to-face Analyzing the amount of exchanges, face-to-face groups produced 204 exchanges while the CSCW groups produced only 29 exchanges in the same period of time during the production phase. The impact of the artifact, and more precisely of the synchronous textbased chat box on the production of exchanges, is apparently very large. The question is : how do groups using CSCW system collaborate when they talk so little? The CSCW groups produce only 14% of the total number of exchanges produced by the face-to-face groups. It is a fact that synchronous text-based communication constrains discussion. Co-authors found it hard to have to write what they could have said much faster. They also found it difficult to point to specific paragraphs and to comment on them through the shared screen. When they started to use the chat box for communication, the CSCW groups were rather glad to discover that they could communicate with each other while writing in the same document in spite of the distance. But, when the initial effect was gone and the more pressing need to understand and coordinate actions and content with each other arrived, the chat-box became more of a problem to manage than an instrument for communication. Co-authors tend to avoid reading messages in the chat box while they write. They in fact try to keep themselves concentrated on the text and avoid any distractions when they compose. To read the other's messages involves stopping the writing activity and shifting their attention to another space on the screen initiating a different activity -. This is a shift from composing text to the very different activity of communicating about the text-. Avoiding the chat box entailed a problem when messages
146
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
needed to be discussed by the group. Moreover, the fact that they avoid to read messages in the chat box generates an extra activity when authors have to take into account comments sent by others; in fact they have to read all the comments sent and try to relate them to the specific parts being under discussion in a particular section of the shared document. The CSCW groups had several problems trying to locate and understand others' questions or comments through the chat box. They also mentioned that the alert signal that announced the arrival of a message interrupted the ongoing activity of elaborating ideas into writing. They had to jump back and forth in the documenting. Groups also mentioned the fact of having to "wait for ages" before other co-authors answered their requests. The impact of the chat box for communication interests us from the point of view of the constrains that the feature introduces into the interdependence of the co-authors' actions and co-production of paragraphs. Most of the research that has investigated the effects of the use of text-based communication has focused on differences between oral and written communication [see 35, p. 140-145]. The focus here differs in that the interest lies in understanding how the artifact affects the interrelation between writing for communication about the text and writing the text. The interest is in the writing technology question as a means for coordination - written communication- and construction of text -writing-. In this sense, the weak production of exchanges of the CSCW groups questions which type of relation co-authors are able to establish and maintain when they are communicating through a chat-box for writing a shared document. Plowman ([24], [25]) pointed out this problem, showing the role that communication plays in the process of co-writing intellectually demanding texts. Plowman thinks that it is catastrophic that certain software for collaborative writing does not offer other spaces for communication than chat boxes. In her text, Plowman points to the interfunctionality of talk and text and the close relationship to human cognition. What is significant in collaborative writing, is that communicating about the text is not a separate activity that should be supported but rather an important instrument of the collaborative writing. Communication is highly interrelated to the writing process. Indeed, it is the relation between text and the communication much more than the communicational space itself that is the essential instrument to be supported by a collaborative writing artifact. The groups using a CSCW tool focus their activity on the organization of their actions while the face-to-face groups focus their collaboration on the content of their text The distribution of the types of exchanges produced by the groups working in different situations shows a significant disparity (see tables 2 and 3). First, face-to-face groups primarily concentrated their exchanges on the "content" of the document, which was interpreted as a sign of co-authors' interest in the ideas being written by themselves and others. Second, they referred to their actions something that was interpreted as necessity to question, modify or adjust their organization of the ongoing situation. Third, they did refer to the artifacts with which they were collaborating. The distribution of exchanges in the CSCW groups differed from the groups working in the face-to-face situation. First CSCW groups concentrated exchanges on their actions. It became essential for them to know about the ongoing organization of the collaboration (cf. who is doing what, when and where in the document). Others' actions became the principal object of the CSCW groups' communication. Second, co-authors referred to the content of the document. This result indicated that the co-elaboration of ideas did not constitute the object of the co-authors activity. Third, co-authors concentrated their exchanges on the artifact in use. If we concentrate on the results related to the categories "actions" and
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
147
"content" we can observe that face-to-face groups through their exchanges, focused and elaborated their common object, that is, the content of the document while the CSCW groups focused on the organization of co-authors' actions. The CSCW groups centered on "how should we collaborate" (actions) while the face-to-face groups centered on "what should we write" (contents). The imbalance between "how should we collaborate" and "what should we write" indicates a cleavage between coordination and integration processes. Collaborative writing through the CSCW tool curiously became more of individual writing. Beck & Bellotti [3] mentioned the needs for developing systems which aim at facilitating information of the situated activity of the co- authors. Dillon [8] pointed out that the design of CSCW tools to support writing must support information about the current state of the document as well as of the state or situation of the collaborators. The visibility of individual actions help to make sense of the others' actions, but is not sufficient. The problem lies in supporting and facilitating the intertwined mechanisms of integration and coordination. Facilitating the interdependency between integration and coordination remains fundamental for the organization and development of collaborative writing. Type of collaboration supported by the CSCW tool In the analysis of the exchanges produced by the groups collaborating in different conditions we were particularly interested in the co-authors' engagement and capacity to evaluate the ongoing dialogue. In the coding, the production of exchanges with two utterances is according to Kerbrat-Orecchioni [17] regarded as minimal exchange and is interpreted as indicating minimal engagement in the conversation. A more significant production, with three utterances, is regarded as a complete exchange and is interpreted as an indication of engagement in the activity [6]. The analysis regarding the forms of exchanges produced by the different groups of co-authors indicates that CSCW groups produce a large proportion of minimal exchanges, while the exchanges of the face-to-face groups are characterized by complete exchanges. These findings are maintained in both the composition and revision phases of collaborative writing. Co-authors collaborating in the face-to-face condition tend to enclose the exchanges with evaluative utterances that is interpreted as a sign of interest and engagement in the group communication. Evaluative utterances have a function that is to specify if the answers were understood or not and to show engagement in the dialogical activity [17]. These findings raise the question of the CSCW co-authors' engagement in their collaborative activity. How are groups actually creating, negotiating, writing and shaping a common text with so little concern? The small number of complete exchanges produced by the CSCW groups points to a contradiction. In learning environments, collaboration is viewed as a technique that stimulates and encourages evaluative utterances. Evaluative utterances play an important role in the development of critical thinking and meta-cognitive processes ([16], [37]) and introduce the difference between awareness and reflection [15]. Reflection when collaborating for creating a common understanding constitutes an activity involving the production of evaluations, justifications and alternatives [15]. It goes beyond answering questions; reflection especially consists of interpreting others' answers -i.e. ideas and texts- and providing adequate feedback on them. The minimal form of the exchange indicates that the use of the artifact might constrain engagement and critical thinking during the writing of a shared document.
148
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
3.6 Summary The most significant difference observed between the groups is the fact that CSCW groups manage their shared document individually while face-to-face groups together co-elaborate the content of the common document. One can thus wonder which is the common object of the collective activity when groups collaborate through an electronic network. The CSCW groups were focused on getting their report completed rather than learning from each other. Without supporting interdependence between co-authors' actions -coordination- and coauthors' paragraphs -integration-, the collaborative activity of the CSCW groups fragmented. Aspects of the collaborative writing activity that traditionally has been experienced as a coherent whole split up as follows: •
The content is dissociated from the communication about the content, i.e. commenting on the text does no directly relate to the text. The CSCW artifact does not support the relationship between comments and text.
•
Individual writing spaces are dissociated from collaborative spaces. To write together is hard with the CSCW artifact. Co-authors individually compose and revise only their own parts of the text. The shared document is the result of different individual parts of text added to the shared file. The CSCW tool does not take into account the interplay between private and collective spaces of action characterizing collaborative writing processes.
•
Individual writing pace is dissociated from a collective writing pace. To write collaboratively is not a matter of sharing a common writing support but rather a more complex process that entails the coordination of individual writing paces. The synchronicity of the CSCW tool presupposes that each co-author plan, compose and revise at the same time.
4 Implications for the design of computer-support for collaborative writing Since the 90's many technical solutions have been developed for supporting collaborative writing activities [see 37]. Despite this, few tools for collaborative writing are used in real settings today. The system CoHaboracio is as an example of a collaborative writing prototype that attempts to take into account both integrative and organized writing activities. Col.laboraci6, which has been developed in our department since 1997, is a web-based system aimed to support the planning and revision phases of the writing process using textbased asynchronous communication among distributed co-authors. The system considers a document as a set of sections to which users can append comments (see figure 2).
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
149
Command frame Comment counter tag
In comparison to others collaborative writing systems requiring a special software installation or the set-up of a network protocol, (e.g. WebDAV [42]), Co.laboracio requires only a Web browser supporting JavaScript and an Internet connection. Some of the features that Col.laboracio presents can be regarded as potential tools to facilitate the complexity of collaborative writing. We describe them in the following eight points : 1 Collaborative writing tools should support focused discussion on two levels. One level taking the document as a whole and the other on a particular part of the document The system Col.laboracio creates for every section of a document a shared space to which co-author can append comments. The commenting feature offered by the system is the core of the communication. Co-authors can make comments in connection with each section, referring to the content of the section and/or to previous comments. Once a comment is appended, it is available to all the co-authors. This dialogue is the base for discussion of plans, coordination, negotiation, and revision focused on that particular section. The content of the section and the comments made so far on it are visible when the user select this section. As every section can be explored individually from the whole document and also its comments, the system supports users to focus on the content of the section in an easy way. Furthermore, the system reserves one space that is called "Ideas for this paper" that has been used by users to organize the document and to coordinate their tasks of the whole document. User studies made on Co.laboracio show that this section was extensively used and in many cases used as a coordination center for the task [30], [31].
150
T. Cerratto and H. Rodriguez/Computer Supported Collaborative Writing
2 Collaborative writing tools should not detach authors from their working environment and allow them to continue to use the word processor they are used to. Col.laboracio does not intend to support collaborative composition and synchronous interactivity. In fact, the composition phase of a shared document is regarded as an individual aspect of collaboration for writing. In this way the system allows co-authors to work in their traditional environment for the composition phase. Asynchronous interactivity is considered more flexible than synchronous in collaborative writing interaction. The system uses the HTML format for saving the document and the comments. This format can be used on different computer platforms and thus to alleviate one of the major problems of collaborative writing : the incompatibility of documents written with different word processor on different computers. 3 Collaborative writing tools should inform about co-authors activity and about the writing pace of co-authors. In Col.laboracio, an email message is automatically sent to the co-author when a comment is made on the common document. The comment is attached to the email and its subject is labeled in such a way that the receiver would recognize that the message is sent by the system and also to which section the comment belongs. Similarly, email is used for notifying the group when a new section is added or removed; this makes the author aware of what a particular member have been doing. This approach of awareness is known as shared feedback [9], i.e. presenting feedback on individual users' activities within the shared space (p. 112). However, receiving a lot of email could be overwhelming. The idea of implementing the concept of coupled/uncoupled awareness [13] has been discussed in future versions of Col.laboracio. Generally speaking, that might mean to give users the possibility to decide how often they want to get an email, and to select the sections that they might be interested in from a perspective of awareness about new comments. Consequently, co-authors could be informed about the ongoing work or others' actions on the document without necessarily entering the system. 4 Collaborative writing tools should allow users to have different views of the document in relation to his/her interest on the document. During the writing process it may happen that one author is more coupled to certain parts of the document. The document in the system is a set of sections that can be handled as modules. Users can select the modules (sections) and form a view of the document. While the author uses the system, the index list presented in the left frame always presents the structure of the document. Co-authors can include a new section, modify its content or the structure of the document, and remove any other section. A section and comments on the section is saved in a file of its own. The system handles the sections as modules, which can be used as a filter mechanism both for having an overview of the document and for creating a version of the document. Co-authors can select a sub-set of sections including or not including comments to provide an overview of the common document. This subset does not necessarily have to be several contiguous sections. For example, one co-author can be interested only in two sections that could be separated in the structure by other sections, say, the introduction and the conclusion of a document. The system allows the user to have a view of the document that includes only those sections. Breaking the linearity that ordinary word processors imposes on document structure deserves further consideration. The awareness mechanism
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
151
discussed in point 3 can also be used to focus attention on the work. 5 Collaborative writing tools should let users share information that may be relevant for the writing task. Comments made in Col laboracio can hold HTML tags. This mean that users could use the tags with two purposes: 1- for formatting reasons; and 2- for including links to the vast information net that the WWW offers. Studies conducted show that co-authors share web links that are relevant to their writing [31]. 6 The integration of a co-author should be supported so that s/he is integrated smoothly to the process by his/her own as far as possible As a whole, the comments that are saved in the system form a dialogue in which users can navigate to obtain awareness of the state of the text and the ongoing collaboration. The comments are presented chronologically and are saved during the whole writing process providing a historical progression of the work. In case a co-author would join the team after the start of the writing, s/he could browse the information that has been kept in the commenting space. This might help the new co-author to understand the process so far. Further more, as comments are associated with a particular section, the new co-author would need only to browse those sections that s/he will be coupled to and thus save time and effort. 7 Collaborative writing tools should provide functionalities that allow co-authors to focus on and exchange information about the content of the document in order to facilitate coauthors to elaborate and transform a common product. When a section is selected in the system, a co-author can view its content and comments. The principle WYSIWIS – What You See Is What I See- is provided by the system and any change of the content is reflected immediately to the others who later access a section. In their communication, co-authors often take account of the whole history of comments in order to know others' reactions to the comments. While other writing tools often lack this feature, we found that supporting the comment history is of great importance to promote communication and awareness among co-authors. The feature facilitates users to engage in the transformation of a shared document. 8 Collaborative writing tools should support private work Providing private workspaces is essential for collaboration. Private spaces for example gives co-authors the possibility to carry out polishing and revision of their contributions before communicating or including them in the shared document ([23], [11]) We believe that providing private spaces could be implemented without locking a segment of the shared text. Also, alternating freely between private and public work should be possible to allow users to choose how they would prefer to communicate through the tool.
152
T. Cerratto and H. Rodriguez/Computer Supported Collaborative Writing
5 Conclusion The study presented in this paper shows a number of transformations that the use of a CSCW tool introduces into collaborative writing and it provides design implications for the development of CSCWriting tools. However, we know that the use of CSCW tools as anticipated by their designers is not enough to make them into instruments for its users, the activity is never defined only by the artifact and the function of an artifact is not an external attribute of this one but comes from an activity of attribution of the subject. Supporting collaborative activities through computer systems is thus not only a problem of supporting shared spaces for common activities but rather also of supporting and facilitating relationships between co-authors through their ideas and texts as well as facilitating the relation with oneself as a co-author. Acknowledgements Thanks to Professor Kerstin Severinsson Eklundh who provided useful comments on this paper. References [1] [2] [3]
[4] [5] [6] [7]
[8]
f9]
[10] [11]
[12]
Beck E. (1993). A survey of experiences of collaborative writing. In Sharpies, M. (ed) Computer Supported Collaborative Writing Springer-Verlag London pp. 9-28. Beck E. (1994). Practices of Collaboration in Writing and their support. Doctoral dissertation. University of Sussex at Brighton. CSRP 340. Beck E. and Bellotti, V. (1993). Informed opportunism as strategy: supporting coordination in distributed collaborative writing. Proceedings of the third conference on Computer Cooperative Work 13, 17 Septembre Milan Italy (Eds). G de Michelis, C. Simone and K, Schmidt., pp. 233247. Beguin, P. & Rabardel, P. (2000). Designing for instrument- mediated activity. Scandinavian Journal of information Systems, Vol.12, pp. 173-190. Beguin, P. (1994). Travailler avec la CAO en ingeniene industrielle : de l'individuel au collectif dans les activit6s avec instruments. Doctoral dissertation. CNAM, Paris, France. Cerratto, T. (1999). Activite collaborative sur reseau. Une approche instrumentale de l'ecriture en collaboration. Doctoral dissertation, University of Paris 8 St Denis, Paris, France. Cerratto, T. (2000). Analyse instrumentale des transformations dans l'ecriture collaborative, suite £ 1'utilisation d'un collecticiel. Proceedings of IC'2000 "Ingenierie des Connaissances", Mai 10-12 . Toulouse. France, pp. 299- 310. Dillon, A. (1993). How collaborative is collaborative writing? An analysis of the production of two technical reports. In Sharpies, M. (ed). Computer Supported Collaborative writing. Springer-Verlag -London pp. 69-85. Dourish P. & Bellotti V., (1992) Awareness and Coordina-tion in Shared Workspaces. The Power of Simple Shared Workspaces. Proceedings of ACM CSCW92 Conference on Computer-Supported Cooperative Work p. 107-114 Ede, l., Lundsford, A. (1990). Singular Texts/Plural Authors: Perspectives on collaborative writing. Southern Illinois University Press, Carbondale and Edwardsville. Ellis, C.A. Gibbs SJ and Rein, Gl. (1990). Design and use of a group editor. In Cockton G. (ed). Proceedings of the IFIP Engineering for Human-Computer Interaction Conference. North Holland, Amsterdam, pp. 13-25. Engestrom, Y. (1999). Activity theory and individual and social transformation. In Perspectives on Activity Theory. (Eds.) Engestrom, Y., Miettinen, R, and Punamaki, R-l. Cambridge University Press. Cambridge, pp. 19-38.
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
[13]
[14] [15] [16] [17] [18]
[19]
[20]
[21]
[22]
[23]
[24] [25]
[26]
153
Fuchs L., Pankoke-Babatz U., and Prinz W. (1995) Supporting cooperative awareness with local event mechanisms: The groupdesk system. In Marmolin H., Sundblad Y., and Schmidt K., editors, Proceedings for the 4th European Conference On Computer Supported Cooperative Work, Kluwer Academic Publishers, Dordrecht, pp 247-262. Haas, C. (1996). Writing Technology: Studies on the Materiality of Literacy, Lawrence Earlbaum Associates, Mahwah, New Jersey. Higgins, L., Flower, L., Petraglia, J. (1992). Planning text together. The role of critical reflexion in student collaboration. Written communication, Vol, 9. No 1. January, pp. 48-84. Johnson, D and Johnson, T. (1991). Learning together and alone. Cooperative, Competitive and Individualistic learning. Englewood Cliffs : Pentice Hall. NJ. Kerbrat-Orecchioni, C. (1992). Les interactions verbales, Tome II, Armand Colin. Paris. Kim, E., (2001). Computer Support for Collaborative Reviewing of Documents. Doctoral Dissertation. Stockholm University and department of Numerical Analysis and Computer Science. Royal Institute of Technology. ISBN 91-7283-182-0. Kim, H. C. and Severinson Eklundh, K. (2001): Reviewing practices in collaborative writing. Computer Supported Cooperative Work: The Journal of Collaborative Computing, 10(2), pp. 247259. Kraut, R., Egido, C. et Galegher (1990) Patterns of contact and communication in scientific research collaborations. Intellectual Teamwork : Social and Technological Foundations of Cooperative Work (eds.). Galegher, Kraut et Egido. LEA, Hillsdale, New Jersey, pp. 149- 172. Kuutti,K., Kaptelinin, V. (1997). Cognitive tools reconsidered: from augmentation to mediation. Proceedings of the Second International Conference on Cognitive Technology, CT'97, Aizu, Japan, Marsh, Hehaniv, Gorayska eds., IEEE Computer Society, Los Alamitos. pp. 31-32. Neuwirth, C., Kaufer, D., Chandhok, R. And Morris, J. (1990). Issues in the design of computer support for co-authoring and commenting. Proceedings of the third conference on CSCW'90. Baltimore, MD : ACM Press, pp. 183-195. Olson, JS., Olson GM, Mack, LA and Wellner, P. (1990). Concurrent editing : the group's interface. In Diaper, D. Gilmore D., Cockton, G. and Shackel, B. (eds). Human-Computer Interaction: Proceedings of Interact'90. Elsevier, Amsterdam, pp. 835-840. Plowman, L. (1993). Tracing the evolution of a co-authored text. Language and Communication vol. 13. nro 3 pp. 149-161; Plowman, L. (1995). The interfunctionality of Talk and Text. In Bruce, B. et Sharpies, M. (eds). Special Issue on Computer Supported Collaborative writing, Computer Supported Cooperative work An International Journal, .pp. 229-246. Posner, I. (1991). A study of collaborative Writing . MSc. Dissertation University of Toronto Canada.
[27]
Rabardel P. (1995). Les hommes et les technologies: contemporains. Paris. Colin.
[28]
Rabardel, P. & Samurcay, R. (2001) (in press) From artifact to instrument-mediated learning. Paper presented at the International symposium 'New Challenges to Research on Learning', March 21-23, 2001, Helsinki, Finland. Rimmershaw, R. (1992) Technologies of Collaboration. In Sharpies M (ed) Computers and Writing : Issues and Implementations. Kluwer Dordrecht. Rodriguez, H., Kim, H-C, & Severinson Eklundh, K. (1999). Poster presented at ECSW'99, Copenhagen, 1999. Rodriguez, H. (2001) Using the WWW as infrastructure for collaborative production of documents. ISBN 91-7283-128-6 Licentiate thesis. The Royal institute of technology, Stockholm, Sweden. Rogers C.R. (1966). Le developpement de la personne. Translated by E.L.Herbert Chapter XIII, pp. 229-237.
[29] [30] [31] [32] [33] [34]
[35]
Approche cognitive des instruments
Saunders, W. (1989). Collaborative writing tasks and peer interaction. International Journal of educational research, 13. pp. 101-112. Severinson Eklundh K. & Linell, P. (1983). The structure of a minimal communicative interaction. In: Karlsson, F, (ed.) 1983. Papers from the Seventh Scandinavian Conference of Linguistic. University of Helsinki, Department of General Linguistic. Severinson Eklundh K (1986). Dialogue Processes in Computer-Mediated communication. A study of
154
T. Cerratto and H. Rodriguez / Computer Supported Collaborative Writing
letters in the COM system. Doctoral dissertation. University of Linkoping, Sweden. ISBN 91-4005164-1. [36]
Schmidt, K and Simone. C. (1996). Coordination Mechanisms: towards a conceptual Foundation of CSCW Systems Design. In Computer Supported Cooperative Work : The Journal of Collaborating Computing number 5, Kluwer Academic Publishers, Netherlands, pp. 155-200.
[37]
Sharpies, M., Goodlet, J., Beck, E., Wood.C. Easterbrook, S. and Plowman, L. (1993). Research Issues in the Study of Computer Supported Collaborative Writing, in Sharpies M. (Eds). Computer supported Collaborative writing Springer-Verlag London pp. 1-7.
[38]
Sharpies, M. (1996). An account of writing as creative design. In C.M. Levy 6 S. Ransdell (eds.). The Science of writing. Hillsdale, NJ : Lawrence Erlbaum. Sharpies, M. and Pemberton, L. (1988). Representing writing : An account of the writing processe ith regard to the writer's external representations. In Cognitive Science Research paper CSRP Nr. 119. School of Cognitive and Computing Sciences University of Sussex, UK. Tammaro, G., Mosier, J., Goodwin, N. and Spitz, G. (1997). Collaborative writing is hard to support: A field Study of Collaborative writing. In CSCW : he Journal of collaborative computing 6. Kluwer Academic Publishers. Netherlands, pp. 19-57.
[39]
[40]
[41] [42]
[43] [44]
Vygotsky L. (1934/1997). Pensee et Langage. La dispute. Translated by F. Seve. Whitehead, E. J. Jr. & Goland, Y. Y. (1999). WebDAV: A network protocol for remote collaborative authoring on the Web. Proceedings of the Sixth European Conference of Computer Supported Cooperative Work, Copenhagen, Denmark, pp. 291-310. Wojahn, P., Neuwirth, C. and Bullock, B. (1998). Effects of interfaces for annotation on communication in a collaborative task. In Proceedings of CHI 98 Los Angeles CA pp. 18-23. Zammuner, V. (1995). Individual and Cooperative Computer-writing and revising . Who gets the best results ?. Learning and instruction, Vol. 5, pp. 101-124.
Cooperative Case Studies
This page intentionally left blank
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press, 2002
157
A Common Work Space to Support the Cooperation in the Cockpit of a Two-Seater Fighter Aircraft Marie-Pierre PACAUX-LEMOINE & Anthony LOISELET Laboratoire d'Automatique et de Mecanique Industrielle et Humaine, UMR CNRS 8530, Universite de Valenciennes et du Hainaut Cambresis,Le Mont Houy 59313 Valenciennes Cedex 9, France Phone : 0033 (0)3 27 51 10 90 - Fax: 0033 (0)3 27 51 10 94 e-mail: {lemoine, loiselet}@univ-valenciennes.fr Abstract: More and more studies underline the necessity today to build common work space to make easier the cooperative activity between different actors, which can be human or artificial whatever work domain. Our paper relates an experiment which aimed at building assistance tools to support this cooperative activity. We present an experimental study that took one's place in a military air base, with twelve operational two-seater fighter aircraft crews. This study deals with the evaluation of two cooperative assistance tools, one of which supports the short-term activity and the other one the long-term activity. After reviewing the theoretical framework of the common work space, we present the field of application and the experimental study. Main results come from the analysis of the answers to questionnaires and the analysis of the coding of the cooperative activity aided by the verbal reports. The results underline the interest for both assistance tools but they have to be improved, in particular by including short-term and long-term assistance tools in a single assistance tools integrated into existing cockpit visualization such as the Head Down Display which sets out the tactical situation. Keywords: CSCW, Shared Workspace, Cooperative Models, Experimental study, Protocol Analysis
1
Introduction
A scientific collaboration between Dassault-Aviation and the CNRS (Scientific Research National Center) began three years ago on the topics of human-human cooperation and its assistance in the field of fighter aircraft piloting. The research program has many goals. The first one consists in studying cognitive cooperation in the cockpit of a two-seater fighter aircraft, between the pilot and the weapon system officer. A preliminary study leads to study their real tasks and to observe their activity in a simulated situation [16]. The second point tries to suggest some ways of designing assistance tools for this cooperation in unexpected conditions (deteriorated situations or new missions) but also to use the cooperation principles to develop assistance tools between the pilot and other actors of the battlefield, for example the
158
M.-P. Pacaux-Lemoine and A. Loiselet/A Common Work Space
controllers of AWACS (Airborne Warning And Control System). Particularly, we hope that the understanding of human-human cooperation may lead to propositions of assistance tools to support cooperation between a remote human operator and one or several UCAV (Unmanned Combat Air Vehicle) which could make a low altitude flight over the battlefield. The cooperation between agents can be co-located but also remote : it is the case of a pilot who flies over the battlefield, and a controller in the AWACS. The pilot has a good idea of the tactical situation thanks to his embedded systems (camera, radar, ...). The controller may give strategic elements in order to allow the pilot to re-plan the mission. Cooperation can be synchronous or asynchronous. Indeed, the pilot and the controller can not always stay in communication because they could be detected by the enemy, because the pilot can be overloaded or because communication device is unavailable. Therefore, during the briefing of the mission, they identify several zones of communication in which the verbal communication is protected. Between these zones of communication each operator works individually, they have not the same information about the evolution of the situation, so when they are in contact they have to adjust their cognitive representation of the battlefield. This updating is today mainly realized by verbal exchanges aided by different radar views. But now, with the data-link improvement, we can imagine that all actors of a battlefield would be able [6] . • To share a display on which they could have the information obtained by an other actor. • To supply the display with his understanding of the situation by providing his information gathering, diagnosis, decision-making and actions in a more or less automatic manner. Thanks to this display, actors would be able : • To take advantage of the results of an other actor activity because the latter has time to obtain them, or because this actor is a specialist in an other area. • To detect a difference between his understanding of the situation and the one of an other actor. • To explain their point of view if they have time to debate. This type of display called a Common Work Space in this paper can take different names in the literature. It is an idea developed by economic, social, human and engineering sciences and mainly reported by Computer Supported Cooperative Work research community.
2
Common Work Space
Many disciplines have already worked on the Common Work Space idea, but before trying to define it, a first step is to give a definition of cooperation. From the psychology point of view, Hoc [10] gives a definition of cooperation according to two concepts : • Interference management : Each one strives towards goals and can interfere with the other on goals, resources, procedure, etc. It would appear when two agents have to discuss a difference in their understanding of process evolution. For Castelfranchi, a "Common World" implies that there is interference among actions and goals of agents : the effects of the action of one agent are relevant for the goals of another : i.e., they either favor the achievement or maintenance of some goals of the other's (positive interference), or threat some of them (negative interference) [3].
M.-P. Pacaux-Lemoine and A. Loiselet / A Common Work Space
•
159
Facilitation : Each one tries to manage the interference to facilitate the individual activities and/or the common task when it exists. It is the action to make easier the activity of the other agent. Facilitation consists of two main activities : the identification of the other agent's activity and helping the other agent, which will be realised by a task sharing. An agent wants to cooperate with another one because the latter, human or artificial one, is able to answer to one of the three objectives stemming from the three forms of cooperation of the sociology point of view [21]. Differences between these forms come from the inability and/or the incapacity of an agent to perform alone a task. The agent wants to increase : • The reliability of the human-human or human-machine system. The debative form implies that the know-how of the agents is the same. The task is not decomposed into sub-tasks but realised by each agent at the same time to underline the differences and to choose the better result. • The capacity of the human-human or human-machine system. The augmentative form appears when the capacity of an agent on his own is not sufficient to realise a task (for example in the case of an overload). The know-how of the agents is similar so the task is decomposed into similar sub-tasks allocated to each agent. • The adaptability of the human-human or human-machine system. The integrative form appears when the know-how of an agent on his own is not enough to realise a task. Other agents with different know-how have to participate to the task realisation. The task is decomposed into many sub-tasks which are allocated to each agent according to his/its know-how. The know-how is the cleverness to solve problem. An agent has the abilities to build a knowhow thanks to knowledge and experiences. It is composed of four main activities [9], information elaboration, diagnosis, intervention decision, and action. These classes of activity will be decomposed into sub-classes describing the evolution of the state of an information. Information Elaboration: It is the perception of the information provided by captors on the process. It is an information gathering goal if the agent tries to organize the information gathering. It is an information recovery if it is recorded in memory. Diagnosis: It is an identification or an inference or a testing. The identification is the interpretation of an information into a categorised one. Inference is also an interpretation, but it is uncertain. Testing is the test of an inference with new information. Decision Making: A decision is schematic or precise and can be evaluated. A Schematic Decision specifies a goal to be reached. It reflects the schematic feature of intervention decision planning. A Precise Decision is used when the decision is fully specified. A Decision Evaluation corresponds to the control or evaluation of one's own action or another agent's action. A process has rarely direct access. In the case of supervision, human operators have to analyse displays which provide a representation of the process. These displays are here called the external current representation (ROE) (cf. Figure 1). The know-how and its interaction with the process through the external current representation allow the agent to build her/his/its internal current representation of the situation (ROI). It is in the mind of the human agent, recorded somewhere in the memory of the artificial agent. The internal current representation is made up of some different attributes: information (stemming from activities of information elaboration); problems (stemming from activities of diagnosis); strategies (stemming from activities of schematic decision making); solutions (stemming from activities of precise decision making); commands (stemming from activities of implementation of solutions) [15].
160
M.-P. Pacaux-Lemoine and A. Loiselet / A Common Workspace
Figure 1 : Model of Cooperation ROI: French acronym for Internal Current Representation RCI: French acronym for Internal Current Frame of Reference ROE : French acronym for External Current Representation RCE : French acronym for External Common Frame of Reference
To cooperate an agent needs to build a know-how-to-cooperate [18]. It is mainly based upon the definition of the cooperation given above. Cooperative activities can also be organized into three levels [7], at the same time implying an increase in the abstraction level and an enlargement of the time span : • Cooperation in action groups together activities that have direct implications in the short-term and that rest on a local analysis of the situation without challenging the internal common frame of reference. This level includes local interference creation, detection, and resolution. It also integrates the anticipation of interference by identifying the goals of the other agents in the short-term. We distinguish three main types of interference : precondition, interaction and mutual control [17]. • Cooperation in planning consists in maintaining or/and elaborating an internal common frame of reference (RCI ) (cf. Figure 1). Its maintenance and elaboration concern common goals, common plans, role allocation, action monitoring and evaluation and, common representations of the environment. So, agents exchange information, problems, strategies, solutions and commands for sharing their/its own internal current representation. The internal common frame of reference and the internal current representation have the same structure. Its elaboration is triggered by the interference which may appear during the comparison between the internal current representation of the agent A and the interpretation of the internal current
M.-P. Pacaux-Lemoine and A. Loiselet/ A Common Work Space
161
representation of the agent B by the agent A (or between ROI B and ROI' A by the agent B). Three forms of interference management may be used by the agent : negotiation, acceptance, imposition. These forms imply, for human agent, cognitive and communication costs which are different. The negotiation aims at reducing the differences between both internal current representations by modifying one of them, on the basis of explanations between the agents. The acceptance is the update of the internal current representation from the interpretation of the internal current representation of the other agent. This acceptance is chosen when the cost of a negotiation is too important or when an agent wants to facilitate the activities of the other. The imposition corresponds to the opposite of the acceptance. • Meta-cooperation, situated at a much higher abstraction level, allows the agents to improve the cooperative activities described above by elaborating long-term constructs, such as a common code to communicate easily and shortly, compatible representations formats, and above all models of one self and of the other agents. The meta-cooperation provides the agent with a model of the other agents. The model allows the agent to build an interpretation of the internal current representation of the other agent (cf. Figure 1). Nevertheless this interpretation can be false. The internal common frame of reference has no external support. As we explain above, it is partly based on the interpretation of the internal current representation of each agent. One way to avoid misinterpretation is to support the cooperation by implementing the RCI on a machine, so called the external common frame of reference (RCE), or also called Common Work Space [15]. The notion of common work space is already proposed by several authors in various work domains. In the case of Air Traffic Control, Bentley, Rodden, Sawyer and Sommerville [1] present a shared work space which provides an adapted presentation of the air traffic to different users, on different machines, to make the use of shared entities easier. In this work, the shared work space is limited to the external environment and process information representation. Other authors enrich the work space by different elements relative to the activity of each agent, and by not only raw information but also elaborated ones. Decortis and Pavard [4] define the shared cognitive environment as a set of facts and hypothesis which are a subset of the cognitive universe of each agent. In the specific field of Knowledge Based Systems, Brezillon and Pomerol [2] pointed out the necessity for a user and the system to share a contextual knowledge i.e. "the sharing of the cognitive representations, an agreement about the contextual knowledge of problem solving, the ability to follow each other's reasoning, the ability to exchange explanation and the control of the interaction by the user". Royer [20] specifies that a system is all the more cooperative since the cooperation implies more levels as perception, analysis, decision and action. Karsenty [12] underlines the relevance of a shared problem representation to support the process of explanation during which co-workers articulate their individual problem representations and thus, build a richer shared problem representation, rend a deeper analysis of the decision space, and thus reach a better decision. Sonnenwald and Pierce [22] deal with Interwoven Situational Awareness which is "the continuous extraction of environmental information, integration of this information with previous knowledge to form a coherent mental picture in directing further perception and anticipating future events". They propose an interwoven situational awareness that includes the individual, intra group and inter group situational awareness for the Control and Command
162
M. -P. Pacaux-Lemoine and A. Loiselet / A Common Work Space
team members who need to collect, synthesize and disseminate information to create an understanding of the current battlefield situation and to anticipate future battlefield events. The situational awareness is based on information about the dynamic work goal and situation (the environmental information : mission, enemy, terrain/weather, troops and time available), the work process (tasks, formal and informal task procedures) and specialized field knowledge (information about battle tactics, decision analysis algorithms or methods, biological chemicals, enemy profiles/characteristics, telecommunications networks, civilian government,...). Jones and Jasek [11] integrate this idea into the building of an Intelligent Support for Activity Management (ISAM) which allows to share goals, contextual information and allocation of functions between the agents. Gutwin and Greenberg [5] highlight the importance to integrate the activity of the other agents and deal with a workspace awareness which is "the up-to-the-moment understanding of another person's interaction with the shared workspace". The shared workspace allows to know who is working, what they are doing, where they are working, when various events happen and how those events occur. This knowledge is provided by the environment, and is exploited to perform an action or to be used to do an exploration. In this context, the goal of our study is to determine the impact of different levels of development of the external common frame of reference, based on the structure of the internal common frame of reference, on the cooperative activities in the cockpit. The first axis of development of the external common frame of reference consists in enriching the representation of team activity. The second axis distinguishes two formats of representation of the activity. The first format is a temporal one, and the second format is a spatial one. The integration of a richer external common frame of reference could improve the cooperative activity, but we have to determine if it does not imply a cost. The agent has : • To update the external current representation on the base of each internal current representation. • To update each internal current representation on the base of the external current representation. • To manage the interference between both external current representation to build the external common frame of reference. Our hypothesis is that an enriched external common frame of reference would support: • The management of the internal common frame of reference. • The communication between the two agents by providing common and objective external representation of the situation. • The management of interference between the operator at the planning and action level of cooperative activity. These elements are evaluated in the following application.
3 3.1
Implementation Work area
The study deals with the cooperation between a Pilot and a Weapon System Officer (WSO) of a two-seater fighter aircraft carrying out an Air-To-Ground mission. An a priori and strict task sharing exists between both members of the crew: The pilot is responsible for the short-term
M.-P. Pacaux-Lemoine and A. Loiselet/ A Common Work Space
163
management of the situation (piloting, firing) whereas the Weapon System Officer manages middle and long-term processes. An important part of the Weapon System Officer work consists in setting up the work of his team-mate. However, in our experimental device the pilot as well as the Weapon System Officer can perform every tasks (dynamic task allocation is possible like in actual work but in a larger extent). A mission is divided into three or four main stages: (i) going near to the target (entering the enemy territory and flying to the target, steering clear of enemy threats), (ii) handling one or two targets (reconnaissance or destruction mission), and then (iii) leaving the enemy territory.
3.2
Experimental platform
In collaboration with Dassault-Aviation, a generic two-seater fighter aircraft simulator was built (cf. Figure 2). It is made up of three workstations.
experimental room
supervisor and wizard of oz workstations
Figure 2 : Experimental platform
One station is put in a one-place cockpit for the Pilot and a second station is put on a table for the Weapon System Officer. Both operator's stations are managed from a third station (composed of three interfaces) also dedicated to the management of scenarios and experiments, and the Wizard of Oz simulation (cf. Figure 2). The operators' interfaces (identical for each crew member) are made up of : First, a big picture screen (cf. Figure 3) putting together the four main displays of a military aircraft cockpit, the assistance tools and some device controls (e.g. automatic pilot) that can be activated by the way of a touch screen. Second, the simulator can be mainly controlled by a Hands On Throttle And Stick (HOTAS) device and by voice-control (simulated by the Wizard of Oz). The simulation of a data-link communication allows to display messages coming from a simulated AirBorne Control, Command and Communication (AB3C) by drawing or writting pieces of information about new threats or new targets. On the upper part of the screen can be seen (from the left to the right), a window displaying the head of the team-mate, a Head Up Display (HUD) presenting the outside world and short term flight parameters, and a window displaying the assistance tools.
164
M.-P. Pacaux-Lemoine and A. Loiselet / A Common Work Space
On the lower part of the screen are, two Head Lateral Displays (HLD) allowing to modify the flight plan (FPL), to see the way-point (BUTS), to prepare weapons (ARME), to detect threats (CM), to manage the radio (RADIO), and to control bomb guidance equipment (VPDL), and a Head Down Display (HDD) setting out the tactical situation (at the center). This big picture defines an External Current Representation of the situation and it is already a shared work space because both crew members see the same information and perceive in the same way the action of each other. The assistance tools aim at enriching this external common frame of reference.
Figure 3 : Big picture screen
3.3
The assistance tools
A long-term assistance tool (cf. Figure 4) can be looked upon as an internal common representation of the situation of both agents. It is displayed in the assistance dedicated window (cf. Figure 3). The content of this assistance is mainly made up by the crew during mission planning. The operators can insert all information they want, along a spatial representation of the flight plan divided up into several pages. The added information are mainly selected from the paper document describing the features of the mission (mission orders, targets, flight plan,
M.-P. Pacaux-Lemoine and A. Loiselet/A Common Work Space
165
flight rules, and so on) and given by the experimenter. During low workload periods of the mission, the pilot or the Weapon System Officer can modify the information displayed on this assistance, by changing pages, by modifying, adding or removing information. The Wizard of Oz performs all these changes on request of the crew member (voice command). The Wizard of Oz also manages the automatic update of the assistance (page changing) according to the progress of the mission. So, the crew builds a common current representation of the situation in order to be ready if a crisis hits.
Figure 4 : Long-term assistance tool, the spacial representation of the flight plan
Figure 5 : Short-term assistance tool, the diary of major tasks
A short-term assistance tool is called a diary of major tasks (cf. Figure 5). This assistance tool can be looked upon as a common illustration of the state of the activity of each other (stemming from the know-how). It is first made up by the experimenters. During the briefing of the mission, the crew can remove, add or modify the information presented on the display. It consists of tasks that the pilot or the Weapon System Officer has to perform during the mission: ATC or AB3C radio contacts, activation or deactivation of functions and function's options, altitude and speed control, weapons' preparation, among others. As well as for the long-term assistance tool, the crew can remove, add or modify tasks (e.g. to bring forward or to put back in time) during the mission, using the voice control simulated by the Wizard of Oz. The latter is also responsible for a more complex task : the continuous updating of the progress state of the tasks displayed. The Wizard of Oz has to follow the activity of the Pilot and of the Weapon System Officer by the way of monitoring the verbal exchanges between them and the evolution of their display (of which s/he has video reproductions) in order to modify the state of the displayed tasks. The progress state is coded by a color, it is yellow if a task is no realized, purple if a task is in progress, magenta if a task is interrupted, and removed if a task is finished. When a task would not be completed at the proper time, the Wizard of Oz has to put
166
M.-P. Pacaux-Lemoine and A. Loiselet / A Common Work Space
the task in red (coding an alarm), and to send a written alert message in the main working window of the crew member responsible for the target task. The voice control is mainly used for managing assistance tools. The pilot or the Weapon System Officer has to push the radio switch and to speak in a microphone using a specific word list to control the updating of the assistance tools. The Wizard of Oz recognizes the command and applies what the crew member asks (thanks to a specific interface), or sends a written error message. This simulation is used to make the control of the assistance tools easier instead of developing a new control display that would overload the display of the operators. It also allows to overcome some technical constraints.
Configuration of the workstation of the team-mate In the same idea, another type of assistance tool is given. The assistance consists in providing the configuration of workstation of the partner. This information is displayed at the bottom of the interface of the assistance tool (cf. Figure 5). One agent is able to know what his partner is able to do and/or to see (because he has the good visualization to do it, and/or to see it). The first information (OFF/ON) indicates if the video camera is selected or not. The second one (CM/BUTS/VPDL/ARME) and the third one (VPDL/HSI/FPL/RADIO) indicate which page is opened on the HLDs. The fourth one gives the name of the assistance tool used (TACHES/ETC). If one page is broken, the word VIDE (empty in English) appears. The aim of this assistance is to allow the crew members to make a new allocation of tasks if one page is broken, or to infer the information the partner has. During the mission, these pages represent cooperation support because they display shared information.
Video picture The Figure 3 (the window on the upper left part) shows the place of the video picture on the workstation in which each crew member can see the face of his partner. This assistance could help the crew members to know in what a state is his partner by analyzing gestures, gesticulations, facial expressions, in addition to his voice which is already a good indicator (intonations, grunts). This assistance tool could allow the maintenance and the updating of the model of the partner. Exchanges of written messages The crew members can send short sentences to the partner. A button on the stick allows the crew members to use a voice control simulated by the Wizard of Oz. A window allows to display the last 15 messages exchanged with a color code to distinct the pilot's messages, from the WSO's messages, from the AB3C's messages. The interest of this type of assistance is the persistence of the message. The operator can read the message when he is not overloaded. This assistance tool is a short-term one that allow the operator to slightly anticipate the information they pass (for example, to prepare a command that must be done in overloaded conditions). It could also be looked upon as an assistance tool to manage interference.
M.-P. Pacaux-Lemoine and A. Loiselet/ A Common Work Space
167
Many tasks for the Wizard of Oz The tasks were shared among three operators. The first one was responsible for the communications between the operators and the simulated system. S/he decided the validity of the voice commands sent by the operators and sent back a message pointing to the acceptance of the command (and a wizard performs the corresponding actions) or an error message. This wizard also managed the long-term assistance tool (automatic updating and updating in demand). At least, s/he controlled the simulation (starting up and stopping, data recording) and activated the breakdowns planned in the scenarios. The second one was responsible for the management of the scenarios and the data-link communications and, played the role of the controllers (ATC and AB3C). S/he could help his/her team-mates (particularly for the management of the task diary) when her/his workload was low. The third one was responsible for the management of the diary of the major tasks: its updating in real time and the simulation of the voice commands (when accepted by the first wizard). Generally speaking, it seems that the simulation of such a continuous process requires at least one dedicated operator. These three operators were not experts of the work domain. Two of them had acquired some field knowledge during their work on the research program (a psychologist-ergonomist and a supervisory control researcher). The third one (a student in psychology who was responsible for the task diary) had been trained by the two others. S/he had a document at one's disposal, specifying the rules of evolution of the state of each task and explaining the actions to perform in response to the voice commands (a similar explanation is available for the management of the long-term assistance tool). 4
Experiment
Twelve crews of French Air Force of the military base 116 accepted to test the assistance tools we proposed. The crews performed three missions in three conditions : • The first one without assistance tool is the reference one. • The short term assistance condition which provides the operators with the diary of major tasks, the exchanges of written messages, the configuration of crew member's workstation assistance and the video picture assistance • The long term assistance condition which provides operators with the spacial representation of the mission, the configuration of crew member's workstation assistance and the video picture assistance. The scenarios were different but required similar tasks. The duration of a scenario lasted about 30 minutes. Order, rank, scenarios and assistance conditions were counterbalanced using a greco-latin square experimental design. After a presentation of our scientific objectives, we gave a quick presentation of the experimental platform. The workplace of the crew and of the supervisor-Wizard of Oz were in the same room, but separated from one to another by a folding screen, because of field constraints. So we had to explain the presence of this additional workplace. We only present the simulation supervisor aspects of this device, letting crews believe that the system is fully functional (not simulated by a wizard).
168
M. -P. Pacaux-Lemoine and A. Loiselet / A Common Work Space
After this, the operators were acquainted with the flight simulation, the assistance tools and the type of mission which was not their usual mission. Each experimental session was divided into five stages : mission planning, preparation of assistance tools, briefing, mission completion, and debriefing (relative to the cooperative activity and to the assessment of the assistance tools). The briefing was managed by one member of the crew, he expressed all the information of the mission file following a precise order. The partner listened and reacted if he had a different interpretation of the information or what it was asked. The briefing implies an allocation of tasks between both members. After the briefing, the crew filled the mission. They had two objectives to destroy, or one objective to destroy and another one to recognize. The spontaneous verbal reports and the video output of both workstation were collected. In addition, we used a method of self-observation reports after task execution ("selfconfrontation") using the video records to control inferences derived from the concurrent verbal reports. They had to answer to different questionnaires aiming at : characterizing the crews, obtaining information about the cooperation in the cockpit, evaluating the flight simulator, the scenarios and of course the assistance tools.
5
Results
The results stem from the analysis of the answers to the questionnaires, the analysis of cooperative activity during mission completion, and the analysis of assistance tool use. 5. /
Answers to the questionnaires
Short-term assistance tool. For more than half of the operators, this assistance tool favors the cooperation by aiding the mutual control. It is mainly red, and not modified during the mission. It seems to be useful to realize the task. The use and the presentation of information were fairly easy but the management of the tasks seems to ask too many time and too many resources to be achieved by the crew members during the mission. The updating of this assistance tool must be automated. It is used as a checklist to follow to achieve the mission in order not to forget any information gathering or diagnosis or action fulfillment. The Weapon System Officers ask to have this type of assistance tool for their medium and/or long term activities. Instead to present the activity during the next three minutes, it could be interesting to give the information before displaying the task in red, by beeping and writing a message in the main visualization. Long-term assistance tool. For half of the operators, this assistance tool favors the cooperation. The experimental procedure implies to use this assistance tool during the preparation of the mission. For nearly all the operators, it is used to perform the mission, and the presentation of the information appears to be satisfactory. The use of the display seems to be very easy. However the crews do not modify it during the mission. No item is added, modified or deleted during the mission. As the short-term assistance tool, this assistance is mainly red, and it seems to be also used for the short-term activity. This tool gives the same information that the notes taken by the crew during the preparation of the mission on the paper map of the flight plan. This assistance tool spares the crew member to ask several information to his partner. For the
M.-P. Pacaux-Lemoine and A. Loiselet/A Common Work Space
169
most part, the crews ask to group the short-term and long-term assistance tools in an existing visualization, the SITAC (French acronym for the Tactical Situation). Configuration of the partner's workstation. For half the operators who answer to the question (only 16 out of 24 do), this assistance favors the cooperation. But they indicate that it is not useful to realize the mission or they do not use it. The presentation of the information is not good, it is perhaps the reason of its no use. However, some crew members react to the word "VIDE" by asking to his partner if he has a problem with his workstation and if he needs help to supply the lack of information. Video picture. All the crews think that this assistance is not useful to realize the mission and it does not favor the cooperation. Only one Weapon System Officer indicates that he used it when he lost his two HLDs, and thanks to this video display he knew when it was possible to ask information to his partner, it was a way to evaluate his workload. Exchanges of written messages. All the operators who answer to the question (8 out of 24), this assistance does not favor cooperation. They also indicate that it is not useful to realize the mission or they do not use it. The mission proposed are not favorable to this type of exchanges. The verbal message is always dominating the written message (easier, and more natural). However this means of communication would be essential when the radio exchanges are forbidden. It would also be a necessity if the partner is not human but artificial as a plane without pilot (UCAV). These subjective results have to be confirmed by other analysis based on more objective data, such as performance analysis and, individual or cooperative protocol analysis.
5.2
Cooperative activities analysis
All the audio-recordings during the mission were transcribed to provide verbal protocol. Aided by the self-observation reports and by the evolution of the tactical and strategic situation, a coding of the verbal protocol was realized. Verbal communication is a privileged means to cooperate but, from a methodological point of view, it is also a means to access to the cooperative activity. For the coding and some analysis of protocols, we used a spreadsheet software which provides editing, searching, and statistical functions. This software is very useful to apply our methodology, without imposing a particular framework, and to elaborate a coding scheme. The coding scheme is building from the cooperative activity model proposed by Loiselet and Hoc [17]. To analyze the cooperative activity, two main entities are coded : the interference which could appear between the goals, the resources, the procedures, ... of the agents, and the common frame of reference management. The cooperative activity of three crews out of twelve are coded. The comparison between the three experimental conditions according to the coding of the cooperative activity gives the following results : • The type of cooperative activity realized in the cockpit is the same in the three conditions. The team cooperation seems mainly to ensue from the prescribed organization of the work.
170
M.-P. Pacaux-Lemoine and A. Loiselet / A Common Work Space
•
The internal common frame of reference content is not affected by the integration of the assistance tools. It is mainly focused on each human operator activities at a low level of abstraction. • The weapon system officer is the most implied in the management of the internal common frame of reference. It is mainly based on the content of the internal current representation of the weapon system officer. This domination is exacerbated in the assisted conditions (for the content relative to the activity of operators). We explain the absence of deep modification of the cooperative activities of the crew by the lack of training of the crew with the assistance tools. In addition, because the cost implied by the assistance use is high, only the Weapon System Officer uses it probably because of his role: contrary to the pilot, he is not continuously involved in the short term management of the mission. As a consequence, the building of the internal common frame of reference becomes more asymmetric and favors the Weapon System Officer's point of view.
5.3
Assistance tools use (action and information gathering)
A coding of the main actions upon the assistance tools underlines that, as the crews explain in their answer to the questionnaires, they never manipulated the assistance tools during the mission. But the coding of the information gathering from assistance tools shows that the crews took information from the short-term and long-term assistance tools. The amount of the information gathering is depending on the experience of the crew. The more experience the crews have, the more they use the assistance tools. This result probably comes from the fact that more experienced crews have a better management of the mission and therefore have more time to read the information presented by the assistance tools. Low experienced crews use all their resources to try to realize the mission, because it is difficult for them and they have not sufficient training with the assistance tools to be helped by them. The weapon system officer seems to gather more information from the assistance tools than the pilot (77% for the shortterm assistance tool, 63% for the long-term assistance tools). This result is coherent with the one from the cooperative activities analysis. The weapon system officer activity is less concerned by the short-term than the pilot, so he seems to have more cognitive resources to read the information provided by the assistance tools. The short-term assistance tool is mainly used during the period of the flight in which the procedural pressure is important (85% of the information gathering from the short-term assistance tool). The long-term assistance tool is used as well during the crew just follow the flight plan (56%) as when they adjust the plan. But the long-term assistance tool is mainly used just before the great constraints of the mission, probably to be sure to be in the right configuration to achieve this critical point of the mission. Operators generally take only one information at each time they read the assistance tools. They certainly take no time to read more information because they have no time to check the current plan. They only take the information they need to perform the current action. The weapon system officers and pilots use the information from the assistance tools for their individual activities as well as for their cooperative activities. In short, the assistance tools support cooperative as well as individual activity.
M.-P. Pacaux-Lemoine and A. Loiselet / A Common Work Space
6
171
Main results and conclusion
The main assistance tools which were used are the short-term assistance tools with the shared diary of tasks, and the long-term assistance tool with the common workspace. The crews seem to find an interest but assistance tools have to be improved. They used these tools during low workload periods, whereas during high workload periods they had not sufficient resources available (especially time) to do so. The updating of these common information must be automated. Our results also point to a lack of training. The crews did not have enough time to learn how to make use of the assistance tools and to embed their use in their usual work procedures as well for information gathering as diagnosis, decision making and action. All information must be integrated into an existing cockpit display. The most interesting display is the SITAC (for tactical situation). This display gives a presentation of the flight plan, the limits of the enemy territory and, the position of all the planes and of the surface-to-air threats. The plan is described by several main way-points and we could join at each way-point a set of essential tasks to achieve the mission. This new presentation of tasks would replace the actual short-term assistance tool. We could also give the possibility to the crew members to add information, whatever it is (a draw, a symbol, a string, and so on), near each way-point like they do with the long-term assistance tool. All these new information can not always stay displayed on the interface without overloaded it. They only have to appear when it is asked by the crew member. The improvement in demand consists in integrated information concerning the know-how and the internal current representation of each agent participating in the process control into the external current frame of reference. The management of them must be automatic in order not to overload human operators. This automatic updating could be realized by a new agent whose role would be the cooperation support, such as a coordinator. It would be easy to detect all agent commands and controls in order to infer at which state of her/his activity s/he is. The latter state would therefore be updated in the external current frame of reference. These improvements have been integrated into a new platform and evaluated during new experiments. Next analysis would allow to have answers regarding the building of better cooperation supports. Acknowledgments This study has been done in collaboration with Dassault-Aviation. We thank the DassaultAviation Istres test base staff Dassault-Aviation Saint-Cloud staff for their collaboration. The experimental field is the Air Force, especially Air Base 116. We thank 2/4 and 1/4 Air Force squadrons for their welcome and their cooperation. We thank also the IMASSA-CERMA staff for their help, J.-M. Hoc and R. Amalberti for their advises during this study and for their comments on a previous version of this paper.
172
M.-P. Pacaux-Lemoine and A. Loiselet / A Common Work Space
References [I]
Bentley, R., Rodden, T., Sawyer, P., Sommerville, I. (1992) An architecture for tailoring cooperative multiuser display. Proceedings of CSCW'92. November [2] Brezillon, P.. Pomerol, J.-Ch. (1999) Contextual knowledge sharing and cooperation. Le Travail Humain tome 62. 3, pp. 223-246. [3] Castelfranchi. C. (1998) Modelling social action for AI agents. Artificial Intelligence. 103, pp. 157-182. [4] Decortis, F., Pavard, B. (1994) Communication et cooperation: de la theorie des actes de langage a I' approche ethnomethodologique. In Pavard B. (Ed.), Systemes Cooperatifs de la modelisation a la conception. Toulouse, France, Octares [5] Gutwin. C., Greenberg, S. (1999) A Framework of Awareness for Small Groups in Shared-Workspace Groupware. Technical Report 99-1. Department of Computer Science. University of Saskatchewan. Canada. [6] Helie, P, & Loiselet, A. (2000). Human-human cooperation in airborne combat system : designing CSCW assistances. In K.Y. Lim (Ed.), Proceedings of the joint conference 4th APCHI / 6th ASEAN Ergonomics (pp.39-56). Oxford, UK : Elsevier. Hoc, J.-M. (2000) From human-machine interaction to human-machine cooperation. Ergonomics, Vol. 43. nº 7, pp. 833-843. Hoc. J.-M. (1999) Conditions et enjeux de la cooperation homme-machine dans le cadre de la fiabilie des systemes. In J.-G. Ganascia , Eds. Hermes, Securite et cognition, pp. 147-164. Paris. [9| Hoc. J.M.. Lemoine, M.P. (1998) Cognitive evaluation of human-human and human-machine cooperation modes in Air Traffic Control . The International Journal of Aviation Psychology. Vol. 8, Nº I. pp. 1 -32. [10] Hoc. J.M. (2001) Towards a cognitive approach to human-machine cooperation in dynamics situations. International Journal of Human-Computer Studies, Vol. 54. pp. 509-540. [11| Jones. P., Jasek, C. (1997) Intelligent Support for Activity Management (ISAM): An Architecture to Support Distributed Supervisory Control. IEEE Systems. Man and Cybertenics. Part A: Systems and Humans. Vol. 27. N° 3, May | 12] Karsenty. L. (2000) Cooperative work : the role of explanation in creating a shared problem representation. Le Travail Humain. tome 63, n° 4, pp. 289-309. [13] Lemoine. M.-P. Debernard, S.. Crevits. I., Millot, P. (1996) Cooperation between humans and machines : first results of an experimentation of a multi-level cooperative organisation in air traffic control. Computer Supported Cooperative Work: The journal of Collaborative Computing. Vol. 5. Nº2–3. pp 299– 321. [14] Lemoine-Pacaux. M.P., Grislin-Lestrugeon, E. (1998) Multi Agents System and Human Machine Cooperation Proc. Of 17th European Annual Conference on Human Decision Making and Manual Control. Valenciennes, France, December 14-16, pp. 101-110 [15] Lemoine-Pacaux, M.P., Debernard. S. (2001) Common Work Space to support the Human-Machine Cooperation in Air Traffic Control. Control Engineering Practice (in press). [16] Loiselet, A., & Hoc, J.M. (1999). Assessment of a method to study cognitive cooperation. In J.M. Hoc. P. Millot. E. Hollnagel, & P.C. Cacciabue (Eds.). Proceedings of CSAPC'99 (pp. 61-66). Valenciennes. F: Presses Universitaires de Valenciennes. [17] Loiselet, A., & Hoc, J.M. (2001). La gestion des interferences et du referenliel commun dans la cooperation : implications pour la conception. Psychologie Francaise. 46, 167-179 [18] Millot. P.. Lemoine-Pacaux. M.P. (1998) An attempt for generic concepts toward Human Machine Cooperation, IEEE SMC, USA, California, San Diego. October 11-14. [ 19] Millot. P. (1999) La supervision et la cooperation homme-machine dans les grands systemes industriels ou de transport. In J.-G. Ganascia , Eds. Hermes. Securite et cognition, pp. 125-145. Paris. [20] Rover. V. < 1994) Partage de croyances : condition necessaire pour un systeme cooperatif ? In Pavard B (Ed). Systemes cooperatifs de la modelisation a la conception, pp. 253-270, Toulouse Octares (p. 129). [21] Schmidt. K. ( 1 9 9 1 ) Cooperative work : a conceptual framework. In J. Rasmussen. B. Brehmer & J. Leplai. Eds. Distributed Decision-Making : Cognitive Models for Cooperative Work. pp. 75-110. Chichester. UK : Wiley. [22] Sonnenwald D. H., Pierce L. G. (2000) Information behavior in dynamic group contexts : interwoven situational awareness, dense social networks and contested collaboration in command and control. Information Processing and Management, Vol. 36. pp. 461-479.
Cooperative Systems Design M. Blay-Fornarino et ai (Eds.) IOS Press, 2002
173
AMANDA Project: Delegation Of Tasks In The Air-Traffic Control Domain Serge Debernard, St6phane Cathelain, Igor Crevits, Thierry Poulain Laboratoire d'Automatique et de Mecanique Industrielles et Humaines (LAMIH, UMR CNRS 8530) Universite de Valenciennes et du Hainaut-Cambresis 59313 Valenciennes Cedex 9, France e-mail: {serge.debernard, stephane.cathelain, igor.crevits,thierry.poulain}@univ-valenciennes.fr Abstract: For a long time air-traffic control is the object of studies to increase the control capacities. Several ways are investigated : improvement of information system, control functions automation, introduction of assistance to controllers,... In the current air-traffic control organization, some cooperative activities appear in the work of air-traffic controllers, between control centers, control position and, into each control position, between controllers. On a control position, the cooperative activities between controllers use a common reference which is the air-traffic represented by the radar screen. Introduce specific assistance based on these common references allow to improve the cooperation between controllers. Moreover the control position is the lowest level in the structure of air-traffic control. On this level, actions are done on the aircraft. An important part of these actions are simple but require a high attention for controllers Increasing the workload of controllers. Introduce basic action assistance allow to reduce the workload of controllers and to maintain into theirs high level tasks requiring their competencies. Keywords: HCI, Shared workspace, Cooperative dialogue system, Experimental study, Dynamic Tasks Allocation, Delegation of Tasks, Air-Traffic Control
1 Introduction This paper deals with human-machine cooperation in the air traffic control domain where highly dynamic situations appear. These situations submit the air traffic controllers, to strong workload variations. Moreover this situation is enhanced by the air-traffic increasing. One way to keep the same level of security, consists in introducing in the human machine system some support tool in order to regulate the air-traffic controllers' workload. Several kinds of support tool have been evaluated in this domain, but we focus our research on the problematic of dynamic task allocation that consists in allocating dynamically the tasks to perform between human operators and on or several support tools. In this paper we present first some previous researches that consisted to evaluate dynamic tasks allocation between air traffic controllers and a support tool which was able to resolve some conflicts between aircraft. The results obtained have shown that dynamic tasks allocation principles allow to obtain a regulation of human workload. Nevertheless, for obtaining an efficient human machine cooperation, it is necessary that all the agents (human and artificial) share their representation of the situation, and the support tool takes
174
S. Debernard et al. /AMANDA Project
into account the strategies of the human operators in order to avoid decisional conflict with human operators. In the next part, we present our new project called AMANDA that consists in evaluating the principle of task delegation where the support tool performs some tasks from a strategy defined by the controllers. Before the design of the support tool, some experiments have been realized in order to characterize the functions that will allocated to the support tool and the contents of the human machine interface. Then, we present this new support tool. This work is realized in collaboration with the CENA (French Air Navigation Study Center). 2
Previous researches: dynamic tasks allocation in ATC
2.1 Human Machine Cooperation One step of the complex system design is the choice of the appropriate level of automation. This step is difficult because the human needs must be considered precisely. A too complete automation can, in a long term debase the human abilities, and in a short term can produce some difficulties for the human operator because he/she can have some problems for having a good "picture" of the process when a failure occurs. On the opposite, an incomplete automation can be not sufficient for keeping the reliability and the productivity of the system, because the human operator can be in an overloaded situation [1]. Automation can be achieved in accordance with two main lines: - The first one consists in the definition of function allocation between the human(s) operator(s) and one or several control systems. This automation consists in allocating statically the tasks or the subtasks to be performed to the agent in accordance with their abilities and their "limits", [2][3]. - The second one, which can be complementary to the first one, focuses on human machine cooperation. The Human Machine Cooperation is a research domain in which several disciplines are interested; a part of them are interested in the description and the understanding of the psychological mechanisms underlying cooperative activities; the other are interested in the definition of tools and of human-computer interfaces for supporting these cooperative activities. It is possible to find in the literature different words as co-action, co-ordination, collaboration, cooperation, [4]. In co-action, the human operators coordinate their actions in accordance with their sub-goals which are dependant from a global goal. So, co-ordination requires some information exchange from the individual activities. Cooperation refers to a situation where human operators work on a same global task and perform operation (action) together (co operation). But of course, cooperation implies co-ordination. So, defining exactly what human machine cooperation is, remains a difficult task. From Hoc [5]: "Two agents are in a cooperative situation if they meet two minimal conditions: - Each one strives towards goals and can interfere with the others on goals, resources, procedures, etc. - Each one tries to manage the interference to facilitate the individual activities and/or the common task when it exists. The symmetric nature of this definition can be only partly satisfied". In this definition, the notion of goal does not refer to the global goal to reach when supervising and/or controlling a process, but to the goal for achieving a particular task. The word "interference" refers to the normal interaction between the activities of several agents, but also to conflicts between the agents, on the results of their activities, or on the means for achieving their tasks. For Millot and Lemoine [6], each agent is characterized by his know-how but also by his know-how to cooperate. This know-how to cooperate is composed of two main
S. Debernard et al. /AMANDA Project
175
functions. The first one allows agents to manage interference between their goals, resources, etc., and the second one allows agents to perform their own activities but taking into account the activities of the other agent for facilitating them. It is clear that in human machine cooperation, it is necessary to improve the interactions between the agents that allow the correct fulfillment of their tasks, and to minimize the possible conflicts. Nevertheless, the Hoe's definition shows the limits of human machine cooperation, especially if the "machine" or computer must facilitate the human activities, because it will be necessary to prevent or detect conflicts with the human operators. Royer [7] specifies that a system is more cooperative when this cooperation integrates several levels, these levels being perception, analysis, decision and action. So, cooperation is not only a coordination of actions between several agents, but depends also on the merging of perceptions, on the confrontation of situation analyses, and on the convergence of decisions. To model these interactions, it is possible to use the three forms of cooperation defined by Schmidt [8]: - Augmentative cooperation: Cooperation is augmentative when agents have a similar know-how but the agents must be multiplied to perform a task too demanding for only one agent. The task is then shared into similar sub-tasks. - Integrative cooperation: Cooperation is integrative when agents have different and complementary know-how and it is necessary to integrate their contribution for achieving a task - Debative cooperation: Cooperation is debative when agents have a similar know-how and are faced with a unique task, and they compare their results for obtaining the best solution. For Grislin and Millot [9], any cooperation situation may be described as a combination of these three forms. But the authors consider that tasks (or sub-tasks) are performed entirely by agents, even if there are some interactions between them before or after the tasks, in accordance to the 3 forms of cooperation. But it is possible to extend this idea, by taking into account the activity level instead of task level [10]. Before to go back over this idea, we will present a particularly mode of human-machine cooperation: the dynamic tasks allocation. 2.2 Dynamic Tasks Allocation Dynamic Tasks Allocation is initially an augmentative form of cooperation. Dynamic Tasks Allocation consists in assisting the human operator, firstly with the integration of an automated system that is able to perform some tasks, and secondly by allocating the tasks between each agent in a dynamic way. So, the automated system must integrate all the functions that are necessary for performing a task entirely, from information elaboration to solution implementation [11]. An optimal Dynamic Tasks Allocation aims at finding a task sharing that optimizes the process' performance and that takes into account the two decision-makers' abilities and capacities. So, Dynamic Tasks Allocation allows to modify dynamically the level of automation for keeping human operators "in the loop": it is because Dynamic Tasks Allocation is also called Adaptive Automation. But the difficulty is to define the set of the shareable tasks in real time that corresponds to the intersection between the set of tasks belonging to the human abilities, and the set of tasks within the assistance tool abilities. In Dynamic Tasks Allocation, the allocation of the shareable tasks is performed by a module called allocator, which the main function is to inform each decision maker about who performs which task and how, Figure 1.
176
S. Debernard et al. /AMANDA Project
Figure 1 : Structure of Dynamic Tasks Allocation
This allocator is controlled by a dispatcher [12]. This function can be performed by an automated system, or by the human operator him/herself [13]: - In explicit Dynamic Tasks Allocation, the dispatchor function is achieved by the human operator with a specialized interface. So, the human operator allocates the shareable tasks with his/her own criteria. - In implicit Dynamic Tasks Allocation, the dispatchor function is achieved by an automated system which performs the allocation in accordance with an algorithm and with criteria defined by the designer. These criteria can require to measure or estimate some human-machine system parameters like the human workload, the system performance, etc. The implicit Dynamic Tasks Allocation has the advantage to discharge the human operator from the management of the allocation. Nevertheless, this mode is more complex to implement than the explicit mode, because the choice of algorithm and criteria requires a serious study for integrating the human operator correctly in the human-machine system. To avoid this difficulty, it is possible to adopt an intermediate mode called assisted explicit Dynamic Tasks Allocation. In this mode, the initial allocation is performed by the automated system, but the human operator can modify it if he/she disagrees with the proposal for any reason [14]. We have implemented on a software platform that integrates a realistic air traffic simulator, and evaluated with professional controllers, different Dynamic Tasks Allocation modes in the ATC domain. 2.3 Air Traffic Control The French ATC is a public service in charge of the flight security, the regulation of air traffic and the flight economy for each aircraft that crosses the French airspace. The security is of course the main objective and consists in detecting and preventing the collision between the aircraft. These situations are called conflicts and controllers must avoid them. Our studies concern the "en route" control more particularly, in which the air space is divided into several sectors. Each of them are supervised by two controllers who have the same qualification, Figure 2. The first controller has a tactical role and is called radar controller (RC). He/she must supervise the traffic in order to detect conflict between aircraft and then resolves them by modifying the initial trajectory of one or several aircraft by sending a verbal instruction (heading, flight level, etc.) to pilots. The second controller
S. Debernard et al /AMANDA Project
177
has a strategic role and is called planning controller (PC). This role consists in avoiding an overload of the tactical level, i.e. an overload of the RC. So the PC must predict the future traffic that the RC will manage, and negotiate the enter and exit aircraft flight level with the adjacent sector PC, in order that RC will not have too much conflict to resolve. So, PC performs "traffic filtering", which has nevertheless some limits because it is impossible to increase the number of aircraft thoughtlessly, and changing aircraft flight level is not a economical solution [14]. Entering traffic filtering trategical level - Coordination between sectors Planning controller Exiting traffic preparation
S
AIR-TRAFFIC CONTROL level Traffic supervision Aircraft guidance Radar controller Tactical
Data integration Figure 2: Main tasks of Air Traffic Controllers So as to face the air traffic increase, one solution consists in giving an active assistance to RC in order to increase the tactical level capacity. Now we present our previous studies performed in collaboration with CENA (French acronym for French Air Navigation Study Center). These studies aim at integrating a Dynamic Tasks Allocation in order to keep the controllers "in the loop". 2.4 SPECTRA VI PROJECT In a first study called SPECTRA V1 [15] [16], a Dynamic Tasks Allocation has been implemented at the tactical level between the RC and an assistance tool called SAINTEX [17]. SAINTEX is able to resolve simple conflicts between only two aircraft (without interfering aircraft in the same space). These conflicts are called shareable conflicts because the two decision-makers are able to resolve them in an autonomous way. SAINTEX has its own strategy and performs completely the resolution task, from the detection to the implementation of instructions, including rerouting. Two modes have been evaluated: an explicit and an implicit mode. The former organization concerns a pre-emptive explicit task allocation. In this mode, the human air traffic controller manages the task allocator through a dialogue interface (Radar Image and Electronic Stripping Interface). He/she is the estimator of his/her own performance and workload, and he/she allocates tasks either to himself or to SAINTEX. Shareable conflicts are indicated with a specific color on the operator dialogue screen and radar image. The deadline for the conflict to become unsolvable by SAINTEX as well as the proposed deviation order are displayed on the strip of the aircraft. Initially, all the aircraft are allocated to the controller. If this one feels overloaded, he/she can select a conflict and transfer it to SAINTEX. In pre-emptive allocation, the controller can modify the allocation of a shareable conflict. But at that time, if the order deadline given by SAINTEX is over, this conflict cannot be solved by SAINTEX anymore. The later organization is a non-pre-emptive implicit task allocation controlled by an automatic management system implemented on a calculator. This allocation depends on the intrinsic abilities of the two decision-makers. For SAINTEX, those abilities are functional ones: It can only solve the easiest conflicts. For the human radar controller, these abilities are related to his/her workload. For the moment, only the tasks demands are assessed in real-time. So, when those demands are too high and exceed a maximum level, the shareable tasks are allocated to SAINTEX. In this case, the conflict with the nearest order deadline is
178
S. Debernard et al. /AMANDA Project
chosen and transferred to SAINTEX which displayed its solving strategy on the strips. In the contrary, the conflict is allocated to the human controller. The scenarios have been created with heavy air traffic to test the usefulness of dynamic allocation. They generated about twenty conflicts per hour of different nature. These experiments have been performed with nine male qualified air traffic controllers. TLX method has been used to calculate the global workload for each controller and for each experiment. The main result of these experiments [15] is that a dynamic task allocation mode (either implicit or explicit) seems to improve the air traffic control task and to reduce the global workload. According to two criteria related to the performance, the economic and security criterion, the implicit mode seems to be the most efficient one. In spite of these interesting results, SPECTRA V1 points out the problem of the acceptance of the assistance by controllers. Because they have got a limited trust in SAINTEX, radar controllers prefer the explicit mode which allows them to take back conflicts allocated to SAINTEX. With this mode, solutions are not imposed to controller so he/she is not only a supervisor. But the problem of the explicit mode is controller's workload which could be increase because radar controller has to insure the additional task of allocation. Thus, radar controller takes a strategic role by detecting and allocating conflicts. These conclusions has led our research to examine the air traffic control organization, to see how we could make to introduce the strategic level in the human-machine system. 2.5 SPECTRA V2 Project In a second study called SPECTRA V2 [18], a Dynamic Tasks Allocation has been also implemented at the tactical level between the RC and SAINTEX. The goal of this project was to keep advantages of implicit allocation - to avoid the workload due to the allocation management - but with the possibility for controllers to take back a shareable conflict. Because the allocation management is a strategic task, this function has been given to the planning controller (PC). So, at the strategic level, a new assistance tool called PLAF (French acronym for Planning of Allocation) has been implemented. PLAF helps the PC to anticipate the influence of the entering aircraft on the existing traffic, and to estimate the future workload of the radar controller. When an overload of the tactical level is detected, PLAF can propose to allocate one or several shareable conflicts to SAINTEX. To justify these propositions of allocation, a dedicated screen displays the tasks detected by PLAF, the conflicts detected by SAINTEX and the period of overload. All these information are presented according to two axes: the time (about 30 minutes of prediction) and the flight levels. Two modes have been evaluated: an explicit and an assisted explicit mode. In the explicit condition, allocation can be controlled either by planning controller or radar controller. Shareable conflicts are allocated, by default, to RC. They can be allocated to SAINTEX before a deadline defined by temporal constraints. In addition to SAINTEX, a traffic and conflict prediction is performed by PLAF. In the assisted explicit condition, PLAF automatically proposes conflict allocations, on the basis of workload prediction. When it allocates a conflict, PC can change the allocation, but RC cannot directly control the allocation, apart from a request to PC. The results of an activity analysis are in favor of the assisted explicit [19][20]. The performance is better, and the cooperation between the agents seems to be more operational. The assistance tools seem to bring the controllers' activities to be more "planful" and less reactive to events. Moreover, the cooperation is more concerned with goals and less concerned with communication. The PC participates more in the decision making by decision suggestion to the RC. It would be the effect of the allocation mode which prescribes a precise allocation of tasks, and the involvement of an agent (PC) to facilitate the cooperation between one artificial agent (SAINTEX) and one human agent
S. Debernard et al. /AMANDA Project
179
(RC). All these results may be explained by the capacity of assistance tools to reduce cognitive workload. An other main point of these results is the information carrying by graphical displays, which constitute a reduced Common Work Space, where SAINTEX and PLAF give their results about the traffic analysis, and where radar controller and planning controller could display symbols to mark important information, especially their conflicts detection. Because the controllers are used to work together, and because they display a subset of their understanding of the situation on the reduced Common Work Space, a controller doesn't need to communicate all the time with the other for taking a decision. But this reduced Common Work Space is not sufficient because it doesn't integrate all the "results" of the controllers' activities, such as information elaboration, diagnosis, decision making, etc. Nevertheless, the SAINTEX' strategy can hamper controllers because they don't share the same representation of a conflict - SAINTEX has its own strategy and RC can take into account another aircraft in a conflict for resolving it in accordance with his/her abilities and workload. Furthermore, in the assisted explicit mode, a complacency phenomenon led RCs to relax their monitoring of the overall traffic, because they were less implicated in the allocation than the explicit mode. The following part presents a new project where a "real" Common Work Space has been defined and implemented for resolving these problems. 3 From dynamic tasks allocation to tasks delegation 3.1 Main Goals of AMANDA Project The AMANDA project (Automation & MAN-machine Delegation of Action) concerns the study, the implementation and the evaluation of an support tool more cooperative than SAINTEX [21]. The human-machine cooperation mode proposed here consists in a delegation of tasks to the support tool. The difference we made between dynamic tasks allocation and tasks delegation is that for the former, the support tool is able to perform the shareable task entirely, whereas in the latter, the support tool performs the task from a strategy given by the human operator. From a strategy, the support tool will try to calculate the solution for the task and will inform human operator. Then, this one could explicitly delegate the task to the support will perform the monitoring and implementation function. The goal of this human-machine cooperation is the same as Dynamic Tasks Allocation: to reduce the human load and to allow human operator, by giving him some time, to focus on complex situation that the support tool can not take into account. But, for an efficient assistance to the human operator, the strategy should be introduced in the support tool enough early. This approach sets us a fundamental difficulty because, for an air traffic controller, a problem (or situation) is not always well defined for giving to the support tool a strategy. In fact, before a human operator can produce a final decision (an operational decision which can be applied on the process), this decision will go from a schematic decision (a strategy in a embryonic state) to the final decision, in accordance with several step of refinement, Figure 3. This refinement is due to the progressive introduction of constraints that will reduce the freedom degrees of the decision (the process situation is more and more precise). In the case of Air traffic control, this refinement is due also to the fact that controllers take into account the situation's uncertainties, as the meteorology for example.
180
S. Dehernard el al. / AMANDA Project
Figure 3: Refinement steps of a human decision
So, the cooperation between human operators and the support tool must be designed in order to take into account several kinds of human decision for delegating a task to the support system, but also by defining the interaction between the two controllers (planning and radar controllers) and the support system in order to take into account the work rules. For predicting the workload of the radar controller (in order to avoid a future overload), the planning controller must detect the future problems in the sector. When he/she detects for example a conflict with two or more aircraft, he/she informs the radar controller. Then, this controller resolves it. In fact, the planning controller "prepares" the traffic for the radar controller, and we think it is possible to give some means to the controllers for a better human-human cooperation. Moreover, if the radar controller has a lot of problems to solve, it will be possible that the planning controller start to introduce some strategies in the support tool. The Figure 4 shows a possible linking of the two controllers' activities.
Figure 4: Refinement steps of a human decision for the controllers
The dialogue support - the human-machine interface - must be well studied, firstly for allowing an easy introduction in the support tool of the definition of the problems and the strategies, and secondly for improving the cooperation activities between the agents. So, the AMANDA study concerns: - The design of the support tool in terms of abilities.
S. Debernard et al. /AMANDA Project
181
- The design of the humans-machine interface which will support the interaction between the two controllers and between the controllers and the support tool. In order to define the characteristics of the support tool and the HMI, an experiment has been realized with professional controllers from CCR (Regional Control Center) of Bordeaux in France. This experiment is called AMANDA V1. 3.2 An Experiment for Defining Human-Machine Cooperation in Air Traffic Control The main goals of AMANDA VI experiment were [21]: 1- To check if controllers will be able to use the support system: we did not want to evaluate the controllers abilities, but to check the adequacy between the support tool use and the controllers' activities. The main questions were "Are controllers able to anticipate their decisions for delegating them to the support tool? Are these decisions enough stable?". 2- To prepare the specifications of the future human-machine interface between the two controllers and the support tool: it is necessary to establish the dialogue modalities between these agents (for example, how a strategy is given by controllers). An experimental platform has been developed for this experiment. This platform integrates a realistic air traffic simulator, but does not integrate any support tool. Each controller performs his/her tasks with a human-machine interface composed of two 21" displays: on the former, a radar image is displayed in real time, and on the latter an electronic stripping can be used. This electronic stripping is common for all the controllers, i.e. each modification on one stripping affects the other. The chosen sector, called C1-C2, includes several points of conflict and the aircraft are often stable. The traffic scenarios were realistic and have been checked by a professional controller. Two kinds of experiment have been performed on this platform. The first one has been designed for the first goal of AMANDA V1 and has been called "anticipation". The sector was managed by two controllers: the planning controller and the radar controller. The experimental method has consisted in breaking periodically the simulation and asking, to the radar controller, to anticipate their decisions for managing the traffic. During the experiments, all the data coming from the platform have been recorded as the verbal communication between controllers in order to perform a cognitive analysis of activities. 7 pairs of professional controllers have been realized this experiment. The second kind of experiment has been called "cooperation". In order to determine the dialogue form between the two controllers and the future support tool, one planning controller and two radar controllers have been managed together the same traffic. But, a static allocation of the aircraft between the two radar controllers has been made and one radar controller could not send any instruction to an aircraft allocated to the other controller. The allocation choice has been realized in order that controllers must cooperate on the same conflict. Six teams of three controllers (2 radar and 1 planning controller) have been realized this experiment, and as the "anticipation" experiment, a cognitive analysis has been performed. The results of the « cooperation » experiment are summarized here and concern only cooperative activities [22]: - Two kinds of cooperative activities have been find: the "cooperation in action" represents only 20% of the cooperative activities, and have some implications in real time; the "cooperation in planing" represents 80% of the cooperative activities and consists in maintaining a common frame of references [23]. This Common Frame of References is a common representation of the controlled environment and also a common representation of the team's resources. So, it includes the common goals, the common plans, the role allocation, and the representations of the process, etc. The cooperative activities that maintain the common frame of references, allow to direct
182
S. Debernard et al. /AMANDA Project
activities in order to minimize decisional conflict between controllers and to avoid reactive activities. The importance of this global level of coordination between agents leads us to think that a support tool must communicate with human operators at the same level by the use of a common frame of references. - For cooperation in action, 60% of these activities was dedicated to mutual control and anticipation of interference, rather than detection and resolution of interference. - For "cooperation in planning", the most of cooperative activities was dedicated to the maintenance of the common frame of references. This maintenance was performed by simple exchange of information and agreement, rather than long explanation. So it is not necessary to implement some functions in the support tool that allow this explanation. Moreover, the controllers will not have enough time for that, due to the time pressure of the traffic. - Generally, the common frame of references concerned some aircraft set rather than isolated one. So, the presentation of information coming from the support tool and controllers, must allow them to communicate at this level, rather than on separated supports as the strips normally used in air traffic control. The «anticipatin » experiment [24] has given two main results that allow to characterize the functions of the support tool: - For obtaining an efficient human-machine cooperation, the support tool must be integrated in the controller's space-problem in order that the information's representation are the same for all the decision-makers. The support tool must allow too controllers to build as they want this space-problem even if the support tool can detect interference on this representation of a situation. For example, when a controller will introduce a strategy in the support tool, this one can detect an another aircraft that will interfere with the solution. So the support tool must enrich the spaceproblem. - The support tool must allow controllers to focus on a particularly problem. But for avoiding that controllers lost their perception of the global situation, it is better that the presentation of a problem is displayed in the same format as the radar image with all the aircraft. Nevertheless, the aircraft included in a problem will be displayed with highlighted color while the other aircraft will be displayed with gray color. The AMANDA V1 experiments have allow us to characterize the functions of the future support tool for delegating tasks to it, and the interaction between controllers and this support tool which will be supported by the human-machine interface. These experiments have shown that when human operators must cooperate, they build and maintain a common frame of references. For obtaining an efficiency human-machine cooperation, an implementation of this common frame of references seems necessary. The next part of this paper describes the functions of the support tool, called STAR, and the contents of the implementation of the common frame of references that we call Common Work Space. 3.3 Concepts of AMANDA The new platform called AMANDA V2, includes two main modules: STAR (French acronym for resolution assistance system), the support tool for delegating task, and the Common Work Space. STAR must perform several functions and must also interact with the controllers. In order to characterize the content of the Common Work Space, it is necessary to study these interaction between these functions and the human activities. Many disciplines have already worked about the CWS idea [23][25][26]. For example, in the case of Air Traffic Control, Bentley & al. [25] have presented a shared work space that provides an adapted presentation of air traffic to different users, on different machines, to make the use of shared entities easier. Decortis and Pavard [23] have
S. Debernard et al. /AMANDA Project
183
defined the shared cognitive environment as a set of facts and hypotheses that are a subset of each agent's cognitive universe. For defining a common work space, it is possible to start from a model of human activities. The Rasmussens's model [27] is composed of four main activities, the information elaboration, the identification of a situation, the decision making, and the implementation of a solution. These classes of activity will be decomposed into sub-classes describing the evolution of the state of an information. The human agent plans her/his own activity in order to manage her/his internal resource according to the time pressure and the stress she/he is affected by [28]. For managing situations, the human agents build a frame of reference which contains different attributes [29]: information (stemming from activities of information elaboration); problems (stemming from activities of identification); strategies (stemming from activities of schematic decision making); solutions (stemming from activities of precise decision making); commands (stemming from activities of implementation of solutions), Figure 5.
Figure 5: Contents of a Common Frame of References
Figure 6: Cooperation between agents through a Common Work Space
To cooperate, human agents pass on information, problems, strategies, etc., for sharing their own frame of reference. In the case of Human Machine Cooperation, the Common Work Space can be the support of cooperation between the human agents and the artificial agents. But it is necessary to define which activities/functions will be performed by the agents and which form of cooperation must be implemented. It is possible to use here the 3 forms of Schmidt for characterizing the cooperative activities around the common work space: augmentative, integrative and debative [30]. In the debative form (Figure 6-a), all the agents supply the CWS with new data (for one task and for one activity/function), and when some interference appear, they can negotiate. In the integrative form (Figure 6-b), only one agent supplies a data of the CWS, and the other takes into account this data for performing the next activity/function. In the augmentative form (Figure 6-c), the agents perform the same activity/function and update the CWS, but for different tasks, in accordance to the allocation of the different tasks. It is clear that the design of a cooperative human-machine system requires to define the effective content of the CWS which depends on the domain and the tasks to be achieved. More over, the use of a CWS generates some new activities from human agent (but also some functions from an assistance tool). These activities (or function for a machine) must allow [29]:
184
S. Debernard el al. /AMANDA Project
-
To update the common frame of reference: correspond to the activities/functions for updating each common frame of reference's attributes from their frame of reference. - To control the common frame of reference: the activities/functions aim to compare the common frame of reference and their frame of reference. These activities correspond to the mutual control and allow the interference detection on one or more attributes (information elaboration activities but on the CWS data, not on the process information). - To manage the interference: firstly, the activity/function corresponds to a diagnosis on the differences between the CWS and the frame of reference of an agent. Secondly, it is necessary to solve the interference (diagnosis, decision making, implementation activities). Three forms of solving may be used by the agent: negotiation, acceptance, imposition. These forms imply for human agent cognitive and communication costs which are different. The negotiation aims at reducing the differences between the CWS and the frame of reference of an agent by modifying one of them, on the basis of explanations between the agents. The acceptance is the update of the frame of reference from the CWS. This acceptance is chosen when the cost of a negotiation is too important or when an agent wants to facilitate the activities of the other. The imposition corresponds to the opposite of the acceptance. The implementation of a Common Work Space between a support tool and one or several human agents brings some constraints, in particular for implying a negotiation. Two human agents, when they negotiate, may use symbolic explanations which are very efficient. An artificial agent needs an explicit explanation on the base of operational information. So, at present, it is very difficult to implement in an artificial agent, some capacities for having a real negotiation with human agents. An other problem will also appear when the human agent has to accept a solution given by an artificial agent if she/he can't modify this solution. For our project, the contents of the common work space have been defined from AMANDA V1 experiments. On the Table 1, each line refers to a human activity, and for each activity are shown: - The generic information produced by the activity/function - The effective contents of the CWS in accordance with the ATC domain. A problem is a set of conflicting aircraft and is called "cluster". A cluster is composed of at least two aircraft in a duel situation (binary conflict) and other aircraft that can interfere with this duel. An aircraft interfere with a duel if the resolution of these initial duel generates an other duel. The definition of a cluster (the aircraft included in this cluster) will be performed by controllers, because it can depend on controllers' strategy. Table 1: The Common Work Space of AMANDA V2
Human Activities Elaboration of information Detection Schematic decision making Precise decision making Implementation of solutions
Information Contents of CWS Initial information none Cluster Problems Strategies for resolving a Directives problem Delayed instruction Solutions Solutions implemented Instruction
A strategy is modeled as one or several "directives" for resolving a problem. For example, a directive can be "turn AFR365 behind AAL347". So a directive doesn't indicate a place and a value as it is the case for an normal instruction. The definition of a directive will be performed by controllers, in accordance with the delegation principle. A solution is a delayed instruction; for example, "turn AFR365 30° to the left at 22:35".
S. Debernard et at. /AMANDA Project
185
From the contents of the common work space, the function of STAR and the interaction between STAR and the two controllers may be specified, Figure 7.
Figure 7: Functions of STAR and interactions with controllers
From a directive given by controllers, STAR is able to calculate a new trajectory for an aircraft in order to resolve a duel. This new trajectory is composed of one or several delayed instructions. Nevertheless, STAR, with the new trajectory, can detect a new duel with interfering aircraft. The new trajectory is entering in the common work space (but it is not delegable) and the interfering aircraft is added in the cluster for preventing the controllers. This situation can appear if the controllers have not see the problem (the interfering aircraft), or more certainly if the controllers have not yet added all the directives (the strategy) for resolving the conflict. From a delayed instruction given by controllers or calculated by STAR, this latter is able to monitor and to implement this delayed instruction, if it is delegated to STAR. But before implementing an instruction, STAR verifies all the time, if the delayed instruction doesn't generate any duel. In this case, the interfering aircraft is added in the cluster. From the point of view of human-machine cooperation, the project AMANDA allows the three modes of cooperation defined by Schmidt [8]: - Augmentative mode: STAR is able to perform some parts of the resolution tasks when they are delegated to it, more particularly the implementation function which requires a monitoring function. So controllers and STAR perform the same activity/function but for different problems. - Integrative mode: STAR take into account the strategy and/or the solution given by controllers for performing the functions necessary to resolve completely a problem. - Debative mode: STAR is able to modify the controllers' perception of a problem when STAR detects some interfering aircraft that are not defined in a cluster. In the other way, the controllers can take into account the solutions calculated by STAR, but they can also integrate into the common work space another directives or constraints if they are not agree with this solution So, the CWS allows controllers to negotiate with STAR, because it is possible to add in a cluster, some new directives. It is also possible to add some geographical constraints on a
186
5. Debernard el al. / AMANDA Project
directive, or to indicate to STAR to not take into account an interfering aircraft for calculating a new trajectory, but of course the controllers must manage alone this aircraft. For evaluating these concepts, a new platform called AMANDA V2 has been developed. We present now this platform. 3.4 The AMANDA V2 Platform 3.4.1 Architecture of AMANDA In this part, we present briefly the platform. It is composed of several workstations linked by a local network, Figure 8. These workstations communicate under the protocol TCP/IP. All the platform was developed with the ADA language, and the interfaces use the protocol X-l1.
Figure 8: Architecture of AMANDA V2
The first workstation is the air traffic simulator. The second one contains the module STAR and manages the common work space. The third workstation allows to backup all the information that circulate on the network in order to perform a replay of an experiment for analyzing it. The two last workstations manage the human-machine interface, and each controller (radar and planning controller) has he/she own interface. This interface is composed of two screens. On the first one, a radar image is presented. The other is used for presenting the CWS. This one is composed of two views which are switchable by controllers. On the first view are presented all the clusters defined by controllers. On each cluster are displayed the aircraft, but also the directives given by controllers and the delayed instructions calculated by STAR. When a controller selects one cluster, a resolution view is displayed, Figure 9. This view allows controllers to add information in the CWS but for the concerned cluster, as aircraft, directives and delayed instruction. This view displays also a filtered radar image which takes into account the aircraft included in the cluster. For each aircraft, the predicted trajectory is displayed in accordance with all the delayed instructions given by controllers or calculated by STAR.
187
S. Debernard et al. /AMANDA Project
Figure 9: Common Work Space interface on screen 2 : Resolution View
3.4.2 Principle of STAR General principle STAR is a set of interactive decision support systems which support three main activities of the controllers for the conflicts resolution between aircraft. These activities are the procedure formulation, the actions execution, and the situation monitoring inside the sector (Figure 10). All these assistance work according to the principle of the tasks delegation. There are different possibilities of delegation according to the level of assistance. The controllers can delegate the complete resolution of a problem or they can delegate only the execution of simple actions. In the first case, the controllers give a complete strategy (under the form of one or several directives) to the actions calculation support tool that calculates new trajectories (under the form of delayed instruction). In the second case, a simple action (a delayed instruction) is given to the assistance that will send it later to the pilot. Assistance to procedures calculation Detection
of an abnormal situation
1i -
Initial
trajectories Procedure
Strategy (Directive)
formulation
Assistance to actions execution and for situation monitoring
Planified trajectories Procedure execution Actions Delayed instructions,
Instructions
Figure 10: Situation of assistance in the controllers' cognitive process
188
5. Debernard et al. /AMANDA Project
directive which includes the aircraft involved in the conflict, the conflict resolution mode, some constraints on the beginning and the end of the resolution, etc. All the resolution modes included in STAR are issued of the experimentation AMANDA V1, to fit with the controllers' resolution strategies. The constraints contained in a directive reduce the space of possible solutions but it stills remains a lot of solutions. There are two steps to reduce this solution space to one solution (Figure 11).
Figure 11: Principle of the targeted exploration
The first step consists in a partial exploration of the solution space (targeted exploration). To explore this space, the support tool calculates a set of possible trajectories. To limit the number of solutions, the support tool calculates trajectories with a time step of separation between themselves. The last step consists in a multi-criteria choice which select the most relevant trajectory regarding to a set of specific criteria. This step is based on the multiple criteria methodology for decision aid [31]. This methodology rests on four level structuring respectively the parameters of the decisions, the consequences on the decision, the preferences of the decisionmakers and the procedure elaborating the decision given to the decision maker. The first level identifies the decision process and the conceivable alternatives called potential actions. These ones are the trajectories previously defined to solve an air-traffic control situation. The set of these potential actions is discrete and perfectly identified. The second level identifies the elements used to evaluate the potential actions: the criteria. The global objectives used by the air-navigation authorities are the security, the economy, the regularity and the efficiency [32]. Several criteria translate these objectives. The minimal separation distance between two planes represents correctly the security objective. The distance of a trajectory suffices to express the economy objective. The regularity is directly lied to the time to transit. The efficiency translates the quality of a trajectory for the controller. Then three criteria translate these quality for a trajectory: the number of deviation, the number of instruction and the greatest deviation. The third level aims at defining the preference between the actions for each criterion and globally for all the criteria. For each criteria, the smallest value is preferred. Moreover the criteria are formed into a hierarchy in the decreasing importance as follow: the number of deviations, the greatest deviation, the number of instruction, the distance of the trajectory, the time to transit and the separation distance. Then the procedure to choose one trajectory is based on a lexicographic order.
S. Debernard et al. /AMANDA Project
189
The support tool for monitoring and implementing instructions These assistance are very simple but they can perform three essential phases: - The monitoring of the conflicts: The uncertainty of the aircraft evolutions can call into question the reliability of a previously calculated trajectory. The controllers must be informed if the calculated trajectory is changed. - The application of instructions: A trajectory is composed of several segments. Each segment corresponds to an instruction which indicates the maneuver to do. The application of a trajectory consists to perform the transmission of all the instructions at the good time. The follow-up: This is the monitoring of the instructions application to the aircraft. An instruction can be forgotten by a pilot or the aircraft can deviate of the previous way because of meteorological conditions. 4
Conclusion
In this paper we presented the AMANDA project. This project aims at evaluating the task delegation principle. This project comes from several studies on application of humanmachine cooperation principle, especially dynamic task allocation, to air-traffic control. Assistance proposed to controllers consists in finalize high level decision of controllers. From a strategy expressed by the controllers, the system defines satisfying sequences of actions and then selects one sequence respecting the controllers preferences. Such an assistance necessitates to introduce new specific information. These information represent some pieces of decision usually manipulated by controllers but never practically expressed. Then the AMANDA project comprises also a human-machine interface issue. Soon the AMANDA project will enter into the experimental stage. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]
[11]
Garland DJ. (1991). "Automated Systems : the human factor". NATO ASI series. Vol. F73. Automation and Systems Issues in Air Traffic Control. pp 209-215. Hollnagel, E., Bye, A. (2000). "Principles for modeling function allocation". Int. J. Human-Computer Studies. 52, pp. 253-265 Sheridan, T.B., (2000). "Function allocation: algorithm, alchemy or apostasy?". Int. J. Human-Computer Studies. 52, pp. 203-216 Boudes, N. (1992). La gestion cognitive du temps dans les systemes socio-techniques de controle de processus. These de doctorat. Universite Toulouse Le Mirail. France. 26 octobre 1992. Hoc J-M. (2001). "Towards a cognitive approach to human-machine cooperation in dynamic situations". Int. J. Human-Computer Studies 54. In press. Millot P., Lemoine M-P. (1998). "An Attempt for generic concepts toward Human-Machine Cooperation". Paper presented at IEEE SMC'98, San Diego, USA, October Royer, V. (1994). « Partage de croyances : condition necessaire pour un systeme coope'ratif ? ». In Pavard B. (Ed.), Systemes Cooperatifs de la modelisation a la conception. Toulouse, France, Octares. pp. 253270 Schmidt, K. (1991). Cooperative work: a conceptual framework. In J. Rasmussen, B. Brehmer, & J. Leplat (Eds.), distributed decision-making : cognitive models for cooperative work. (pp 75-110). Chichester, UK : John Willey and Sons. Grislin, E., Millot, P. (1999). Specifying artificial cooperative agents through a synthesis of several models of cooperation. Seven European Conference on Cognitive Science Approaches to Process Control (pp 73-78), Villeneuve d'Ascq, 21-24 Septembre 1999, France. Debernard S., Hoc J-M., (2001). "Designing Dynamic Human-Machine Task Allocation in Air Traffic Control: Lessons Drawn From a Multidisciplinary Collaboration". In M.J. Smith, G. Salvendy, D. Harris, R. Koubek (Ed.), Usability evaluation and Interface design: Cognitive Engineering, Intelligent Agents and Virtual Reality, volume 1. (pp. 1440-1444). London: Lawrence Erlbaum Associate Publishers. Debernard S., Vanderhaegen F., Debbache N., Millot P, (1990). "Dynamic task allocation between controller and AI systems in air-traffic control". 9th European Annual Conference on Human Decision Making and Manual Control. Ispra, Italy.
190
[12] [13] [14] [15]
[16] [17] [18] [ 19]
[20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31 ] [32]
5. Debernard et al. /AMANDA Project
Crevits I., Debernard S., Denecker P. (2002). « Model building for air-traffic controllers workload regulation ». European Journal of Operational Research, Vol. 136/2. Pp. 324-332. To be Published in 2002 Millot, P. (1988) Supervision des procedes automatises et ergonomie. Edition Hermes, Paris. Crevits, I., Debernard, S., Vanderhaegen, F., Millot, P. (1993). Multi-level cooperation in air traffic control. Proc. Fourth International Conference on Human-Machine Interaction and Artificial Intelligence in AeroSpace. Toulouse, France, September 28-30. Debernard S., Vanderhaegen F., Millot P., (1992). "An experimental investigation of dynamic task allocation between air traffic controller and AI system". 5th IFAC/IFIP/IFORS/IEA Symposium on Analysus, Design and Evaluation of Man-Machine Systems. The Hague, The Netherlands. June 9-11, 1992. Debernard S., (1993). « Contribution a la repartition dynamique de taches entre operateur humain et systeme automatise : Application au controle de trafic aerien ».These de doctorat. Universite de Valenciennes. France. 28 Janvier 1993 Angerand, L., & Le Jeannic, H. (1992) Bilan du projet SAINTEX, Note R92009. Centre d"Etudes de la Navigation Aerienne. Crevits I., (1996). « Repartition dynamique de taches dans les procedes complexes. Modelisation de la repartition anticipee : application au contro1e de trafic aerien ».These de doctorat. Universite de Valenciennes. France 28 Janvier 19%. Lemoine, M.-P, Debernard, S., Crevits, I., Millot, P. (1996) Cooperation between humans and machines : first results of an experimentation of a multi-level cooperative organisation in air traffic control. Computer Supported Cooperative Work: The journal of Collaborative Computing. Vol. 5, N°2-3, pp 299321. Hoc, J.M., Lemoine, M.P. (1998) Cognitive evaluation of human-human and human-machine cooperation modes in Air Traffic Control. The International Journal of Aviation Psychology. Vol. 8, N°l, pp. 1-32. Debernard S., Crevits I., Carlier X., Denecker P., Hoc J-M. (1998). Projet AMANDA : Rapport final de la premiere phase. N° CENA/NR98-185. Convention de Recherche LAMIH/CENA 96/C007. Septembre 1998 Carlier, X. & Hoc, J.M. (1999). Role of a common frame of reference in cognitive cooperation: snaring tasks in Air-Trafic-Control. Communication presente a CSAPC'99. Villeneuve d'Ascq, France, Sept. Decortis, F., Pavard, B. (1994) Communication et cooperation : de la theorie des actes de langage a I'approche ethnomethodologique. In Pavard B. (Ed.), Sysiemes Cooperatifs de la modelisation a la conception. Toulouse, France, Octares Denecker, P., Carlier, X., Morineau, T., & Hoc, J.M. (1999). Projet AMANDA - Note intermediaire 2.1 (Rapport technique). Athis-Mons, F: CENA. Bentley, R., Rodden, T., Sawyer, P., Sommerville, I. (1992) An architecture for tailoring cooperative multi-user display. Proceedings of CSCW'92. November Jones, P., Jasek, C. (1997) Intelligent Support for Activity Management (ISAM): An Architecture to Support Distributed Supervisory Control. IEEE Systems, Man and Cybertenics, Part A: Systems and Humans, Vol. 27, N° 3, May Rasmussen, J. (1983). Skill, rules, and knowledge ; signals, signs and symbols, and other distinctions in human performance models. IEEE Transaction on Systems, Man, and Cybernetics (pp.257-266), SMC15,N°3. Hoc, J.M. (1996) Supervision et controle de processus: la cognition en situation dynamique. Presse Universitaire de Grenoble, Collection Sciences & Technologie de la connaissance. Pacaux-Lemoine, M.-P., Debernard, S., (2001). "Common work space for human-machine cooperation in air traffic control". Control Engineering Practice. Accepted the 5th January 2001. Lemoine, M.P. (1998) Cooperation Homines-Machines dans les procedes complexes : modeles techniques et cognitifs pour le Contrdle de Trafic Aerien. Memoire de Doctorat Specialite Automatique des Systemes Industriels et Humains. Universite de Valenciennes, France Roy B., (1985), Methodologie multicritere d'aide a la decision, Economica, Paris, 1985. Colin de Verdiere D., (1996), Quel optimum pour la gestion du trafic aerien, Centre d'Etude de la Navigation Aerienne, Not NT96575.
Cooperative Systems Design M. Blay-Fornarino et al- (Eds.) IOS Press, 2002
191
Recursive articulation work in Ariadne: the alignment of meanings Marcello Sarini1 and Carla Simone2 'Dept. of Computer Science, University of Torino, Corso Svizzera 185, 1-10148 Torino, Italy, e-mail:
[email protected] 2 Department of Informatics, System and Communication (DISCO), University of Milano-Bicocca, Via Bicocca degli Arcimboldi 8, 20126 Milano, Italy. e-mail: carla.simone@disco. unimib. it Abstract. Articulation work is a recursive phenomenon. In fact, the management of the articulation of cooperative activities is conducted as a cooperative effort which, in turn, needs to be articulated. We call this type of articulation recursive articulation. Recursive articulation may involve all aspects of coordination. They concern the local cultures and skills as well as the local conventions and procedures that are objectified in local coordination mechanisms. In this paper we want to address the recursive articulation that involves local cultures and skills and the related conventions in order to achieve the alignment of local points of view for cooperation to occur (alignment of meanings). From empirical studies we derived the idea that the alignment of meanings can be obtained by the combination of two phases: the first one, where users are active in (partially) solving the dis-alignment of meanings and a second one where this effort is paid back since the supportive technology alleviates the communication problems. On this basis, we implemented a specialized module, called Reconciler, which is a technology to interactively build the correspondences to solve conflicts related to concepts used in communication during cooperation between communities.
1
Background arid motivations
In [1] we discussed the conceptual foundation underlying Ariadne, the notation we have developed for the construction of tools supporting articulation work. This conceptualization is based on the analytical distinction between cooperative activities and articulation work. The relationship between them can be described as follows: articulation work is the additional effort needed in cooperative settings in order to align, coordinate, mesh, etc. the distributed cooperative activities characterizing the specific field of work. The notion of Coordination Mechanism (CM) was proposed to describe the specific means actors use to manage the complexity of articulation work in cooperative work settings characterized by complex tasks interdependencies. The notion of Coordination Mechanism incorporated in Ariadne is in particular useful to capture the distributed nature of articulation work: in fact, Ariadne enforces the view that CMs have to be identified by looking at the behavior of local communities, whose members are tied together by quite well understood interdependencies, and that the larger organization can be viewed as the composition of such communities cooperating to achieve the organizational goals. In addition, the interacting communities cooperate with the support of the Compound CMs constructed, at any level of aggregation, through the composition of the local ones. In particular, Ariadne allows (of course, without
192
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
imposing) the bottom-up construction of the technological support of articulation, at any level of composition. This capability implements the type of adaptability that refers to local cultures and skills as well as local conventions and procedures [2]. While locality fits the need to manage adaptability in terms of modification of the CMs respecting the autonomy of the involved communities, at the same time it makes more difficult the maintenance of the mutual consistency of the involved CMs in relation to the modifications. This tension is constitutive of the distributed nature of cooperative work and cannot be avoided. The only possibility is to provide support to manage this complexity in a more effective way. among the others means, through a suitable technology. To this aim it is fundamental to consider that articulation work is a recursive phenomenon in that the management of the articulation of cooperative activities (i.e., its definition and maintenance) is conducted as a cooperative effort which, in turn, needs to be articulated [1]. We call this type of articulation recursive articulation. The above considerations have at least two consequences. On the one hand, recursive articulation is part of the cooperative effort and has to be explicitly considered as well. On the other hand, the notion of CM can be applied again to the design of a technology supporting recursive articulation. This point of view is not so common in CSCW. All attempts to support adaptability of CSCW systems envisage an activity devoted to the management of changes: this latter however is commonly seen more as a specific functionality of the CSCW systems under concern than as a special type of articulation work. In other words, the focus is on the implementation of the changes not on the process defining them. The cooperative nature of this effort is usually overlooked. This is exactly what we aim at in our research effort. Recursive articulation may involve all aspects of coordination (see Figure 1). As recalled above, these aspects concern the local cultures and skills as well as the local conventions and procedures that are objectified in the local CMs.
Figure 1: Recursive Articulation
In a previous work [3,4] we considered the recursive articulation required by the management of conventions and procedures defining the flow of control of distributed cooperative activities (that is, the recognized protocols in Ariadne's terminology). The proposed approach consisted in a specialized CM supporting the negotiation and propagation of changes of the communicative behavior of the underlying CMs. Distributedness was considered very seriously by taking into account that the actors involved in the negotiation of changes observe the whole working arrangement from different perspectives, and therefore have partial and dis-homogeneous views of the communicative behavior of the actors belonging to different local communities. The CM
M. Sarini and C. Simone /Recursive Articulation Work in Ariadne
193
supporting this type of recursive articulation was equipped with functionalities governing the impact of local changes on the whole system behavior as well as protocols helping people to make the negotiation converge to a common solution. In this paper we want to address the recursive articulation that involves local cultures and skills and the related conventions. The research presented here attacks this problem from the point of view of the languages actors use to construct local CMs to objectify their cultures and skills and the related conventions for sake of cooperation. According to Ariadne's terminology, a language consists of a (sub)set of basic categories of articulation work together with their relationships as well as of their instances according to the specific work setting. Like in our previous work, the attention is on the communicative behavior of actors across local CMs. Unlike it, the focus is on the meaning of the categories and their relations constituting the content of the communicative actions, and not on the flow of control linking them in the protocols. In both cases, the problem is about the alignment of local points of view for cooperation to occur: however, since the objects of the alignment are different (protocols versus meanings) it is necessary to go deeper in the analysis of the space of possible solutions. In [5] we discussed the alignment of meanings with reference to the role of classification schemes in cooperation. Although in this paper we consider more structured descriptions than classifications of (articulation) work categories, the same arguments apply. They can be summarized as follows: 1) the working situations considered by the field studies dealing with alignment of meanings are very different and require the integration of different (organizational and technological) solutions; 2) the alignment can be made easier by the presence of some specific factors (like, social pressure or professional languages) but in any case local jargons and habits survive the effort to normalize the local points of views; 3) the alignment of meanings is a process that evolves with the communities using them. Autonomous behaviors and mutual learning processes (happening through communication and cooperation) dis-align and re-align meanings, respectively, in a dynamic way. By consequence, the design of a support of alignment of meanings has to consider both organizational and technical aspects. Moreover, it cannot impose a unique 'external' view and should support the dynamic interplay of "naturalization and the memberships" [6]. Finally, it should not aim at "glass-box technology or pure transparency" since both are "impossible" due to the fact that "anomalies always rise when multiple communities of practice come together, and useful technologies cannot be designed in all communities at once" ([6] p. 311). The paper is organized as follows: the next section provides a more precise explanation of what we mean by the articulation devoted to the re-alignment of meanings. Section 3 describes the experimental setting of an initial implementation dealing with a special case of alignment of meanings. Section 4 describes a more general case where the integration of the related approach in the Ariadne framework and underlying ABACO architecture is presented together with its current implementation. The conclusions identify the main open problems. 2
About alignment of meanings
The idea that the recursive articulation aiming at alignment of meanings has to be explicitly considered was derived from empirical investigations in the framework of the Politeam project [7]. In that experience, the effort to recompose dis-alignments was mainly performed during workshops where users, with the help of technologists, discussed the linguistic misunderstandings specifically induced by the introduction of a document management technology supporting their cooperation. Moreover, it was recognized that the outcomes of that re-alignment were not naturally incorporated in this technology, and users
194
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
were in charge to manage the established clarifications outside the system. From that experience, we derived the idea that the alignment of meanings can be obtained by the combination of two phases: the first one, where users are active in clarifying the problems and negotiate partial solutions; the second one, where this effort is paid back since the supportive technology alleviates the communication problems using of the collected information. On this basis, we implemented a specialized module, called Reconciler [8], which is a technology to interactively build the correspondences to solve conflicts related to concepts used in communication during cooperation between communities. The basic idea is to provide users with a framework where they can cooperatively discuss various kinds of semantic conflicts and establish correspondences between entities, attributes with their domains, and relations (the language the Reconciler provides to describe the local views) to (partially) solve them. Then an algorithm uses these correspondences to elaborate communication messages in order to improve their understandability by the receiver. The kind of communication misunderstandings leading the collaborating actors to start a reconciliation process influences the type of correspondences to be established. One kind of misunderstanding in communication can be related to the different terminologies actors are using during their collaboration. Since each of them communicates in terms of her professional domain, the construction of a shared space of understanding is needed to work jointly. In this case the actors using the Reconciler module can cooperate to clarify the meaning of the professional terms so that each actor, not skilled in the profession of her interlocutor, can construct the paraphrases' of the related professional terms according to what she perceives as their key features. The Reconciler is then able to elaborate the terms with the previously established definitions so to improve understanding during communication among the actors [9]. Another kind of misunderstanding can arise during inter-community collaboration since each actor describes the common space on which they will collaborate in very different ways. Descriptions serve like "maps" drawn using a set of concepts in which different levels of granularity, different names and structures represent the different but not irreconcilable perspectives the communities have on the same shared "territory" (see [10] for the metaphor of maps as artifacts giving different descriptions of a single common territory). Furthermore, since each local view represents consolidated conventions, each community wants to maintain it without "be enforced" to create and use a unique standardized global view. As a consequence of the above facts, the use of local descriptions in communication may generate misunderstandings. If a term used in a message is unknown or differently used by the receiver, we want to give her a chance to interpret it in her context. This is the main goal of the functionality incorporated in the Reconciler. From the implementation point of view, the Reconciler [8] was implemented as a stand-alone module characterized by the high modularity of its internal structure and by well defined information interfaces between sub-modules. This architectural choice makes the Reconciler module easy to integrate in different contexts. This possibility is valuable for two reasons: first, to achieve adaptability to different user needs and technological environments; second, to start the experimentation [9] of the Reconciler functionality as a fundamental step to identify features to be incorporated in the user interface and to evaluate the capabilities of the Reconciler algorithm in real communication contexts. This development path is not fully achieved and only partial implementations have been realized. In the following we give a description of what we have now.
1 From now on, according to the "jargon" used by users during experimentation, these paraphrases are called definitions.
M. Sarini and C. Simone /Recursive Articulation Work in Ariadne
3
195
The experimental setting
In order to start the experimental phase as soon as possible, we identified a context of use where the involved actors did not have a long tradition in local (intra-community) cooperation and any specific technological support to it. However, the users were acting as professionals discussing a shared problem by using their professional language and expertise. Their language is mainly expressed by entities enriched with attributes: hence, in relation to the richer capabilities of the Reconciler, relationships were not considered. Moreover, correspondences are between the terms of a profession and their descriptions from the point of view of the people of a different profession. Here below we sketch the scenario and the related user interfaces that have been constructed as the front-end to the core functionality of the Reconciler module. Two roles, the Chief Technology Officer Developer (CTO) and the Project Manager (PM) of a software development company, have to build a joint proposal for developing and marketing a new software product: a task manager for a PDA (personal digital assistant). In the first part of the experiment, the CTO and the PM use the Definition Interface (see Figure 2) to cooperatively establish the definitions for the various terms they are likely to use in constructing the proposal for the new product. Using this interface it is possible to define new concepts and associate with them new definitions or re-use definitions of other concepts acting on the Define As button. In the latter case it is also possible to better specify the definitions by enriching them with attributes. For instance in Figure 2 the concepts HTML and WML are both defined as WEB PROGRAMMING LANGUAGE. WML is specified with the attribute <usage: pda/mobile phone>, while HTML contains <usage: pc>. At the end of the definition process all the resulting information is saved in a persistent format used by the Reconciler module in the communication phase.
Figure 2: the definition interface
196
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
In the second part of experiment, the CTO and the PM decide which features of the system to include in their joint proposal. In so doing they communicate through their Communication Interfaces to exchange messages that can contain the concepts clarified in the previous phase. These interfaces, which are a specialization of a chat tool, split the screen in different areas (see Figure 3): • the Concepts list contains the list of the concepts defined in the previous phase according to the specific role's point of view, • the Message History panel keeps track of the messages exchanged by CTO and PM, • the Outgoing Message panel which contains the current message to be sent.
Figure
the message interface for the PM
The messages are a combination of free text (as for normal chat messages) and concepts selected from the concept list. The chosen concepts are inserted in the message text automatically between square brackets to remember the users that they are not simply words, but something they have already discussed before and the system handles in a special way. Once the message is written, the Send Message button activates the elaboration phase. The message is parsed to recognize the concepts that are transformed by the Reconciler algorithm according to the definitions of the previously constructed shared space of understanding. Since OCR is the only concept to which SOFTWARE TO RECOGNIZE HANDWRITING is associated (see Figure 2), the algorithm substitutes in all the messages OCR with SOFTWARE TO RECOGNIZE HANDWRITING (and vice-versa). In case of oneto-many correspondences, as for example in Figure 2 where HTML and WML are both
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
197
defined as WEB PROGRAMMING LANGUAGE, the definition interface, as explained above, allowed more specific definitions formulated using the usage attribute. So when the PM is writing a message for the CTO talking about WEB PROGRAMMING LANGUAGE if she specifies the value of the attribute usage as pda/mobile phone then the algorithm uses this information in a specific heuristic and transforms WEB PROGRAMMING LANGUAGE as WML solving the ambiguity for the multiple definition. Once all concepts are transformed, the message is sent to the other role. The experimental version does not use all the capability of the Reconciler algorithm: its purpose was mainly to test how users accepted the two phases process of reconciliation and their impacts on the communication interface that become somehow more complex that usual chat tools. Since the outcomes were altogether positive [9], we were encouraged to continue our integration effort toward the use of more complex schemes.
4
Integrating alignment of meanings in Ariadne
If local communities use a technology to support the articulation of local activities and the communication with the other local communities, it is natural to assume that the description of the used concepts are constructed by using the functionality made available by this technology. Consequently, it is natural for us to consider the Ariadne notation as the candidate technology and to integrate the Reconciler functionality in this framework. In the following we give the information about Ariadne, which is essential for this purpose. Ariadne is a user-oriented notation providing a set of Categories of Articulation Work (CAW) to define coordination mechanisms which represent different local conventions or procedures. Every single category can be considered as a class representing some aspects of articulation work, and is characterized by a set of attributes referring to some properties of the category or expressing relations with other categories. In Ariadne coordination mechanisms are composed of a coordinative protocol (an integrated set of procedures and conventions stipulating the articulation of interdependent articulated activities) and an artifact (a distinct and permanent symbolic construct which explicitly and permanently represents the state of the protocol). Due to the flexible nature of the notation, coordination mechanisms can be modeled using different languages according to the different set of Categories of Articulation Work chosen for their description. Furthermore their definitions can be completed in a fully incremental way, specifying the values of the attributes belonging to the categories of the chosen language with different levels of granularity at different times. Using Ariadne two or more coordination mechanisms can be composed to define a compound coordination mechanism, hi this way it is possible to model inter-community collaboration articulating distributed activities described by different coordination mechanisms. Parts of these activities are expressed in terms of communication among actors belonging to the different communities. This communication has to be understood by the involved actors in order to guarantee collaboration. The above mentioned misunderstandings are generated by the different ways in which local communities define their local coordination mechanisms using Ariadne. The solution to these communication problems requires the dynamic shift to a higher level of articulation where the involved parts can discuss about their current different local points of view in order to get mutual understanding. When the involved parts reach an agreed level of mutual alignment of meanings they can go back to their usual collaboration and use the information previously defined to improve their mutual understanding.
198
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
Figure 4: the articulation of meanings 4.1
The Reconciliation CM
In Ariadne this shift to a higher level of articulation is obtained by activating a specific coordination mechanism, called from now on Reconciliation CM, handling the process of reconciliation to align the different views each CM represents (see Figure 4). During this reconciliation process, portions of the coordination mechanisms employed in communication for articulation are 'materialized' in the artifact included in the Reconciliation CM, as explicit correspondences that can be used to avoid repeating the same negotiation in the future. In this sense, correspondences constitute a dictionary, which can be more or less complex depending on how the descriptions of the two different communities are "in conflict". When the involved actors go back to their usual activities, the defined correspondences are used by the communication support to re-elaborate the messages so that the human interlocutors can interpret them more easily, since they are reformulated in terms of the pertinent local descriptions. Here below we describe these aspects in detail. According to the Ariadne notation, where a coordination mechanism is composed of a protocol and an artifact, the Reconciliation CM includes as protocol the communication pattern governing the reconciliation process and as artifact, called from now on Reconciliation Artifact, the materialization of the correspondences the involved Roles established in this process.
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
199
4.1.1 The Reconciliation Artifact: constructing the schemes The Reconciliation Artifact contains the information of the local CMs and the dictionary constructed in the reconciliation phase to be used by the Reconciler module as part of the Reconciliation CM. Since the Reconciler module has been conceived of as a component to be used in combination with different cooperation supports, the above information has to be transformed in the representation language used by the Reconciler, namely as entities, attributes and relationships among them. The result of this transformation is called scheme. The transformation process has to take into account that in Ariadne the construction of coordination mechanisms is divided in two steps: the selection of the CM language and the CM definition according to it. The two steps generate a partial scheme and its completion, respectively. The choice of the language of the CM is about the selection of the categories used to describe the CM. In order to build the partial scheme, every category is translated into an entity of the scheme. For what concerns its attributes2: the attribute of type name becomes the key attribute name of the entity corresponding to the category; all attributes of type content become attributes of this entity keeping the same name; the attributes of type reference become relations between entities. Specifically, the relation is between the entity (representing the category in the scheme) the attribute belongs to and the entity (or entities) referred to by the value of the attribute. In Ariadne, for every attribute of type reference in a category there is, in the category referred to by the value of the attribute, an attribute of type reference that describes the inverse relation. For example, in the category Role the attribute responsible for refers to Tasks, and in the category Task the attribute under responsibility of refers to a Role. In the scheme the two unidirectional relations become a single bidirectional relation whose cardinality is the composition of their cardinalities. In the above example, the resulting bi-directional relation responsibility has cardinality 1:N since the relations responsible for between Role and Tasks has cardinality (1 :N) and under responsibility of between Task and Role has cardinality (1:1). The categories representing resources are handled in a slightly different way. Since they are especially involved in the reconciliation process we consider them in more detail. Ariadne makes available four kinds of resources: Informational, Material, Technical and Infrastructural Resources. The latter become four entities of the scheme carrying the same name. Attributes of type content., like location and state whose value is a data-frame directly become attributes of those entities. Attribute description of type content, which is different in each kind of resource, can be either a scheme, typically in the case of Informational Resources, or a data-frame, typically in the case of Technical Resources. The first possibility is the most interesting one from a reconciliation point of view. Hence, we consider only the Informational Resources in the explanation of the global scheme generation. They will be referred simply as resources. Moreover, we will call sub-schemes the schemes associated to the attributes description.
2
It is out of the scope of the present paper to give all details about the attributes. For the purpose of this paper it is enough to know that the attributes of type content are data structures while the attributes of type reference are references to other categories.
200
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
4.1.2 Completing the schemes The definition of a CM concerns the construction of the instances of the categories through the specification of the values of their attributes. Instances do not appear in the scheme. Instead, the values of their attributes become the domains of the attributes of the corresponding entities3. Specifically, the value assigned to each attribute of type reference is the Cartesian product of the domains associated to the name attribute of the involved entities.
Figure 5: the resulting schemes involving resources
Again, (informational) resources are considered in more detail. Since in the language phase the choice of a resource does not include the specific sub-scheme associated to its description attribute, in this definition phase the attributes of the resource are handled differently. Every value assigned to the attribute name is translated in the scheme as an entity carrying the same name. For example, if a resource is named complaint document, the scheme contains an entity complaint document in addition to the entity informational resource and defines an is-a relation between them. This amounts to consider this naming as a specialization of the resource: we refer to the specialized category by resource subcategory and the related entity as resource sub-entity. The latter inherits from the resource its attributes. A typical attribute of type reference is used by a Task: this attribute becomes a relationship between the sub-entity and the entity Task (see Figure 5). Moreover, the definition of a CM entails the definition of the attribute description by specifying the corresponding sub-scheme that becomes part of the general scheme. Since Ariadne requires the user to express an entity of the sub-scheme serving as connection with the resource subentity, the transformation automatically adds in the global scheme a standard relation, called component, with cardinality (1:1), to realize the above connection (see Figure 5). This completes the sub-category management. Obviously, the definition of a CM creates instances of sub-categories. Again, the latter do not appear in the scheme (dotted boxes in Figure 5). Instead, for all of them, the values of their attributes become the domains of the attributes of the connected sub-scheme, apart from the names of the instances of the resource sub-entity that become the domain of the attribute name of the resource sub-entity (bottom right of Figure 5). 3 The scheme does not consider additional information about the instances of the categories since this is not required by the functionality currently implemented in the Reconciler module.
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
201
4.1.3 The protocol: materializing the correspondences The main purpose of the protocol of Reconciliation CM is to support users in the reconciliation process, which is materialized in the correspondences between schemes elements. These correspondences have to be incorporated in the above described artifact to construct the whole information needed in the communication phase. As for any other CM, the protocol can be defined by the involved users. For sake of testing the integration we are aiming at, we defined the simplest protocol: all the users involved in the reconciliation process can communicate among themselves to define the correspondences, while a single Actor assumes the Role of Manager of the Reconciliation Artifact and is in charge of keeping it updated. This choice is however very reasonable since it leaves the users free to organize the reconciliation in a quite flexible way, in principle also by communicating outside the system: e.g., in a workshop like in the Politeam case [11]. In [8] we described how the Reconciler module guides the users in identifying and solving the dis-alignments between the local points of view. Here, we recall that this is achieved considering different types of conflicts. Taking inspiration from the conceptual modeling approach [12] the conflicts are: Naming conflicts (including Synonym and Homonym), Category conflicts, Structural conflicts (including Type and Dependency conflicts) and Unit conflicts. Notice that during the reconciliation process entities and attributes coming from the partial schemes associated to the languages to define the two involved CMs, mainly generate Naming and Unit conflicts. Instead, in the case of entities and attributes coming from the sub-schemes associated to the attributes description, the correspondences may involve all the types of conflicts that we are able to deal with, since the involved schemes are fully user defined. Moreover, since the domains related to all attributes are, in principle, dynamic as they collect the set of all values specified at definition time, the definition of the correspondences has, in principle, to be updated accordingly. This concludes the description of the Reconciliation CM. 4.2
Using correspondences in communication
The next step is to consider how the information collected through the Reconciliation CM is used by the Reconciler algorithm to support the comprehension of the communication exchanged by the involved Actors through their Communication Interfaces (bottom part of Figure 4). In Ariadne the communication contains the receiver and the message. The receiver can be an Actor or Role name. The message contains one action and a set of objects on which the action has to be performed. In the current implementation there are three types of actions: tell, ask, perform command. This limited choice is irrelevant from the reconciliation point of view since the latter involves only the objects mentioned in the message (we will comment on this point in the conclusions). The objects are specified using a semi-structure in an SQL-style, which combines entities, attributes, relationships and conditions. For example a message could be: perform a check on documents where the date is greater than 1-1-1999 and author is responsible for Prepare Budget. If the receiver is a Role its value is elaborated according to the correspondences among the domains of the name attributes of the entity Role. Nothing has to be reconciled if the receiver is an Actor. The elaboration of the message is performed by the Reconciler algorithm. This algorithm is composed of three subsequent steps.
202
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
1) The algorithm parses the content of the message in order to recognize the concepts used by the sender. Taking into account the semi-structure of the message, the recognized concepts are grouped into entities, attributes, relationships and conditions. 2) Then the algorithm searches for the correspondences involving the elements inside of every single group, by using the information established in the previous reconciliation process. The algorithm is driven by entities because people mainly use them to describe concepts. For every entity there are three possible situations: there are no correspondences in the other scheme (zero correspondence), there is exactly one correspondence in the other scheme (one-to-one correspondence), there is more than one correspondence in the other scheme (one-to-many correspondences). In the first case the algorithm does not propose any solution. For the one-to-one correspondence the entity is substituted with the corresponding one of the other scheme. For the third case the algorithm uses a set of heuristics based on the resolution of the conflicts proposed during the reconciliation process. The aim of these heuristics is to choose, if possible, an entity of the other scheme maximizing similarity with the current one. At this point, the same search of correspondences is made for the other elements. The heuristics for the solution of the ambiguities related to one-to-many correspondences between attributes and relations are based on the previously selected correspondences between entities and on how the current elements are linked to the latter. Furthermore for what concerns conditions, the algorithm elaborates the values of the attributes expressed in the conditions according to the solution of Unit conflicts provided by the users during the reconciliation process. 3) The algorithm substitutes the concepts recognized during the first step with the corresponding concepts found in the previous step. Notice that not all concepts could be substituted: this happens for zero correspondence and for one-to-many correspondences when the heuristics are not able to select a unique correspondence. In the first case the Communication Interface of the receiver (Figure 4) keeps in the message the element expressed in the language of the sender and notifies the receiver that it is a practical source of misinterpretation. In the second case the sender is asked to better specify the element in the message through its Communication Interface in order to overcome, if possible, the unsolved ambiguities. 4.3
The impacts on the underlying architecture
The Reconciler module is integrated in the current ABACO multi-agent architecture to provide the reconciliation services needed to support inter-community collaboration. Figure 6 shows the same structure as Figure 4: the main difference is that here boxes represent (and names refer to) ABACO agents, each one corresponding to a category of articulation work in Ariadne, but for the Reconciler agent which is automatically created inside each compound coordination mechanism. In ABACO, every Actor has a user interface allowing human actors to select the working space associated to their current Role and to handle the related duties. Hence, a human actor, with the appropriate responsibilities can, among the others, define and activate another CM, and specifically a Reconciliation CM. Figure 6 shows the situation of a compound CM composed of n CMs and focuses on the information flows between agents in the reconciliation process involving CM1 and CM2. • In Flow 1 an Actor involved in CM1 activates the process. This causes the Roles and the Reconciliation Artifact involved in the Reconciliation CM to be activated too. • In Flow 2 the Reconciliation Artifact retrieves from the Reconciler the schemes and the correspondences needed to support the current process of reconciliation between the two CMs. The Reconciler creates the schemes related to the two CMs as soon as the
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
203
definition of the CMs is terminated, according to the transformation process explained in the previous section. The correspondences are the results of the previous reconciliation processes between the two CMs which the Reconciler maintains in its data structures. In Flow 3 the Roles involved in the Reconciliation CM start the negotiation to find a mutual alignment of meanings. At the end of the negotiation, the Manager fills the resulting information in the data structures of the Reconciliation Artifact through the Reconciliation Interface. In this way the process of reconciliation is objectified in the Reconciliation Artifact in terms of the established correspondences. In Flow 4 the Reconciliation Artifact passes to the Reconciler the new information so that the latter can update the part of its data structures related to this particular pair of CMs. At this point the phase of alignment of meanings is terminated.
Figure 6: the integration of the Reconciler in ABACO
Now this information can be used to support human actors in understanding their communication for collaboration. • Flow 5 corresponds to the case where a Role belonging to CMi writes a message to be sent to a Role belonging to CM2 through the Communication Interface of the Actor associated to it. Before to redirect the message, the Reconciler checks whether its internal data structures contain information about the two involved CMs.
204
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
•
Flow 6: Since a compound CM can compose more than two CMs and each CM can activate several messages at the same time, the Reconciler has to manage all these communication in parallel. This duty is in charge of the Message Dispatcher which activates a specific thread for any communication across pairs of CMs. Each thread starts a local copy of the Reconciler algorithm to elaborate the message. • Flow 7: Once the message is transformed the Message Dispatcher redirects it to the receiver agent of the other CM. Notice that Flow 5 is bi-directional because in case the information of reconciliation is not complete and some conflicts cannot be solved, the user associated to the sender Actor/Role agent is asked by the Reconciler to solve ambiguities by using its Communication Interface. Moreover, the proposed architecture manages the possible parallelism of the reconciliation phase with the elaboration of the messages exchanged by the reconciled CMs. In fact, the former is in charge of the appropriate Reconciliation Artifact that serves as a temporary working memory while the latter is in charge of the Reconciler that represents a persistent memory where the reconciliation information is stored and subsequently used for the elaboration of the messages. 4.4
The current level of integration
The integration of the Reconciler in ABACO is not fully accomplished. According to our incremental approach toward a more complete solution, we developed an interface guiding users in the construction of their local schemes. This intermediate step allows to extend the experiment described in Section 3 to more complex situations without imposing a specific technology generating them.
Figure 7: the Definition Interface of the Reconciler module: an example of synonym and homonym correspondences
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
205
However, in this case the Definition Interface has to take into account the richer structure of the schemes and the fact that they can raise all types of conflicts we mentioned in section 4.1.3. To this aim we are able to re-use the interface we developed for the Reconciler module (see Figure 7). This interface is split in two parts: the one related to the first scheme, let's say A, and the other related to the second scheme, let's say B (in Figure 7 the schemes are related to two different university administration offices which have to collaborate). In the current implementation the interface is unique for the two schemes. The scenario of use of the reconciliation process supported by the Reconciler is supposed to be a negotiation involving people of the two communities where a facilitator is in charge to record the negotiated correspondences through the above interface. Each scheme is presented as the set of its entities, the related attributes and relationships among entities. Furthermore for each scheme there are sections devoted to show the conflicts the users have recognized. At the beginning these sections are empty. As the use of Reconciler proceeds, they are filled in with the correspondences among concepts in the scheme. The first kinds of correspondences to consider are those related to Naming Conflicts: that is identification of homonyms and synonyms. Once the involved roles determined this kind of correspondences, the system automatically detects the other kinds of conflicts related to the latter. Then the involved roles are called to solve them according to the various functionalities the interface provides. Once the schemes and the correspondences are constructed, they can be used in the communication phase: again, through an ad hoc Communication Interface if local communities are not supported by any specific cooperation technology, or through the Communication Interface incorporated in the Actor agents of ABACO. These communication interfaces are presently under development as well as the automatic schemes generation and completion when the Ariadne/ABACO environment is used.
5
Conclusions
The paper discussed the recursive nature of articulation work by distinguishing between the articulation needed for re-aligning protocols and meanings. While the former was the topic of a previous paper, the second one is the focus of the research effort presented here. The approach we propose to re-align meanings stresses the need to support this form of articulation work in terms of both the outcomes and the process generating it. This led to the identification of two main phases in this process: a first one, where reconciliation is constructed by identifying the sources of misunderstandings and the correspondences that materialize the outcomes of the negotiation of meanings. In the second phase the correspondences are used, with the support of a specialized technology, to improve mutual understanding in communication. The goal in the background of this approach is to let cooperating actors use their local languages (that are a mix of their professional languages and of the common languages developed during their cooperative experience) and to use the system to facilitate mutual understanding by transforming 'foreign' concepts into concepts that are part of the local culture. The resulting technology can be used in different scenarios, ranging from pure communication between professionals without any additional technological support up to situations in which the proposed technology has to be integrated in the technology currently used by local community to support their internal cooperation. As reference technology, we selected the Ariadne/ABACO framework we have developed to support articulation work oriented to protocols. The paper shows the main steps we have to perform in order to achieve this goal.
206
M. Sarini and C. Simone / Recursive Articulation Work in Ariadne
This difficult research path is not concluded. In order to verify the basic assumptions and to get hints on how to develop user interfaces, we started the experimentation of the technology in a simplified scenario. The outcomes encourage us to proceed toward a richer functionality using all the capabilities of the underlying transformation algorithm to be integrated in existing support to cooperation. In our opinion, the integration principles presented here, although specialized to a specific framework, can be re-used for other types of technologies since the only strong precondition is that the related information is expressed in terms of entities, attributes and relations: a standard modeling language that covers classical ER- as well as Object Oriented approaches. The paper discusses also how the design of the proposed technology can support incremental experimentation in increasingly complex scenarios. To this aim, the most significant open problem is the design of the Communication Interface that incorporates richer supports of communication without affecting its usability. Of course, also the capabilities of the transformation algorithm have to be verified in the field. We are confident that the approach we have taken will allow us to reach a good trade-off between richness and usability. In fact, the proposed functionality is not an imposed substitutive technology. On the contrary, it is a support local communities decide to use to improve a learning process oriented to increase their mutual awareness of their local points of view. Acknowledgments The authors want to thank Gloria Mark for the many discussions about the Reconciler and for her fundamental role in making its experimentation possible. References [1] Schmidt, K. and C. Simone (1996): Coordination Mechanisms: Towards a conceptual foundation for CSCW systems design. Computer Supported Cooperative Work (CSCW). An InternationalJournal, vol. 5. no. 2–3. [2] Divitini, M. and C. Simone (2000): Supporting different dimensions of adaptability in workflow modeling. CSCW Journal- Special Issue on 'Adaptive Worflow Systems', vol. 9, no. 3–4, pp. 365-397. [3] Donatelli, S., M. Sarini, and C. Simone (2000a): Negotiating propagation of changes in interorganizational workflows. Special Issue on Flexible Workflow Technology driving the Networked Economy, IJ of Computer Systems Science and Engineering, vol. 15, no. 5, pp. 357-370. [4] Donatelli, S., M. Sarini, and C. Simone (2000b): Towards a Contextual Information Service supporting adaptability and awareness in CSCW systems. In COOP 2000, Sophia-Antipolis (Fr), ed. R. Dieng, A. Giboin, G. De Michelis, and L. Karsenty, pp. 83-98. [5] Simone, C. and M. Sarini (2001): Adaptability of classification schemes in cooperation: What does it mean? In ECSCW2001, Bonn, ed. W. Prinz, M. Jarke, Y. Rogers, and K. Schmidt. Kluwer Academic Publishers, pp. 19-38. [6] Bowker, G. C. and S. L. Star (1999): Sorting things out: classification and its consequences. Inside Technology, ed. W.E. Bijeker, W. B. Carlson, and T. Pinch. Cambridge, MA: The MIT Press. [7] Mark, G., L. Fuchs, and M. Sohlenkamp (1997): Supporting groupware conventions through contextual awareness. In ECSCW97, Lancaster, ed. J. Hugues, W. Prinz, T. Rodden, and K. Schmidt. Kluwer Academic Publishers, pp. 253-268. [8] Simone, C., G. Mark, and D. Giubbilei (1999): Interoperability as a means of articulation work. In WACC'99, San Francisco. ACM Press, pp. 39–48. [9] Mark, G., V. Gonzalez, M. Sarini, and C. Simone (2002): Reconciling different perspectives: an experiment on technology support for articulation, (in this volume) [10] Kent, W. (1978): Data and reality: basic assumptions in data processing reconsidered. New York: North Holland. [11] Wulf, V. (1997): Storing and retrieving documents in a shared workspace: experiences from political administration. In INTERACT'97, ed. S. Howard, J. Hammond, and G. Lindgaard. Chapman & Hall, pp. 469-476. [12] Batini, C., M. Lenzerini, and S.B. Navathe (1986): A comparative analysis of metodologies for database schema integration. ACM Computing Surveys, vol. 18. no. 4. pp. 323-364.
Cooperative Knowledge Management and Context
This page intentionally left blank
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press, 2002
209
Cooperative Organizational Memories for IT-based Process Knowledge Management August-Wilhelm Scheer Frank Habermann Oliver Thomas Christian Seel Institute for Information Systems (IWi) Saarland University Im Stadtwald, Building 43.8 D-66123 Saarbruecken (Germany) Phone: +49 (0) 6 8 1 / 3 0 2 – 3 1 0 6 Fax: +49 (0) 6 8 1 / 3 0 2 – 3 6 9 6 E-mail: [scheer/habermann/thomas/seel]@iwi.uni-sb.de Abstract A critical factor of success for the realization of innovative business strategies is the knowledge about procedural organizations. Consequently, the management of business process knowledge has become a central economical challenge. This paper demonstrates the development of a computer-based organizational handbook to efficiently organize the process knowledge within the Information Technology (IT) domain and to support IT-based business process improvements. As a solution serves the concept of the Organizational Memory that can pragmatically be considered as a medium to learn about the past and make knowledge applicable for the future. Keywords Knowledge Management, Impacts of Information Technology, Information Systems, Organizational Memory System, Business Process Improvement, System Design
1 Introduction During recent years, the amount of scientific articles surrounding the topic 'knowledge within a company' has multiplied. The significance of knowledge in the management re search sector results to a lesser extent from a product's theoretical point of view. In fact, the general perception is that in times of market expansion, the command of organizational knowledge becomes a critical success factor and accelerates this development [55]. Thus, the current focus on knowledge within a company is based on a central organiza tional-theoretic problem: How to design organizations that they can meet the demands of the external environment? [34, p. 42]. The research on this topic covers the areas of organizational development, innovation management and organizational learning about which exist a multitude of concepts. Knowledge Management is a relatively young discipline that is synthesized in these extensive theoretical scientific papers. It focuses on the competitive role of knowledge within a company and deals with the modified question: "How to plan,
210
A.-W. Scheer et al. /Cooperative Organizational Memories
organize and control organizational knowledge to secure the productivity of a company permanently?" [17, p. 44]. Admittedly, an analysis of knowledge management projects has shown that the utilization of theoretical knowledge management concepts is difficult in practice. Below there are a few examples of such problems: • Simplification: During the design of knowledge management systems, knowledge types and processes are simplified which usually results from lack of time. Therefore, the developed systems do not support solving the primary problem. An example is the employment of gross phase models for the control of knowledge processes. • Deficit of theory: Though conclusions and recommendations are derived from theory, the expected outcome cannot be reached. An example is the theoretically justified statement that the right knowledge has to be available 'at the right point in time and place' in order to lead to a desired effect, yet it has already been observed that in practice the op posite can occur. Such effects are usually caused by unknown interactions and unex plored cognitive and social coherences. • Variety of theory: The partly mixed theoretical approaches can lead to the attempts of companies to find their own 'best' theoretical concept, which subsequently serves as the basis for system development. In such cases, there is a certain danger that either the projects stagnate at an early stage or an already adopted draft has to be discarded in the implementation phase. The aim of this article is to describe an Organizational Memory System (OMS) that supports the management of knowledge. This includes the design of business processes using modern information technology. The computer-based information system was developed at the Institute for Information Systems, Saarland University, during the research project 'De velopment of a computer-based organizational handbook for IT-based process improve ments', funded by the Deutsche Forschungsgemeinschaft (DFG). 2 Process Knowledge and Information Technology From 1990, the role of information technology (IT) has been seen as the "enabler of process innovation" [16, p. 47] that has reached the minds of organizational decision leaders. No other resource raised so many hopes for developments in the area of business process management. Whereas the expectations on IT focused primarily on the radical change of business operations, the continual improvements of business processes were becoming increasingly the focal point of interest. The high expectations on IT systems are also reflected in monetary measurements. No other resource has caused so many investments during the last decade. Admittedly, experience has shown that the acquirement of new IT does not necessarily lead to an improvement of business processes. Notably within the administration and service sector as well as the non-producing sectors of industrial companies there was no evidence found of a positive correlation between the IT-expenses and productivity. Paradoxically, it could even be proved that an increased use of IT can have negative effects on the business processes [9]. The availability of IT is nevertheless the deciding factor for realization of innovative business models. More important than IT, which is usually already available or can be pur chased in a cost-efficient way, is the knowledge of business-economical potentials and sequenced organizational effects of the use of IT [45]. In the case of Electronic Commerce, it is for example simply not enough to make an online order form available for the customer via the Internet. Only if on the one hand the logistic processes of a substantial flow of goods are changed according to new requirements and on the other hand the sequence of organizational resources in the 'back office' are respectively reorganized. Electronic Com-
A.-W. Scheer et al. / Cooperative Organizational Memories
211
merce can be successful beyond pure rationalization effects [46]. The knowledge about business procedures has thus become a critical resource and the management of business process knowledge (short: process knowledge) a central challenge. A brief study of the literature already clarifies that the basic idea of the development of computer-based systems for the management of process knowledge is not in the center of interest. This has already been proved theoretically and practically, e.g. based on process model repositories for business process reengineering and the implementation of business economical standard software [44], It is therefore of interest to show the possibility of how a knowledge management system is designed and how it meets the requirements of certain fields of application. As mentioned above, the system described in this article is used particularly for the management of knowledge about the design of business processes. This knowledge domain is called 'IT-based process improvement'. The scope of the system illustrates the so-called '7-S-modeF [36, p. 93] shown in Figure 1.
Figure 1: Delimitation of the application field 'IT-based process improvements'
The seven components 'Systems', 'Shared Values', 'Style', 'Structure', 'Strategy', 'Staff and 'Skills' as well as their interactions, characterize the complex system of an enterprise. Each component and each connection between the components can be seen as a subject of organizational measures. The developed system represents IT-related knowledge to be used to improve business processes. Vice versa, the effects of the business processes on the organization of information technology should not be taken into account. 3
Systems and Technologies for the Management of Process Knowledge
To support the management of process knowledge a broad variety of systems and technologies is available. Examples for suitable applications are • Know-how databases [51], • Data warehouses [27; 18; 43; 28], • Model databases like 'VISIO' by Microsoft, 'Rational Rose' by the Rational Software Corporation and the ' ARIS-Toolset' designed by the IDS Scheer AG, • Hypertext solutions [12; 23] and methods to build up hypermedia-knowledge networks like concept mapping [21] and mind mapping [10], • Knowledge maps [38, p. 104], like 'yellow pages' [11] or • Artificial intelligence (AI) systems [6] like Case-based Reasoning (CBR) systems [54] or Expert systems [4].
212
A.-W. Scheer et al. / Cooperative Organizational Memories
These exemplary categories are overlapping. None of the mentioned systems is based on one single technology, but in fact they are complex 'technology bundles' [29, p. 163]. Certain systems also show conceptual similarities. For example, model databases can save knowledge as reference models and can employ CBR-mechanisms to adapt reference models. During the introduction of mySAP.com solutions this is done by a question/answerdialogue through which pre-defined templates in the form of so called 'collaborative business scenarios' can be adapted to the demands of a business reality [26, p. 6]. Further application examples for technologies mentioned above can be found in [8], [22] and [30]. In theory and practice, further technologies are discussed in relation to knowledge management, e.g. Recommender and Pointer systems [40] and frequently asked questions (FAQ)-systems [5]. Yet these systems are mostly specialized versions of the presented systems and do not render an original contribution to the management of business process knowledge. 4 Organizational Memory 4.1 Concept of the Organizational Memory Although the role of IT as a support of knowledge management is generally accepted and, as illustrated above, numerous publications deal with the development of corresponding hard- and software systems, certain disagreements about the characteristic properties of knowledge management systems can be found. Representatives of Al research e.g. already claim that the terminology of their designed 'knowledge-based systems' can serve as a tool of knowledge management. Furthermore, experience has shown that expert systems only cover a very limited field of application because of their strict formal requirements. The relevance of knowledge-based systems for the coverage of important challenges like e.g. organizational design tasks is therefore questioned [33, p. 441]. More concrete is the concept of the Organizational Memory System that receives a stronger interdisciplinary point of view. Admittedly, Organizational Memory Systems are defined to different parameters and sometimes reduced to the features of a data memory. The term Organizational Memory (OM) is the subject of US-American research since the 60's. Numerous scientific disciplines like communication research, psychology, sociology, Al research and business administration have dealt always since the beginning with this topic from very different angles. In this regard, widespread theoretical work has been completed, but so far, no generally accepted interdisciplinary concept for the OM could be found. At least all efforts of definition have the causality between memory and learning in common. So, the OM is seen as a medium to learn from the past and into the future. This will thereby increase the efficiency and effectiveness of an organization. In this sense, an OM can generally be defined through the features of its processes and respectively its functions that handle certain knowledge contents and strike for the goal of efficiency and effectiveness. [53, p. 61; 48, pp. 19, 22]. 4.2 Functions of an Organizational Memory An Organizational Memory can be understood as a process that is not cognitive - in contrast to the human memory [48, p. 26]. Numerous authors have characterized these organizational knowledge processes and tried to point out the central functions. They are outlined as follows: • Documentation of knowledge: the acquisition and the detection of relevant knowledge media and the documentation of knowledge contents. Gathering of knowledge refers to
A. - W. Scheer et al. / Cooperative Organizational Memories
213
acquisition of employee knowledge as well as to the detection of already documented knowledge contents e.g. via new media. • Integration of knowledge: the semantic linkage and the physical storage of knowledge objects based on a logically integrated meta structure. This structure should be open and expandable so that different knowledge contents and media can be combined. • Utilization of knowledge: the search and navigation for the relevant knowledge to solve a problem, its selection and distribution to the involved persons via networks as well as the presentation of knowledge contents. In the broader sense, the application of knowledge also includes the retrospective evaluation of the used knowledge. 4.3 IT-Support of an Organizational Memory Numerous authors have dealt with the defined features of an OM, yet there are just few publications that outline the technical support and implementation of the OM concept as a central topic. According to Lehner, a reason for this is that only by now suitable technology-concepts are available enabling the implementation of OM requirements [31, p. 48]. An OM is designed with the goal to support organizational learning. Firstly, it is under stood as the presupposition to use experience and knowledge of the past for future business decisions. Secondly, it is in itself part of organizational learning and has changed over time. However, it is worth to note that, with this interactive relation, OM as well as organizational learning still has further effects. In this context Walsh and Ungson speak about the "use, misuse and abuse" [53, pp. 74-77] of an OM. Since OM can be understood in various ways [53, pp. 74-77], it is difficult to define OMS and to distinguish it from other systems like Databases, Data Warehouses, Model Repositories as well as from general organizational resources like organizational handbooks. One of the few attempts of theoretical explanation is outlined by Stein and Zwass. They define an OMS based on the OM definition of Walsh and Ungson [53] as the amount of all technical resources that is used to support the functions of an OM [49, p. 95]. Among other things, Lehner also defines an OMS by its functional features [30, p. 167]. If the definition also takes the textual alignment of a certain field of application into account, an OMS can be understood as a computer-based system that combines different basic technologies like modeling, database and retrieval technologies to continually gather and integrate the relevant knowledge of a well-defined field of application. This includes all types, levels and contents of knowledge and makes it applicable for further business decisions [53, pp. 6164; 49, p. 95; 1, p. 213]. Figure 2 illustrates the interaction of the OMS components.
Figure 2: Components of an Organizational Memory System
214
A. - W. Scheer et al. / Cooperative Organizational Memories
Examples for systems that could be defined in a broad sense as an Organizational Memory System are • Process Warehouses [44], • Handbook of Organizational Processes [33; 3; 14], • QuestMap [47; 13], • AnswerGarden [3; 2] and • Workflow Experience Databases [25; 41; 51 ]. 4.4 Architecture of an Organizational Memory System As shown in the overview of already existing OMS the systems and technologies mentioned above can be used to support the functions of an OMS. It is obvious though that its development does not falter from a deficiency in suitable technologies - "the next generation of knowledge support systems will be limited rather by our creative imaginations, than by the availability of technologies" [20, p. 14]. There is an almost non-manageable diversity of technologies, which offer potentials for the realization of an OMS. The difficulty is to find out the relevant system features, to choose suitable technologies and to integrate them in a way that creates a logical overall concept. The question to be answered is therefore: "How to design an OMS to meet the requirements of a certain field of application." [33, p. 441] The architecture shown in Figure 3 according to [35, p. 138] outlines the components of an OMS as well as its functional interaction. Similarities to the architectures of database and expert systems are obvious.
Figure 3: OMS-Architecture
On the organizational layer of the architecture, e.g. the role of the knowledge manager is clarified or the scope of development in gathering knowledge. By the use of the technological platform, the integration of the OMS into the operative application systems is carried out. On this basis, the usage interface of the OMS is defined (interaction layer). The usage interfaces should be adaptable to the demands of single knowledge owners by providing the possibility to create personalized profiles. In this way, they can create a personal portal to document, integrate and utilize knowledge. Supporting tools for this challenge are designed on the tool layer. The design of the suitable tools takes place on the conceptual layer via the OMS meta model. It describes the abstract structure of the elements of the respective
A. - W. Scheer et al. / Cooperative Organizational Memories
215
field of application which should be supported by the OMS i.e. knowledge owners, their organizational coordination, knowledge forms, documents, contents and their relation to each other. In that, the formulation of the meta model is a significant, intellectual challenge during the development of an OMS. The importance arises out of the fact that the metastructure delivers the implementation model for the physical data storage. Finally, on the physical layer, the scheme of a certain database or hypermedia system is derived from the meta model. Commercial knowledge management systems like KnowledgeX, Grape Vine or Hyperwave represent integrated technical solutions to support single OMS functions as for example mind mapping for the integration of knowledge or agent-based knowledge search and distribution. They are not oriented in certain fields of application and do not possess a concept for implementation. They rather offer a pre-selection of available technologies than a textual solution for the problem. A company for example which wants to organize the process knowledge of its product design, the internal consultancy or the customer relationship management, has to design its knowledge management in a conceptual way. A preimplemented technology choice can have hindering effects [42, p. 14]. First of all the field of application should be described and on this basis, the decision about the essential knowledge management tools should be taken. Only in this way, the demands can be checked and be fulfilled by a commercial solution [52, p. 57]. In the following this design task for the application field of 'IT-based process improvements' is demonstrated exemplarily on the basis of the meta model at the macro level. Firstly, those objects of an application field are modeled with their respective relations that have to be managed by an OMS. On this basis, the concrete requirements to the design are defined i.e. the OMS components to gather, integrate and implement knowledge are conceived. The overall aim is to design a system, which helps to efficiently organize the knowledge of IT and to support IT-based improvements of business processes. Thereby the term 'IT-based process improvements' identifies variations of business processes that increase their efficiency by the use of IT. 4.5 Meta model of an Organizational Memory System A central aim of modeling is to increase the understanding of the field of application and to offer design proposals based upon this. This task includes the clarification of relevant terms and the fixing of a uniform terminology. The term 'Process Improvement' could e.g. not only be interpreted as the procedure that aims on improving a business process but also to the result of such a procedure, i.e. it could be interpreted as an improved process itself. While the meaning of this term can often only be guessed from the context of an essay, there has to exist a unique definition within the scope of system development. Macro models [7, p. 229] make the first step towards this definition. They represent models which can be analyzed in detail and are suitable for roughly structuring a complex field of application as well as to give an overview of the relevant modeling components. In the Unified Modeling Language (UML), the package diagram [19, p. 115] performs the task of macro modeling. The notation reference should be made to the specifications of the Object Management Group [39]. On an abstract level, the package diagram presented in Figure 4 shows the basic interdependency of the components of IT-based process improvements. The relations between the packages are indicated by arrows. They refer to the fact that there exists at least one object class in each associated package that supports the relations. An 'organizational unit' can express an 'improvement proposal', which can lead to an 'improvement project'. The 'improvement project' is turned into concrete 'improvement measures' that are also carried
216
A. - W. Scheer el al. / Cooperative Organizational Memories
out by 'organizational units'. An improvement proposal should already contain hints to a concrete solution method, i.e. contain planned improvement measures [50, p. 23]. The relation between 'improvement proposal', '-project', and '-measure' is consequently transitive. In the context of an 'improvement controlling', the conducted 'improvement measures' and '-projects' are rated according to economical aspects.
Figure 4: Macro model of IT-based process improvements
The components 'improvement proposal', 'organizational unit' and 'improvement controlling' are most independent of a special improvement domain. They are designed in a way that they can be implemented almost unchanged into other fields of application of the improvement management. The concrete field of application 'IT-based process improvement' is characterized by the component improvement domain. This component marks the technical hot spot of the framework. If the framework should be reused for another subject area, it would have to be replaced. When designing the associated OMS micro models one should keep in mind that these models should be principally reusable for different fields of application of improvement management. This reutilization of model and system modules is an essential goal of the component-based system development [37]. Special focus on the application field of 'ITbased process improvements' takes place within the package 'improvement domain' (see Figure 4). Because of its special features and the interaction with other packages, it brings together the overall model and gives early hints of the system concept. A detailed overview of the micro-model would go beyond the scope of this article, thus it shall be referred to [24]. 5
Realization of the OMS-Functions
Based on the main functions of OMS as well as the OMS meta model for the field of application 'IT-based process improvements', the following, supporting OMS tools were designed: documentator (knowledge integration), mind mapper (knowledge integration) and improvement process generator (knowledge utilization). The basis of this overall concept is the operator concept. 5.1 Operators as Design Patterns The design of the function 'knowledge documentation' directly depends on the organizational integration of the system. For certain OMS applications, the introduction of a 'chief knowledge officer' (CKO) analogue to the chief information officer (CIO) has proved to be
A.-W. Scheer et al. / Cooperative Organizational Memories
217
successful [17]. He is meant to locate relevant knowledge of a field of application, gather it and make it utilizable by adding meta data [1]. This organizational concept however soon may reach its limits concerning the huge amount of users and the special diversion of the knowledge. Thus, the field of application aims to gather the knowledge directly 'at source'. Through the implementation of semi-automated documentation tools IT knowledge owners should be able to send their contents immediately to the OMS. Therefore, it is presumed that all IT-based improvement measures can be traced back to a finite number of so-called 'operators'. These operators are to a certain extent design patterns for future design steps to realize an IT-based process improvement. The operators are created by the use of enterprise models. Thus, individual corporate models or reference models can be used to show the general industry-typical design possibilities of IT. To obtain all relevant operators it has to be abstracted starting from a certain field of application e.g. chemical industry, banking or civil service. The definition of the operators takes place on the meta level. Operators are paste-, delete- and change-operations of objects, attributes and domains of the elements of the system 'IT-based process improvements' (see Figure 5). Meta level (abstract design pattern)
OPERATOR
Operator identifier Operator Name Operator_Description Operator_Generate() Operato_Delete() Attribute_Change() Operator_Part() Operator_Specify() Abstraction of application contents Application level (typical design pattern)
• Technical implementation of new business-economical standard software for sales • Introduction of a new data base system for construction (Germany) • Introduction of a workflow management system for personnel administration (USA) • Integration of workflow management systems for personnel administration (for the whole enterprise) • Introduction of new version of an Office application (for the whole enterprise) • Upgrade functions of an individual software solution for inventory management
Instance level (concrete design pattern)
• Upgrade function 'portfolio management' (project 02/1993) > Upgrade function' receipt posting' (project 08/1995) • Upgrade function 'product search' (project 03/2000) • Introduction of MS Office 95 (project 09/1996) • Introduction of MS Office 97 (project 01/1999) • Introduction of MS Office 2000 (project 05/2000)
Figure 5: Operators to document process improvements
As result of this procedure, there arises a large number of text modules, respectively all samples for the design of IT-based process improvements, which can be arranged, searched and applied according to certain criteria. With the operators being derived from business process models via different levels of abstraction (see Figure 5), they are inter-subjectively comparable to the created knowledge descriptions that go along with them. Simultaneously,
218
A. - W. Scheer et al. / Cooperative Organizational Memories
the textual operators are understandable and applicable for knowledge owners of all levels of qualification without knowledge of modeling methods. On the meta level, the repository of the OMS is defined to an operator-based knowledge gathering. The meta structure should be flexible and expandable so that the operator concept can be adapted to the individual demands of a company. These adaptations contain textual as well as formal changes of the operators. To illustrate the adaptability and expandability of the operator concept, the respective object methods have been modeled in Figure 5. Through splitting and specifying operators, their structures are identified. These can be arranged and built up according to certain criteria. The application level in Figure 5 shows examples of typical operators, which can be used as samples for gathering IT-based process improvements. The given operators result from several specification and decomposition steps and are to be found on different levels of detail. The operator 'Technical implementation of new business-economical standard software for sales' includes e.g. the specifications 'software resource/standard software/business-economical/sales' and the decomposition within the technical implementation. To introduce a workflow management system, a structural relation of the operators is presented on the level of application. By use of typical design patterns of the application level onto concrete improvement projects, the IT-based design knowledge is gathered. This happens on the instance level. In Figure 5 there are given concrete project histories for two operators of the application level to exemplify this. Within the OMS meta model this operator concept has been integrated by implementing the object class OPERATOR (see Figure 5). The association class 'operator structure' describes the relations between the operators attained via specialization and decomposition. Because of the associations between the operators and improvement projects, concrete improvement measures can be gathered. Since the operators are developed on different levels of detail, improvement measures can also be gathered from different technical levels of aggregation. 5.2 Documentator - Knowledge Documentation The documentation of knowledge is not ending with the operator concept. A list that has been created via operators contains only one part of the knowledge about an IT-based process improvement. To comply with the improvement project completely, this knowledge description has to be supplemented with those knowledge objects, which have been identified within the meta model. It has to be registered for example, which organizational unit submits an improvement proposal and how to rate the efficiency of the proposed project. Moreover, the OMS should allow the integration of all documented knowledge objects. Thereby several forms of integration can be differentiated. The basis of the technical integration is the developed meta model. It is the conceptual model of the OMS repository and enables to link project-related answers of the so-called 'W-Questions' (who, why, when, where and so on) in a semantically correct correlation. Those links can be designed flexibly via hyperlinks. In the context of the technical integration, the OMS has to enable the implementation of different media like text, sound, picture and film. This refers e.g. to the graphical presentation of project results, audio and video records of project meetings and textual project reports. The core elements of an IT-based process improvement to be registered have been already marked in Figure 1 in form of a causal relation: IT and business processes correlate to each other like reason and effects of the design decision which is indicated by the arrow. The arrangement criteria to be followed when documenting IT-based process improvements
A.-W. Scheer et al. / Cooperative Organizational Memories
219
are thus the information technology, the business processes as design objects and the design decision. Due to the existing multi-causality (e.g. changes of the IT can cause several effects on the process sequences), the arrangement criteria can be interpreted as dimensions of the cube presented in Figure 6, which fill the space of IT-based process improvements. One ITbased process improvement to be registered includes one or more sub-cubes. To improve the information processing of a single technical function (intra-functional), an isolated ITinvestment could e.g. be used (upper left sub-cube in the fore) or an overall processcomprehensive project could be carried out which concerns all IT functions (sub-cube on the right).
Figure 6: Dimensions of IT-based Process Improvements
These considerations are also reflected in the user interface of the OMS presented in Figure 7. The menu bar is split into the basic functions of the OMS: OMS-user ('User'), improvement projects ('Project'), search and navigation of knowledge ('Search') and help to use the OMS ('Help'). The operational range of the OMS is divided into the 'Explorer' (on the left) and the 'Viewer' (on the right). The Explorer shows all projects of the IT-based process improvement. They can be arranged and searched according to certain criteria. Explorer and Viewer are connected logically: an improvement project, which has been chosen in the Explorer is shown in detail in the Viewer and can be manipulated there. The map 'Overview' summarizes all basic project information. The right part includes data about the project leader, the project time, the project process and the project status. On the left side of the map the central elements of IT-based process improvements are characterized: business processes, IT resources and the manner of the design measure (c.f. Figure 6). The title of the improvement project is defined via these central elements in a way, which is uniquely identified and usable in applying knowledge. The assembled process title always shows the structure '[CHANGE] [IT-RESOURCE] within a [BUSINESS PROCESS]'. The active project of Figure 7 is called 'New web server for customer order processing'. The components of this title have been chosen by the user when creating a new project via the buttons 'Business Process', 'IT Resource' and 'Operation'. They present the interfaces to the operators, IT resources and project databases of the OMS repository. Apart from the naming, the user can also describe improvement plans. The map 'Operations' allows defining measures that are essential for realizing the improvement process. They are named according to the same model like the improvement projects based on the operator concept. Thus, improvement projects, sub-projects and
220
A. - W. Scheer et at. / Cooperative Organizational Memories
measures have the same structure and can be connected and searched for according to the same criteria. This is the substantial quality of the OMS to reach an effective knowledge application.
Figure 7: User Interface of the OMS
The concept indicates a limit to the number of elements of a measure plan. For these purposes, improvement processes can show the character of sub-processes and contain a measure plan in turn. This matches the decomposition of project knowledge. With the project title and the measure plan, the structured center of the knowledge description has been defined. It can be connected flexibly with further knowledge objects. These are references to the project associates (map 'Members') as carriers of implicit knowledge and the implementation of explicitly gathered knowledge documents (map 'Documents'). 5.3 Mind Mapper - Knowledge Integration The presentation of different logical relations between the elements of the repository is the task of the Organizational Mind Mapper (see Figure 8). Nodes of the resulting organizational mind map are process improvements and improvement processes which have been described via the documentation function of the organizational handbook. The mind mapper detects the semantic relations relevant for the process improvement between them and displays them. From the knowledge point of view, it is to some extent a tool for enterprise modeling. An essential goal is the formulation of suitable criteria to integrate different forms of knowledge and to connect different improvement measures to a multidimensional organizational 'knowledge map'. Thus, it is not only important what are suitable network criteria from an organizational point of view, but also the net has to be technically navigable and searchable.
A.-W. Scheer et al. / Cooperative Organizational Memories
221
Figure 8: User Interface of the 'Mind Mapper'
Due to the application purpose of the OMS, the traced goal is also to present network criteria which are not strictly formalized. Users should be able to navigate within semi-formal structures. Examples for network criteria are e.g. reasons, goals of improvement measures, time references, success effects or initiators of process improvements, as well as improvement process responsibilities, costs, lead time or tools used for the improvement process. 5.4 Improvement Process Generator - Knowledge Utilization The multidimensional-networked process improvements that are registered by the Documentator and the Mind Mapper should be usable as starting points for the definition of new improvement projects. The 'Improvement Process Generator' has to be designed in a way the documented process improvements can be used to draw conclusions for the management of future improvement processes. Thereby, improvement projects should be samples for the deviation of new improvement structures and support future improvement processes. The concept of suitable navigation and search is a presupposition of the knowledge application. The OMS should support the re-utilization of already saved improvement projects through a preferably active and context-sensitive knowledge supply [1]. An improvement project that is stored within OMS is regarded as helpful if it contains certain knowledge descriptions contributing to the coverage of a current knowledge requirement. The user chooses the documented process improvements as starting point and 'surfs' within groups, indexes and lists. If a similarity can be found between knowledge contents during this 'organizational mining', the documented improvement projects can principally be used as 'reference models' for new improvement plans. In contrast to the popular term of the reference model, an already documented improvement process identified as similar is not a 'common practice solution', i.e. not an abstract procedure for a certain type of problem. The documented process that is used as a template as well as the corresponding new
222
A.-W. Scheer el al. /Cooperative Organizational Memories
improvement project, represents an individual specification. Thus it has to be quoted to a certain extend as a 'best local practice solution'. The project knowledge can be browsed according to the mentioned explorer respectively mind map structures (Figure 8). The presentation of the found knowledge follows the maps 'Overview', 'Details', 'People' and 'Documents' (Figure 7). The search criteria implemented into the project explorer to identify improvement projects are presented in Figure 9. The specifications of the search criteria arise in the project explorer as a form of sub-indexes. By means of the criteria and their specification, an amount of stored projects can be divided during several steps and thereby identify suitable improvement projects.
Figure 9: Criteria to identify Improvement Projects
Using the mentioned criteria, the total amount of stored improvement projects is divided into sub-amounts. In case of a large amount of projects, this form of knowledge search can lead to several ramifications. Therefore, it is meaningful to limit the search area before starting the knowledge navigation, e.g. in form of a well-directed search. For these purposes, a simple full text search has been implemented which enables to insert the search item in the explorer field. All project titles can be found which contain the search item. The result serves as a starting point of further knowledge navigation. For the modulation and adaptation of the discovered solution, as example of intended improvement projects, the OMS-tools can be in turn used to gather knowledge. By this, they do not only fulfill the tasks of the actual registration of ongoing improvement projects but also of the ideal description. The controlling of improvements can be supported by comparing ideal and actual situation respectively past projects, which serve as examples, and new projects, which are based on these examples. 6
Organizational Memory - Future Challenges
The majority of managers and staff members consider the information and communication technology as the decisive competition factor of their company. All parties are aware that the working environments and business processes need to be changed to utilize this competition factor. In the last decade the willingness to actively start this change, to be always informed about new technologies as well as the motivation for continuous learning have permanently increased [15]. In this paper, organizational memory systems have been treated as a means for supporting this organizational change. Thereby IT-based changes were the focus of interest.
A.-W. Scheer et al. / Cooperative Organizational Memories
223
For this field of application, a concrete solution has been developed based on a semantic model. The result is a semi-structured hypermedia system, which connects explicit and implicit knowledge. The center of the system solution is the operator concept that has been designed based on process models via several levels of abstraction. It offers orientation to the users and helps them to systematically gather process knowledge and make it applicable. The developed OMS concept has been prototypically implemented at the Institute for Information Systems at Saarland University. It is almost impossible to provide a thoroughly functional and textually sophisticated application system to support the IT-based process improvements of a company. Moreover, the necessity consists in allowing for the 'expansion' of the OMS within the company. That is to say that the contents as well as the expert terminology have to be adaptable to create the required acceptance of the user and to sharpen the awareness for IT-based process improvements. In general, the OMS should offer a basic functionality that initially allows a usage independent from the objective 'process improvement'. That means the design has to be created in a way that the system can serve as an effective aid for the company right from the beginning. By using the basic functionality, the organizational members should however be prepared for the advanced functionalities which are going to support the process improvements. Therefore, the basic functionality of the OMS needs to be chosen in a way that it supports an existing operational task and implements the additional usage of the process improvement. The documentation function of the OMS meets these demands. The documentation of organizational measures is important for a company even without the objective target 'process improvement' - be it to inform the staff, to present or evaluate reorganizational measures to either management or staff association, to instruct and train new staff members or certificate according to ISO 9000 standards. At present there is still a lack of experience about the manageability of an OMS as well as how to adapt it to technical advancements like e.g. new Internet standards and efficient designs. Future challenges representing the complex topic OMS are notably viewed within the organizational implementation of organizational memories and the analysis of the effects of organizational memory systems [30, p. 443]. Organizational memory approaches are doomed to fail when either the development of the organization or its staff does not work efficiently. Cultural and infra-structural correlations of companies like e.g. the staff's willingness to learn will continue to be indispensable in successful implementation strategies of organizational memory systems. References [1] Abecker, A.; Decker, S.; Kuhn, O.: Organizational Memory. Informatik Spektrum, 21 (1998) 4, pp. 213214. [2] Ackerman, M. S.: Answer Garden: A Tool for Growing Organizational Memory. Sloan School of Management, Cambridge, MA, 1994. [3] Ackerman, M. S.; Malone, T. W.: Answer Garden: A Tool for Growing Organizational Memory. Proceedings of the ACM Conference on Office Information Systems 1990, 31–39. [4] Amelingmeyer, J.; Strahringer, S.: Expertensysteme als Werkzeuge fur das Wissensmanagement. HMD, 36 (1999) 208, pp. 80–92. [5] Burke, R.; Hammond, K.; Kulyukin, V.; Lytinen, S.; Tomuro, N.; Schoenberg, S.: Question Answering from Frequently Asked Question Files: Experiences with the FAQ Finder System. AI Magazine, 18 (1997) 2, pp. 57–66. [6] Blasius, K.-H.; Burckert, H.-J. (ed.): Deduktionssysteme. 2nd edition, Oldenbourg, 1992. [7] Booch, G.: Objektorientierte Analyse und Design. Bonn 1994. [8] Borghoff, U.; Pareschi, R. (ed.): Information Technology for Knowledge Management. Berlin et al. 1998.
224
[9]
A-W. Scheer el al. /Cooperative Organizational Memories
Brynjolfsson, E.: The Productivity Paradox of Information Technology. Communications of the ACM, 36(1993) 12, pp. 61–77. [10] Buzan, T.; Buzan, B.: The Mind Map Book: How to Use Radiant Thinking to Maximize Your Brain's Untapped Potential. New York 1996. [11] Chrapary, H.; Rosenow-Schreiner, E.; Waldhor, K.: Das Elektronische Organisationshandbuch. In: Lutze, R.; Kohl, A. (ed.): Wissensbasierte Systeme im Buro. Mumchen 1991, pp. 295–312. [ 12] Conklin, J.: Hypertext: An Introduction and Survey. IEEE Computer Magazine, 20 (1987) 9, pp. 17–41. [13] Conklin, J.; Yourdon, E.: Groupware for the New Organization. American Programmer 9 (1993), pp. 38. [14] Crowston, K.: A coordination theory approach to organizational process redesign. Organization Science 8(1997)2. [15] CSC Ploenzke AG, infas GmbH (ed.): Auf dem Sprung in die Wissensgesellschaft - Chancen und Risiken angewandter IT fur die Arbeitswelt von heute. Kiedrich 2000, URL: http://www. cscploenzke.de/de/study/20001025_studie_wissensgesell/pdf/wissensgesell.pdf. [16] Davenport, T.: Process Innovation: Reengineering Work through Information Technology. Boston MA 1993. [17] Davenport, T.; De Long, D.; Beers, M.: Successful Knowledge Management Projects. Sloan Management Review, 39 (1998) 2, pp. 43-57. [18] Flade-Ruf, U.: Data Warehouse - nicht nur etwas fur GroBunternehmen. In: Hannig, U. (ed.): Data Warehouse und Management - Informationssysteme, Markt, Technik, Anwendungen. Stuttgart 1996, pp. 25-31. [19] Fowler, M.; Scott, K.: UML konzentriert: die neue Standardmodellierungssprache anwenden. Bonn 1998. [20] Gaines, B. R.; Rappaport, A.; Shaw, M. L. G.: Combining paradigms in knowledge engineering. Data & Knowledge Engineering, 9 (1992) 1, pp. 1–18. [21] Gaines, B.; Shaw, M.: Concept Maps as Hypermedia Components, URL: http://ksi.cpsc.ucalgary.ca/articles/ConceptMaps/CM.html. [22] Gentsch, P.: Business Intelligence: Aus Daten systematisch Wissen entwickeln. In: Scheer, A.-W. (ed.): Electronic Business und Knowledge Management - Neue Dimensionen fur den Unternehmungserfolg. Heidelberg 1999, pp. 167–195. [23] Gloor, P.; Streitz, N.: Hypertext und Hypermedia: Von theoretischen Konzepten zur praktischen Anwendung. Berlin et al. 1990. [24] Habermann, F.: Organisational-Memory-Systeme fur das Management von Geschaftsprozesswissen. doctoral thesis, Saarland University, Saarbruecken 2000. [25] Habermann, F.; Wargitsch, C.: IMPACT - Workflow Management System for Coordinated Process Improvements (German), Proceeding of the D-CSCW'98, Stuttgart 1998, pp. 65–78. [26] IDS Scheer AG (ed.): Erfolgreiche Umsetzung von mySAP.com-Ldsungen, ARIS for mySAP.com. Version 5.0. White Paper July 2000, URL: http://www.ids-scheer.com/sixcms_uploao/media/35/aris for_mysapcom_dt.pdf. [27] Inmon, W. H.: Building the Data Warehouse. New York 1992. [28] Inmon, W. H.: Data Warehouse and Data Mining. The On-Line Executive Journal for Data-Intensive Decision Support, 1 (1997) 1, URL: http://www.tgc.com/dsstar/971007/100001.html. [29] Krcmar, H.: Informationsmanagement. Berlin et al. 1997. [30] Lehner, F.: Organisational Memory, Konzepte und Systeme fur das organisatorische Lernen und das Wissensmanagement. Mflnchen, Wien 2000. [31] Lehner, F.: Organisational Memory-Systeme - Eine Dimension der Informationssystem- und Organisationsentwicklung. In: Hummeltenberg, W. (ed.): Information Management for Business and Competitive Intelligence and Excellence, Proceedings der Fruhjahrstagung Wirtschaftsinformatik'98. Braunschweig, Wiesbaden 1998, pp. 43-54. [32] Malone, T. W. et al.: Tools for inventing organizations: Toward a handbook of organizational processes. MIT Center for Coordination Science, 21 st Century Initiative Working Paper No. 5, 1998, URL http://ccs.mit.edu/21c/mgtsci/index.html. [33] Malone, T. W.; Crowston, K.; Lee, J.; Pentland, B.; Dellarocas, C.; Wyner, G.; Quimby, J.; Osborn, C. S.; Bernstein, A.; Herman, G.; Klein, M.; O'Donnell, E.: Tools for inventing organizations: toward a handbook of organizational processes. Management Science, 45 (1999) 3, pp. 425–443. [34] Oberschulte, H.: Organisatorische Intelligenz- Ein Vorschlag zur Konzeptdifferenzierung. In: SchreyOgg, G.; Conrad, P. (ed.): Managementforschung 6. Berlin, New York 1996, pp. 41–81. [35] Ovum Ltd. (ed.): Knowledge Management: Applications, Markets and Technologies. Analyst report, London 1998. [36] Pascale, R.; Athos, A.: The Art of Japanese management. Harmondsworth 1981. [37] Pree. W.: Komponentenbasierte Softwareentwicklung mil Frameworks. Heidelberg 1997.
A.-W. Scheer et al. / Cooperative Organizational Memories
225
[38] Probst, G.; Raub, S.; Romhardt, K.: Wissen managen: Wie Unternehmen ihre wertvollste Ressource optimal nutzen. Frankfurt am Main 1997. [39] Rational Software Corporation, Inc. (ed.): OMG Unified Modeling Language Specification, Version 1.3. June 1999. [40] Resnick, P.; Varian, H. R.: Recommender systems. Communications of the ACM, 40 (1997) 3, pp. 5658. [41] Rolles, R. et al.: Design of Improvement Processes: Workflow Systems in Learning Environments (German). In: A.-W. Scheer and M. Nuttgens (eds.) Electronic Business Engineering, Heidelberg 1998, pp. 725-743. [42] Ruggles, R.: Tool Time: The Next Generation. In: Ernst & Young LLP (ed.): Center for Business Innovation, March 12th 1996, URL: http://www.businessinnovation.ey.com/mko/html/know_org.html. [43] Scheer, A.-W.: Data Warehouse und Data Mining: Konzepte der Entscheidungsunterstiitzung. IM Information Management, 11 (1996) 1, pp. 74-75. [44] Scheer, A.-W.: ARIS - Vom GeschaftsprozeB zum Anwendungssystem, 3rd edition. Berlin et al. 1998. [45] Scheer, A.-W.; Erbach, F.; Thomas, O.: E-Business - Wer geht? Wer bleibt? Wer kommt? In: Scheer, A.-W. (ed.): E-Business - Wer geht? Wer bleibt? Wer kommt? 21. Saarbrtlcker Arbeitstagung 2000 fur Industrie, Dienstleistung und Verwaltung. Heidelberg 2000, pp. 3–45. [46] Scheer, A.-W.; Habermann, F.: Electronic Business: Vom ,Why now?' zum ,Let's Go!'. IM Information Management & Consulting, 14 (1999), pp. 7-14. [47] Shum, S. B.: Negotiating the Construction and Reconstruction of Organisational Memories. Journal of Universal Computer Science 8 (1997) 3, Special Issue on Information Technology for Knowledge Management, pp. 899-928. [48] Stein, E.: Organizational Memory: Review of Concepts and Recommendations for Management. International Journal of Information Management, 15 (1995) 2, pp. 17–32. [49] Stein, E.; Zwass, V.: Actualizing Organizational Memory with Information Systems. Information Systems Research, 6 (1995) 2, pp. 85–117. [50] Urban, C.: Das Vorschlagswesen und seine Weiterentwicklung zum europaischen KAIZEN. 2nd edition, Konstanz 1994. [51] Wargitsch, C.: Bin Beitrag zur Integration von Workflow- und Wissensmanagement unter besonderer Berucksichtigung komplexer Geschaftsprozesse. Doctoral thesis, Friedrich-Alexander-UniversitSt. Erlangen-NUrnberg 1998. [52] Warschat, J.; Ribas, M.; Ohlhausen, P.: Wissensbasierte Informationssysteme zur Unterstutzung wissensintensiver Prozesse in Unternehmen. HMD Praxis der Wirtschaftsinformatik, 36 (1999) 208, pp. 53-58. [53] Walsh, J., Ungson, G.: Organizational Memory. The Academy of Management Review, 16 (1991) 1, pp. 57-91. [54] Wess, S.: Fallbasiertes SchlieBen in Deutschland - Eine Ubersicht. KI Kunstliche Intelligenz, 6 (1992) 4, pp. 46-51. [55] Wiig, K.: Managing knowledge: A survey of executives perspectives. Arlington 1989.
226
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press. 2002
Towards a "Knowledge-Based Marketplace" model (KBM) for cooperation between agents Jean-Pierre Cahier* - Manuel Zacklad* *Laboratoire Technologie de la Cooperation pour 1'Innovation et le Changement Organisationnel (Tech-CICO) Universite de Technologie de Troyes (UTT) 12, rue Marie Curie - BP 2060 10010 Troyes cedex http://w3-tech.univ-troves.fr E-mail: {jean-pierre.cahier,manuel.zacklad} @utt.fr Abstract: E-Marketplaces are not usually studied as places for cooperative work between suppliers and buyers, involving knowledge and creation of new knowledge. The present work starts from a theoretical point of view, in order to build a model of cooperation that we have entitled "Knowledge-Based Marketplace" (KBM). e-Marketplaces catalogues proceed from a twofold problem of modelling information and knowledge from multiple points of view and from multiple experts. Buyers and sellers each bring complementary expertises to "co-construct" the catalogues. As a first step, in this paper, we justify and start to develop that model, by studying it through an experiments program of a concrete case of KBM, with a particular interest in the possibilities for new knowledge to emerge from the collaboration processes.
Keywords: Cooperative Work. Knowledge-Based Marketplace (KBM)
1 Introduction 1.1 Objectives Our work takes place as a first step of a long-term project, focusing on the theoretical study of a cooperative model of a "Knowledge-Based Marketplace" (KBM). This model links human agents who are suppliers and buyers of resources, in the wide sense. The cognitive activity of these agents is particularly considered according to the resources they are proposing or seeking. The model especially focuses on the cognitive aspects of design and exploitation of resources catalogues by designers and users. Resources, including associated concepts and descriptive attributes, are simultaneously i) the objects of the cooperative work and ii) the points of reference of agents' activity, in terms of knowledge and language. This generic model is used in order to build "Knowledge-Based Marketplace" (KBMs) for concrete applications and domains. Such a KBM allows cooperation between agents exchanging resources through a catalogue with description possibilities from multiple points of view. In a KBM, agents will be allowed to modify resources description and classification, either in order to differentiate merchants' offerings (sellers) or to facilitate
.
J.-P. Cahier and M. Zacklad / Towards a "Knowledge-Based Marketplace " Model
227
their research (buyers). Cooperative activity focuses on both description and transaction systems. In our context, e-marketplaces catalogues proceed from a twofold problem of modelling information and knowledge from multiple points of view and from multiple experts. Buyers and sellers each bring complementary experts' views to "co-construct" the catalogues. We also focus on the nature of knowledge used by agents in these cooperations, with a particular interest in the possibilities for new knowledge to emerge from the collaboration processes. To state the model more precisely, our approach consists of operating it on an example, based on an e-marketplace case in a precise domain (for selling training services to the computer community). In that case, we present a laboratory apparatus, allowing experimental study of cooperating human agents. 1.2 Structure of the paper Part 2 first brings elements to justify and specify the model, and look at it through several fields, including Computer Supported Cooperative Work (CSCW) and Knowledge Engineering. Cooperative work in a KBM depends strongly on many problems of ontology sharing. Part 3 describes our method, based on several steps respectively dedicated to modeling and experimenting the KBM we take as an example. Considering the « work in progress » status of our research, we make a distinction between achieved steps and ongoing steps, in our program for studying several cooperation approaches on the KBM. The experimental part involves participation of student groups inter-acting as buyers with KBM for completing various orders. So we are trying to validate a first hypothesis, based on the possibility of creating and sharing knowledge from cognitive agents cooperation in a KBM context. Part 4 describes the KBM taken as an example, and the data included. That KBM for a modelling and experimental purpose addresses the field of training products and services. It is derived from an on-line catalogue proposing syndication of many French training suppliers in the fields of office automation, information systems and networks (1200 training suppliers, 500 training topics, 18 000 "training module" products). Part 5 resumes the first results at the present step of the study. They concern the semiformal modelling of the KBM based on the concept of «training module ». This modelling approach is based on a domain expertise, through the analysis of the talk of a buyer, in simple cases of interaction between the buyer and the products board. These results are used to complete the experimental apparatus designed to test the first collaborative hypothesis we advance concerning the KBM model. According to that hypothesis, performance and relevance of resource mining by user should be enhanced on a KBM, by the user ability of creating heuristic attributes in a « continue capitalization mode, and sharing them asynchronously with other users. Finally, part 6 presents temporary conclusions and perspective elements from these first results. 2 Backgrounds and characteristics of the KBM model The theoretical "Knowledge-Based Marketplace" model (KBM) that we study is a generic model, exceeding the purely economic application domains. In contrast it is a complex system justifying a strong interdisciplinary aperture [1]. Among the many matters and scopes that could be involved that for, we note the sociological approaches of knowledge management [2] and the cognitive economy [3], with which gateways could be established. We have chosen to focus on Computer Supported Cooperative Work (CSCW), Knowledge
228
J.-P. Cahier and M, Zticklcid/Towards a " Knowledge-Based Marketplace" Model
Engineering and Multi-Agent Systems.These three matters have just begun to take into account e-marketplaces, i.e. from a cooperation point of view, to manage knowledge at the extended enterprise level [4]. Multi-agent systems are used for modelling and simulating marketplaces from the point of view of economics [5] or of the supply chain management [6], but not from a generic collaborative and cognitive perspective. Among the backgrounds, motivations towards a KBM model are strongly stressed, in the scope of Management Sciences, by the numerous problems associated with the development of B2B and B2C e-commerce techniques [7] [8] [9] [10]. In particular, for emarketplaces, these trends stress a lot of aspects such as dynamic content services, consumer and retailer shop bots, shopping and notification agents, one-to-one services... Offer value is more based on singularity, reactivity and client profiling, with positive feedback loop exploration [11] in the context of internet-based Customer Relationship Management (« e-CRM »). New resulting conditions for buying make the price more reactive and relative [10]. Detailed knowledge of products, quality of service, delays, degree of personalization are often more important than price to determine the added value brought by a seller [12]. The KBM approach could also help to bring an answer to issues encountered by emarketplaces dealing with a very large number of products, when those products present a large diversity of trade names and characteristics. A lot of designers of those emarketplaces need to improve the solutions allowing users to collaborate more efficiently, enhancing the product research modes from both parametric [11] and semantic points of view. In their requests users wish to associate arguments concerning concepts, domain knowledge and more trivial data (dates, prices, labels of quality, places...). The KBM approach seems to fit the strong needs of the industry. 2.1 The basic cooperations of the model The KBM model can be applied as soon as different actors need an inscription and exchange environment in order to propose resources they are owning or producing, or to find resources that they need, with an activity of choice among comparable resources. The more similar the resources are, the more important it is that agents can use precise knowledge to differentiate them. An efficient support to cooperation, taking into account cognitive aspects, is more necessary in the commercial field, because cooperative work concerning the resources occurs between actors who have interests and strategies that can be strongly complementary or opposite, even in a drastic manner. These conflict aspects are inherent characteristics of cooperation that a KBM has to deal with. Actors integrate in their strategies the communication about the needs and the resources offered, whose shared property is being very particular and detailed, considering knowledge and language stakes as weapons in the practice of buying and selling. The minimum KBM model includes each combination between the three elements involved: the buyer (s), the seller (s) and the resources organized on the marketplace inscription support (product board, buying or selling portal). Considering only the collaborations that refer to the exchanged thing, all configurations (1–1)and (l,n) from the actors may be involved, as seen in the following illustration (Tab.l). So the KBM model emphasizes processes and uses surrounding the basic marketplace information system.
J.-P. CahierandM. Zacklad/Towards a "Knowledge-Based Marketplace"
Model
229
Tab.l - Cooperation cases between agents in a KBM
topics involved at the present step of the study
Processes and collaborations on a KBM (« Knowledge-Based Marketplace )
Examples of collaborative processes on the KBM chosen as an example, in the field of training . A buyer analyses the customized training cursus he needs, according to the skills that he already owns. He compares that analysis with the proposed training modules, their Confrontation between Buyer and the products scheduling and prerequisites, etc. board . A buyer chooses a learning module or a sequence of leaning modules . In order to choose, a buyer can re-use classes and criteres resulting from previous activity of other buyers with a similar or near profile Collaboration between Buyer and Buyer (s) . Before or during the training course, through an e-mail or chat service, a buyer reads the through the products board comments of previous clients of the training sellers. During the training action (i.e. in elearring context) buyers or participants can also exchange information with other buyers or participants in the same trainign module (in order to refine or tune the cursus, to exchange documents, etc.)
Confrontation between Seller and the products . The author or ttie seller of a training object registers and describes it on a shared device board (according to either an unified terminology or a more sophisticated approach accepting ontology sharing) . A seller uses the products board for analysing his competitors' offerings and situating his own product offering . A seller organizes cooperation with third parties, i.e. certification authorities, free-lance Collaboration between Seller and Seller (s] trainers, in order to harmonize their common offerings or product vocabulary in the through the products boarc catalogue . Groupware participation of the seller into a validation circuit (i.e. : modifying of consensual knowledge or ontologies, vertical standardization workgoups...) .Thebuyersendsageneralrequesftowardsa panel training sellers selected through the KBM) . "Push" of information or e-learning contents towards users who have subscribe to certain topics . On-line helpdesk, information feedback loop in order to improve quality of training services, e-CRM, tning between customer and seller ontologies Collaboration between Buyer (s) and Seller (s] through the products board . Help to explicit and memorize arguments (i.e. Questions, Options, Criteres) exchanged between buyers and sellers about resources: attachment of remarks or arguments into particular attribute or issues of training modules .negodation about attributes of a learning object (content, date, teacher, options...) . Collaboration through the products board between actors, who are both either simultaneously proposing or searching for training : training in the frame of projects, knowledge-sharing networks.
We have included in the table, in front of the processes which make up the model, the concrete collaborative forms in the KBM case studied as an example, according to the methodology detailed in the following section (see section 3). (As we will explain, our method proceeds by steps, because collaboration modes proposed in Tab.l are numerous and complex. In order to successfully implement them in the context of a KBM, the proposed method begins with the deep analysis of a simple case - i.e. a buyer facing the products board. Even if that case doesn't directly address cooperative work, it constitutes a step that can provide a better knowledge of the cognitive activity of the buyer faced with a product board. This step is necessary to build more sophisticated cooperation cases, such as knowledge transmission « from buyer to buyer » through the products board, a case that we shall present in section 5). 2.2 Conflicts between ontologies in cooperative work on a KBM An advantage provided by a KBM could be the giving of adapted tools to user, in order to deal with a very big quantity of concepts and properties used to describe competing resources. In addition, concepts corresponding to products (i.e. the «learning module » concept in our example of KBM) are involved in several processes and roles, and used by numerous agents carrying an unlimited variety of points of view. The success of the agents
230
J.-P Cahier and M. Zacklad / Towards a "Knowledge-Based Marketplace " Model
(i.e., for a buyer: finding the best product) depends largely on the integration of many ontological systems or on sharing problems, taking into account the taxonomies and the vocabulary that agents use in different contexts. This problem of interoperability is for example studied, in the field of Computer Supported Cooperative Work (CSCW), by Carla Simone [13][14][15], who explores the metaphor of producing/consuming knowledge in distributed systems in a new sense. This author studies how, in the articulation work devoted to activity coordination, contexts can be conciliated at a level dealing with ontologies, knowledge and business rules. It is possible to distinguish terminology conflicts (homonymy, synonymy...), categorization conflicts (various senses for descriptive concepts) and conflicts between types. Top-down approaches, based on an unique repository and using abstraction, contrast with approaches based on observation of a more complete reality, from multiple and partial points of view mixing various levels of detail and using abstraction in a « continue capitalization)) mode. In that last case, ontology sharing is progressively built in the cooperation process: that seems us to be the more attractive scenario for a KBM. These works and many other ones, including research's works in the fields of knowledge engineering, multi-agent systems and information systems [16], [17], [18], [19], [20], [21], contribute indirectly to highlight a central characteristic of the KBM model: most emarketplaces are open systems, which do accept (even research) a plurality of beliefs and naming practices aiming at the resources in play, including their identifying and qualifying properties. Dissonances do necessarily appear between ontologies used by interacting agents, and context problems in the cooperative work. That is a recurrent issue that occurs in all the attempts to articulate activities of cognitive agents who have to synchronise their activities at semantic and pragmatic levels. For the emarketplaces, this problem concerns all the modalities listed in Table 1, starting with the more basic cases we have chosen to focus on during the first steps of our experimentation. That openness of the KBM model means that a unique actor, i.e. as ontology administrator, cannot really seek to impose a unique point of view, or even a set of pre-established points of view that could be sufficient. In contrast we are faced with both multi-points of view and multi-experts issues [22], as described in the knowledge engineering field. 3 Methodology We have designed and we are constructing an experimental apparatus involving several steps (Fig 1), in order to implement a set of demonstrations, modelling and experiences associating real users in a KBM. Our reference to CSCW (Computer Supported Cooperative Work) implies an important focus on collaboration between real actors in working-like situations. In the present phase of the work, steps 6 and 7 are in progress, and we focus in this paper on steps 3, 5 and 6. Globally the part in progress « Experimentation of the KBM» aims at a better understanding of individual and collective cognitive activities supporting a KBM. That phase was prepared by a semi-formal modelling step, in order to clarify some classification keys, to guide the building of buying portals fitting into the experimental goals.
J.-P. Cahier and M. Zacklad / Towards a "Knowledge-Based Marketplace " Model
231
The sequence of experiments aims to involve several collaboration modalities listed Tab. 1 in a progressive manner. Steps 5 to 8 focus on the situation of buyers faced with the KBM. At every step, the products board plays the core role, as the pivot of collaborating and cognitive activities in the KBM. The products board presents the advantage of allowing both relational and epistemic dimensions [23] of a KBM to be explored. In helping the user to compare products, to collect information or to consult other users' advice, a KBM allows this user for example to estimate the influence of his seller's credibility (relational dimension) or the influence of products description criteria (epistemic dimension). 3.1 Modelling During the semi-formal modelling phase (step 5 in Fig.l) that prepares and leads up to experiments, the main object studied is the product. In that modelling, according to cognitive psychology, we consider the concept of the product, carrying the representations in the mind of the user faced with the product. At the same time, on a more objective manner, the product is modelled referring to its concrete inscription environment on KBM (inside the products board). That preparatory modelling step, whose main results are summarized in section 5, unifies these two aspects. In order to achieve that modelling, we have first studied a domain expertise from a cognitive point of view, through the analysis of the talk of an expert (buyer). We have studied simple cases of interaction between the buyer and the products board, this expert being faced with different orders to the products board of the KBM. In the context of this paper, one of the authors who knows the domain has played the role of the expert. We have used the notion of point of view, in order to specify the many incomplete perspectives that the expert applies to the concept and its properties. The expert uses those points of view in order to describe his buying activity, or suggest to use them to classify the concepts corresponding to products, in general situations or in the context of various buying orders. 3.2 Standard and heuristic attributes In the method that we have followed, what the expert said was analysed in order to highlight several potential classification systems for a same concept, according to the point of view that expert chooses or recommends. So each point of view addresses a set of classes allowing to classification of all or a part of considered concepts.
232
./.-P. Cahier and M. Zacklad / Towards a "Knowledge-Based Marketplace " Model
In the definition of a product, we make a distinction between standard attributes and heuristic attributes. The first category corresponds to basic properties of the product (i.e. for a "training module" product: its date, its name, its duration, its price). A second category of heuristic attributes corresponds to properties that are more dependent on domain knowledge and have a more « subjective» connotation (the goal of the training, the expected skills, the professions of the trained people, the quality of service...). Starting from similar values for the same attribute, one can group products into different classes. Each point of view is based either on standard or heuristic attributes, whose values allow the classes to be defined. In a first approach, we have modelled the product from multi-points of view but monoexpert mode. We then enhance the scope to a multi-experts mode. As a matter of fact, in the following of the laboratory experiment (step 6 to 11, see Fig. 1), we consider the users of the KBM as different experts, each with the ability of creating new product classes. using heuristic attributes representing their specific user's needs. 3.3 Laboratory experiment For the sequence of experiments (step 6 to 11, see Fig. 1), the general idea consists in mobilizing different student groups, as users of the KBM. Students interact through the system with the products board, to complete different orders. The performance of their activity and the relevance of their results are measured. The experiment makes various collaboration, registration and classification devices to link users and the products board concepts of KBM (see i.e. [24] for this type of experiment). Especially, at a first step of the experiment (step 6, see Fig.l), this applied frame will allow the creating of "heuristic attributes" by users and the asynchronous sharing of the resulting heuristic classes on the experimental apparatus to be tested. So we could evaluate the hypothesis of making knowledge emerge in a « continue capitalization» on a KBM. 4 The experimental KBM The building of a KBM, as a model and a laboratory experiment, has started from a realistic operational data set. Our example is situated in the field of services emarketplaces, and is derived from a commercial base syndicating offerings of many competitors proposing training modules for French-speaking computer people and office automation end-users. The following board (Tab.2) gives some examples of the concepts of a « training module » with some of the attributes used to describe them, as those concepts appear after preparation in the focused KBM. The source, considering the initial syndicated catalogue, was voluminous (600 key-words, 500 official training themes in a 4-stages hierarchy, 1200 training organisms of all sizes, 18 000 registered training modules). It was necessary to extract and fit up a more usable set of information into the KM. In the next point of this paper, modelling and experiments deal with a subset of a few hundreds of training modules, from several organisations. Themes are also limited, in the field of computer people training in topics such as Corba, UML, XML and Java. Attributes listed on Tab.2 are price per day, duration, mean delay before next occurrence of the training module, name and location of the learning seller, certification labels at organisation or module levels, etc.
J.-P. Cahier and M. Zacklad / Towards a "Knowledge-Based Marketplace" Model
233
Tab.2 Some products inside the studied KBM, with some of their standard attributes (Example of request under keyword « Java »)
It must be noted that a current activity, in the KBM seems to be generally very sensitive to buyer's profile and to his context. It is clear for activities such as identifying a set of training modules, which sufficiently fit in terms of topic and other attributes contributing to a thematic tuning. Therefore a too trivial request approach, i.e. in terms of either classical text mining techniques or pure RDBMS techniques (Relational data base management systems), cannot usually answer a real world user's request efficiently. For example, even if the buyer is searching for a training module precisely in order «to develop B2B applications in Java», clearly he will not mechanically evaluate only the unique listed module whose name however seems to perfectly fit his need. 4.1 Knowledge creating and sharing in choice activity The modelling carried out in analysing the activity of a buyer faced with the products board, allowed several phenomena to be highlighted, which we choose to use in order to experiment the creating and sharing of heuristic attributes by buyers in the KBM.
234
J. -P. Cahier and M. Zacklad / Towards a "Knowledge-Based Marketplace " Model
Fig. 2 Choosing structure for the concept« Learning Module », as indicated by the expert
In his act of choosing, the buyer considers the concept of product according to several points of view. A indicated by the expert, the choosing structure of the « Learning Module » concept leads to the creation (Fig. 2) of 4 heuristic attributes, corresponding respectively to the emphasis placed on the criteria of thematic, of calendar (dates and duration), of cost or of quality and reputation (product and seller).
Fig.3 Classes corresponding to points of view « by learned notions » and « by project's goal », as highlighted by the expert
The expert considers in turn the point of view by theme according to 4 other points of view. These were defined extensively by the domain expert in looking at 50 modules for learning to Java language (numbered from 1 to 50) inside a subset of the studied sample: The expert has distinguished 2 classes corresponding to his points of view « by learned notions » and « by project goal» (Fig.5). In a similar manner, the expert has classified every learning module of the sample of the KBM according perspectives respectively "by tool" and "by profession". These 4 perspectives were then considered as interesting in order to help to classify learning modules aimed at computer people more generally.
J.-P. Cahier and M. Zacklad / Towards a "Knowledge-Based Marketplace " Model
235
On the other hand, considering the scheduling of his activity of selection, as a first step the buyer appears to search, in the great majority of cases, from both the points of view of « Theme » and « Calendar ». He doesn't focus at that phase on the points of view of "Cost" and "Quality/reputation" which will interest him only later. So we note a two-phase system, with a first phase that we have called « elimination » (conducting towards an "intermediate selection" from 5 to 10 products) and a second phase entitled « refining ».
Fig. 4 Activity diagram of the buyer, with segmentation into 2 phases
In the first phase of "elimination" the will to identify a group of training modules that fit well is preponderant. In contrast, in the second phase of "refining", buyers research firstly of all to optimise the choice between multiples criteria [25]. We have represented that division into two phases in the buyer's activity diagram (Fig. 4). We have advanced the hypothesis that such a division could be generalized to a important subset of purchase order processing - excluding cases such as searching failure or discovery of new elements, for which it could be necessary to return to the "elimination" phase (see feed-back loop, on Fig.4). That hypothesis includes the existence of an intermediate selection board that then guides us towards designing an experiment, which could benefit from that particular situation. Since in numerous cases the buyer at the end of the "elimination" phase uses an «intermediate board » representing a subset near his need from a product identification point of view, it is a solution to create a new point of view, corresponding to a heuristic attribute of this particular user. For example, inside a sample of 50 Java training modules that he consults, a buyer has by several means selected an «intermediate board » of 5 modules, which respond to his particular point of view. This particular point of view could represent a synthesis between a point of view by a predefined topic («Initiation for Java developer ») and the will to emphasize certified modules and cursus (« labels », as standard attributes). This resulting « intermediate board » corresponds to a heuristic attribute created by this particular buyer. It represents a new knowledge, created by this user, that could be interesting to capture, to name and to memorize in a "continue capitalization" mode, so as the buyer could find later that selection again, or to share with other users in a cooperation dimension. Experimenting the sharing of « heuristic attributes » between multiples experts The experiment's apparatus then becomes as shown on Fig.5: in a first time («initial step »), multiples points of view proposed by the expert, such as theme search « by learned
236
J.-P. Cahier and M. Zacklad/Towards a "Knowledge-BasedMarketplace"
Model
notion» or « by project goal » (see Fig.2) allows a classifying device to be set on a first "initial" on-line e-marketplace (SITE-1) to search and select on the KBM. With that device, a first buyer (Buyer 1) or a group of buyers shall apply a set of research orders. For each buyer, the intermediate board at the end of the "elimination" phase is saved, semantically qualified by the request of the user, named and considered as a new point of view resulting from an « heuristic» attribute. Integrating these new classes to aid selection. an « enhanced » classifying device (SITE-2) is then built. That modified e-market place will be tested, with the same orders and with another set of orders, by a second group of buyers (Buyer 2).
The possible gain in performance is measured by comparing buyers of 1st and 2nd groups (dealing with similar orders), and also, with another set of orders, between subjects using the "initial" or the "enhanced" device. So the experiment could allow to best performance and relevance of user selection activity to be verified, in the case of dynamic introduction of heuristic attributes, especially after several iterations.
6 Conclusion At the date of drafting this paper, the program of experiments is still going on, as indicated on Fig. 1. The modelling we have carried out has firstly allowed outlining of a multi-point of view and mono-expert navigation structure. It is useful for building, at a later stage, emarketplace prototypes that could include classification devices using explicit domain knowledge resulting from cooperative work. That is the case, for the experiment that we are now conducting, to create and asynchronously share heuristic attributes, and further to experiment the other facets of the cooperation model we have begun to describe for a Knowledge-Based Marketplace. So we can switch into a dimension of multiple points of
J.-P. Cahier and M. Zacklad / Towards a "Knowledge-Based Marketplace " Model
237
views and of multiple expertise [26], these multiple experts being the actors directly involved in KBM collaborations. Extension of these experiments needs a more formal level of modelling, with the help of a knowledge representation language (based on conceptual graphs). If it was confirmed, progress for realizing a KBM in a domain could also be useful to progressively clarify the characteristics of the KBM model, with a cooperation paradigm inside. KBMs could bring a dimension of generated and shared knowledge, stimulating new types of economic exchanges and product innovation. References [1] Intelligence de l'Economie et Economic de la Cognition, Bourgine P., Le Moigne J.-L., Introduction/preface a la seconde conference "L'Economique et L'Intelligence Artificielle" (CECOIA II), Paris 1990, ISBN 2-9036780-3 [2] Poitou J.-P., La dimension collective de la gestion des connaissances: cadrage theorique, Chapitre 7, p. 149–184, in [27] [3] L'economie Cognitive, Walliser B., editions Odile Jacob, 2000 [4] Boughzala I., Zacklad M., Matta M., Ingenierie de la cooperation pour gerer les connaissances dans une entreprise etendue, 1C 2001 [5] Sandholm T.W., Multiagent systems. A Modern Approach to Distributed Artificial Intelligence, in Distributed rational Decision Making, p.201–258, The MIT Press [6] Kozlak J., Nawarecki E., Demazeau Y., Koning J.-L., Simulation multi-agents basse sur les mecanismes de marche pour la distribution des ressources, Actes MOSIM'Ol, Troyes, 25–27 Avril 2001 [7] Nieuwbourg P., D'Hondt H., Places de marche sur Internet, nouvelles regies pour le commerce du XXIeme siecle, Editions BNTP www.BNTP.COM, www.PlacesDeMarche.net, ISBN 2–9515654–0–2 [8] David Roddy, Leo Obrst & Adam Cheyer, Competing in the Evolving Landscape of B2B Digital Markets B2B bots, interoperability and ontologies , http://www.tradeum.com/pages/frset-3forum_white.html [9] Brynjolfsson E., Smith M.D., The Great Equalizer ? Consumer Choice Behavior at Internet Shopbots, MIT Sloan School of Management, Cambridge, Ma, July 2000, (document MIT on line, section economie) [10] Pensel J.-L., Les places de marche virtuelles, 1'acces a de nouveaux outils transactionnels, Actes du 6eme colloque de 1'AIM, pages 97–110, Nantes, 7 au 9 juin 2001 [II] interview de Sherif Danish(societeSaqqara), par J-P.Cahier, « Gestion de catalogues Web: L'avenir du B-to-B est chez le fournisseur", in "Le Monde informatique", n°900, 15 juin 2001, pages 32-33, http://www.weblmi.com/techno/2001/900%5F32%5FgestiondecataloOO.htm [12] Morrison D., Wise R., Beyond the exchange, the Future of B2B, Harvard Bus. Rev., Nov-Dec.2000 [13] Simone C., Giubbilei D., Mark G., Interoperability as a Means of Articulation Work, Proc. of Conf. on Work Activities Coordination and Collaboration (WACC'99), ACM Press, Feb.22–25, 1999, San Francisco [14] Simone C., Unifying or reconciling when constructing Organisational Memory? Some Open Issue, extended version of ECAI2000 Workshop on KM/OM [15] Schmidt K., Simone C., Coordination Mechanisms: Towards a conceptual foundation for CSCW system design, Computer Supported Cooperative Work (CSCW) An International Journal, vol.5, n°2–3, 1996 [16] Dieng R., Comparison of Conceptual Graphs for Modelling Knowledge of Multiple Experts: Application to Traffic Accident Analysis, Rapport de recherche INRIA N°3161, Avril 1997 [17] Castelfranchi C., Conflicts Ontology, Computational Conflicts : conflict modelling for distributed intelligent system , Muller H.J., Dieng R., Springer, (2000) [18] Sadek D., "Dialogue et cooperation" , in "Cooperation et Conception", sous la direction de G. de Terssac & E.Friedberg, ed. Octares, 1996 [19] Hurwitz S.M., Interoperable Infrastructures for Distributed Electronic Commerce , http://www.ontology.org/main/papers/csc-ont-eng.html et http://www.atp.nist.gov/atp/98wp-ecc.htm [20] Euzenat J., Building consensual knowledge bases: context and architecture, dans Nicolaas Mars (ed.), Towards very large knowledge bases (actes 2nd international conference on building and sharing very largescale knowledge bases (KBKS), Enschede (NL), (10-13 avril) 1995) IOS press, Amsterdam (NL), pp!43155,1995 [21] Prince V., Vers une informatique cognitive dans les organisations, Le role central du langage, Ed. Masson, Paris, 1996 [22] Ribiere M., Representation et gestion de multiples points de vue dans le formalisme des graphes conceptuels, These de Doctorat de l'Universite de Nice - Sophia Antipolis, Paris, avril 1999
238
J.-P. Cahie r and M. Zacklad / Towards a " Knowledge-Based Marketplace" Model
[23] Zacklad M., « La Theroie des transactions intellectuelles, une approche gestionnaire et cognitive pour le traitement du COS »,, Intellectica, vol.30, n° 1, 2000 [24] Lewkowicz M., Zacklad M., "Analyse cognitive des impacts de 1'utilisation d'un collecticiel base sur une methode de resolution de problemes", in Actes IC-2001, pp-215–232, PUG.Grenoble. [25] Roy, B., Methodologie multicritere d'aide a la decision, editions Economica, 1985 [26] Canier J.-P., Zacklad M, , « Approche cognitive de la co-construction d'un catalogue multi-points de vue et multi-experts », Actes EGC 2002, Montpellier du 21 au 23 Janvier 2002, p.221–226, Hermes [27] Zacklad M., Grundstein M; & al, , "Management des Connaissances, Modeles d'entreprise et applications", Editions Hermes Science, 2001
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press, 2002
239
Putting CSCW in a Business Context Implications for Research and Systems Design Per-Arne PERSSON Swedish Army Command, P.O Box 901, S-745 25 Enkoping, Sweden per-ame.persson @ atk.mil.se Abstract Modern management usually means highly controlled organizations, relying on sophisticated control mechanisms. Workflows, large integrated management information systems (MIS), global business processes and network(ed) economies belong to this situation. Research and industries search for improved use of information technology (IT) for the integration of human practices and various kinds of coordinating mechanisms. Some form cooperative systems where computer supported cooperative work (CSCW) is organized. Socially satisfactory use of IT applications may be overshadowed by the contextual business conditions and views on rationality, design, and organizing. CSCW must not be treated as a happening on the way towards profit or an obstacle for profit or efficiency. In order to establish a theoretically sound platform for independent research, avoiding technology-driven business and research development, CSCW has to be understood, possibly reevaluated, and put into context: modern society, its institutions and businesses. CSCW can be consciously designed, applied ad hoc, or a strategy for managing power, meaning and norms. This paper provides a theory background and includes findings from an ethnography, accounts from work design within military management and command work, suggesting research approaches for better understanding of technology and work design principles. Keywords: CSCW, work practice, command work, ethnography, management.
1. Introduction, about IT and Rationality After more than a decade of research on organizing and FT, CSCW as an approach for research and as a perspective on work has gained considerable ground. However, the role of IT in workplaces is still worth attention, ever more so because of the continuing implementation of IT, making IT "natural" and new frontiers for computing taken for granted. Some approaches are related to top-down control interests while others aim at changes in work practices. Unexpected consequences or frictions are common. Berg argues for the importance of empirical research contributions and formulates some of these concerns [1], meaning that it is an important theoretical issue to investigate the generative power of such technologies while circumventing technologically determinist or socially constructivist accounts. It is a challenge to conceptualize what it is such tools do in a relational way: without exclusively attributing this activity to either the tool itself or to the humans working with it. (ibid., p. 373) Moreover, such an understanding, he claims, enriches the de-skilling/empowerment accounts nourished by common perceptions of IT-use. What promotes (as in Berg's case from within medical care) replacement of paper-based practices by computer-based systems? Maybe there are other ways of formulating concerns or research issues. If less paper is the objective change, what more is at stake? Symon et al. mean that studying work coordination in context and the interplay between formal and informal practices can give an adequate understanding of what is really going on [2]. Hayes, too, underlines the need for such research to include contextual
240
P.-A. Persson / Putting CSCW in a Business Context
factors [3]. Thus, researchers have to learn where to look in society and in modem workplaces, and get support for the interpretation of their findings. The question Why (cooperative, IT)? has not yet been exhausted. Few empirically grounded theories accompany consulting and engineering enterprises. Instead various 'IT rationales' are at play in organizations. Brunsson goes further, discussing action rationality and decision rationality respectively, and rational or irrational actions [4]. I would say there are actions (decisions) of several rationalities at hand, each one related to superimposed constructs. The work context in this study, for example, is "Information Society", implying a certain rationality. a specific vision as regards our economic and social future, even possible to see as an ideology distracting us from important issues [5]. The research object is military command work where formal procedures are abundant, but. as we shall se. CSCW has certain roles. 2. Understanding Risk Society, Technology, and Human Action In order to proceed, we have to look at society, technology, and work in a general sense. Bryant and Jary who discuss Anthony Giddens' contributions [6], see three alternatives for control in modern 'risk society', the 'runaway world'. The alternatives are: recover or secure control, fix the big picture; resign oneself to loss, return to the private and personal; or go for limited and local control, there is no big picture. Fix bits and pieces as and when you can. One condition in risk society is the loss of secure foundations for knowledge, including erosion of authorities, at the same time as unconditional (probably) trust in society's abstract systems (for example money) and technology is a prerequisite. Orlikowski [7] who draws on Giddens' theory of structuration elaborates on two aspects of technology: first, the duality of technology, second, that technology is interpretively flexible: actions that constitute the technology are separated from actions constituted by technology. Many technologies are perceived as fixed structures but practice shows that even "black box" technology has to be comprehended and activated by human agency to be effectual: in such interaction usage shape technology and its effects, forming an ongoing balance act: When humans act in organizations, they create and recreate three fundamental elements of social interaction (Giddens. 1976:104): meaning, power and norms, (ibid., p. 64)
These elements are highly interactive and not separable in practice, in work, much of which is involved in this interplay, especially in management and control efforts. Technology thereby constitutes tools, media, input and products. Orlikowski uses concepts such as the design mode and the use mode where technology is constructed technically and socially respectively. For reseachers, it is imperative to study such interpretations and transformations, for example the 'junction' named CSCW. Understanding societal and organizational contextual conditions, opens for a deeper understanding of work. 3. Looking at Work Nardi and Engestrom [8] underline how understanding invisible work is crucial to designing and managing organizations. Invisible but valuable work is often eliminated when organizations are restructured and work is reorganized. To make work itself visible, becomes important: Much work is visible. It yields to being mapped, flowcharted, quantified, measured. When planning for restructuring or new technology, visible work is the focus of attention. It is the only work that is seen, so efforts to restructure center on how visible work can be manipulated, redrawn, reorganized, automated or supported with new technology. But a growing body of empirical evidence demonstrates that there is more to work than is captured in flow charts and conventional metrics.(ibid., p 1)
P.-A. Persson / Putting CSCW in a Business Context
241
Kensing and Blomberg [9] mean that ethnographically-inspired fieldwork techniques integrated with participatory design (PD) techniques provide insights into unarticulated aspects of the work and shared views on it. They see PD as a maturing area of research, the emphasis of which can be at the individual, organizational, or national arenas/levels. The first level, according to Kensing and Blomberg, so far has been at the centre of attention. To work contextually means to see how these levels interfere with each other. PD, they say, has not been coupled to cooperative work, but to cooperative strategies for system design. Power as distributed in organizations reinforce the tendency to build managers' goals and strategies into new systems, promoting workers' conflicts with managers/owners. Visibilization might mean to involuntarily choose sides because what is visible can be controlled, for better or for worse. Hayes means that dominant assumptions of how organizations and information systems should be designed have been challenged during recent years [3]. Improvisations and workarounds, for example, may become more prevalent and noticeable with the advent of new technologies such as CSCW applications. They open new relations and may, more than single user technologies, cause instability and uncertainty. I conclude that we do not know whether new applications and work-arounds are goals or the consequences of new control technologies. Let us return to Berg, who seems less enthusiastic over what CSCW, carrying 'problematic weight', has achieved [1]. He means that the human-centred theoretical focus, and the call for transparent tools that support the work that humans do tends towards rather powerless technologies: Only modest technologies, whose functionings are transparent to the user, whose workflows should not be too explicated and detailed, and whose structure should be re-designable by the users themselves seem to fulfil these criteria. (p. 391)
If this is the situation in some organizations, how then can the achievements of CSCW be more rewarding, and well as any research related to them? Some inspiration is possible to get from early organizational research, and this will then lead to a look at modern managerial control, defining the freedom of action for both research and CSCW. 4. Organizational Systems, Cooperative Systems and Research When "systems thinking" was a new discipline in the post WWII years, Peter Selznick analyzed and described organizations as formal structures representing rationally ordered instruments for the achievement of stated goals [10]. He conceptualized an organization as 'a system of consciously coordinated activities or forces of two or more persons' (p. 261). "Structural expression of rational action" includes a pattern of coordination, a "systematic ordering of positions and duties" defining a chain of command for the integration of specialized functions. He saw delegation as a primordial organizational act, requiring continuous elaboration of formal mechanisms of coordination and control. Organizing means a persistent pressure for the institutionalization of relationships, eliminating the uncertainties connoted with individual actors' preferences, also making them interchangeable. Thereby the formal structure becomes possible to manipulate into an "instrument of rational action". Selznick realized, however, that formal structures never succeed in "conquering the nonrational dimensions of organizational behavior" (ibid.). An organizational structure contains analytically distinct but empirically united economic and adaptive social structures. In Selznick's time, evidently, the attention of management technicians and students of public and private administration was often focused at the economy-organization. However, Selznick concluded that the organization as economy was necessarily conditioned by the organic states of the total concrete structure, "outside of the systematics of delegation and control" (p. 263), which made it impossible to separate control and consent.
242
p. -A. Persson / Putting CSCW in a Business Context
Figure 1. Ongoing transformation of a cooperative system
In practice the economy-organization may be technically manipulated in terms of efficiency and effectiveness, which have reciprocal consequences for the other structure(s) but just looking at the economic aspect of an organization provides inadequate tools for control of the whole structure. Therefore it becomes necessary to view formal organizations as cooperative systems, thus widening the frame of reference of those concerned with the manipulation of organizational resources, in other words managers and controllers at various levels. I conclude that cooperation is the term that signifies keeping both control and consent in mind, finding mechanisms that include conflict resolution rather than open hostilities, articulating different opinions, initiating outright 'repair work' in case of breakdowns. In large organizations, deviations from the formal system tend to be institutionalized, and instead unwritten laws or informal associations are established. They are then transformed and given a formal structure, as part of an ongoing cycle of deviation and transformation (Figure 1). In cooperative systems, in order to remain functioning and stabilize a disequilibrium, transformations include size, organization, goals, leadership, doctrine, effectiveness. What concerned Selznick was restoring formal authority, gaining respectability, legitimacy, reestablishing stability of formal authority, while keeping the control-consent twins in mind. What he did not know much about (even if he had acknowledged them) is of course CSCW systems. What we as researchers should be concerned about is a) how to do research in organizations seen as cooperative systems consisting of economic and social components, b) what purposes this research serves, and thus who may benefit from the results, and c) how to make IT a "more fullfledged actor in the construction of a more desirable world" [1, p. 393]. In other words, how can research be exploited as part of the organizational practices and processes? To what extent can research find a critical perspective to look at these practices, interpret them, thereby contributing to ideas and concepts applicable in CSCW and in organizations? As regards the first concern (a), because economic and adaptive social structures are "empirically united". research methods should allow analytical distinctions between them. Let us look at management practices half a century after Selznick's conclusions. 5. Current Trends in Management Control Bourgignon discusses modern management control, arguing that basically it uses a Taylorist approach of scientific management, corresponding to a reinforcement of control [11]. This reinforcement ("what counts is what is counted", ibid., p. 7) happens in firms where simultaneously there is an increased and officially acknowledged demand for autonomy, creativity and initiative. This dichotomy probably causes frictions. Formal (or control) regulation and informal (or autonomous) regulation coexist. The former is written in an organization's rule systems or in function descriptions, becoming recognized management principles, while autonomous regulation rests on effective relations possible to discover behind "official fiction"
P. -A. Persson / Putting CSCW in a Business Context
243
(ibid., p. 6). Eventually, the new methods enlarge the scope of management control in three directions: • non-financial dimensions of performance, often called qualitative, • means of action, • interdependencies in organization, complementing the responsibility-focused performance measurement. Control and autonomous regulations are juxtaposed. Negotiation tends to reduce their opposition, which all the same is never totally eliminated. An organizational control rule intervenes in the recognized "actors' games". Its effects may be those intended, but also unexpected ones, reinforcing an autonomous rule-set - even if this is different - and with it the deviant behaviour it intended to replace or counter. Management control formalizes organizational expectations in budgets and plans. Bourgignon remarks that "planning, budgeting and MBO (Management by Objectives) are areas of negotiation in which control regulation and autonomous regulation are competing through the opposition of local and central interests" (ibid., p. 7). Labouret, analyzing accounting and budget practices, supports this view and concludes that budgets may be tools for rational choice, processes selecting and retaining information, or political tools [12]. In addition they are standardized but highly rational representations of reality, according to economy-control principles. Concluding from the discussion by Schaffer et al., they even represent a superior rationality [13]. The ideal organizational design thus may be one where articulation work, clarifying relations and connecting work activities [14], is unnecessary because business structures and processes already deliberately coordinate workflows. The remaining application areas within firms for CSCW might be for strategic management, accounting and negotiation, or CSCW for persons who do not fit into the ordinary workplace. Finally CSCW might be an approach for learning and a way to attract those who do not accept traditionally organized work, but constitute a small but valuable resource. We shall have a closer look. 6. CSCW Rationales and Participatory Design Maybe CSCW in a broad sense - work mediated and supported by computer technology objectively is a common and unproblematic style of how work is organized but the rationales behind CSCW may vary considerably. The common understanding today probably is that computer systems are tools which need to be designed to be controlled by the people using them, support developing work activities including communication, not making them more rigid. Selznick's view on organizations as cooperative systems makes it possible to regard "enforced CSCW" as a 'manipulative cooperative strategy', a way to establish tight control by explicitly working with social aspects, using 'rational' technologies. CSCW then becomes a strategy for articulation work when the divergence between actors in organizations is great (ibid.), when smooth functioning is necessary as a way to transform deviations into structure (figure 1). Alternatively, CSCW can be a legitimizing approach to avoid tight managerial control while keeping issues of power and autonomy unsettled. The double emphasis on productivity and quality when designing modern business organizations, especially as PD and CSCW-approaches, is somewhat contradictory. Scandinavian design ideals, as discussed by Kyng, are that computer systems should enhance workplace skills rather than degrade or rationalize them, to pay attention to that which is often left out of formal specification, for example tacit knowledge [15]. Suchman describes the history of design in the US as one where computer science and systems development were viewed as strictly technical and commercial arenas, "dedicated to the advancement of technology and
244
P. -A. Persson / Putting CSCW in a Business Context
improvements to business productivity and profitability" (p. 46) with an increasing commodification of computer artifacts [16]. Suchman claims she and some other researchers intervened in the progress of systems research and design and succeeded in establishing the place and relevance of work in anthropology and sociology to CSCW's concerns. A related opinion of hers is that the complexity of CSCW led to the use of the social sciences. Her term "design in context" aims at building a cooperative design practice grounded in workers' knowledge, requiring new contexts for design where the resources are defined and dimensioned according to their interests. Suchman finishes with "it is...the creation of genuinely new and autonomous spaces for workers' control over technology direction and design, that poses the greatest challenge to us all (p. 51)." Recalling modern management, to what extent can we expect it to allow 'genuinely new and autonomous spaces'? If what Selznick said was true, we can expect breakdowns to occur because of the reduced flexibility and the augmented formalized control of the "social system" with the help of economic control mechanisms. Turning towards theory, the current state of affairs reflects what Giddens saw in 'risk society' [6]: High modernity involves the disembedding, or lifting out, of social relations, practices, mechanisms, and competencies from their specific, usually local, circumstances of time and space ('locales'), and their extension, thanks to the developments in communications, over much wider spans of time and space (p. 22).
We have to consider to what extent the social adaptive system can cope with this separation from the local economy-organization, if the social one can in some way maintain links with the latter when in a 'remote-control mode'. Then, if ever, and in case of breakdowns, 'good' CSCW is justified. Cooperation becomes a primary issue in order to 'repair' what did not function, even over large distances. Also, whenever negotiations are initiated, it is likely that the economy-organization may be less relevant, even in the budgetary procedures and negotiations. We shall return to the question whether CSCW is a rational management response to new control requirements, a strategy to preserve autonomy, or the possible mechanism to establish cooperation, simply put: a process for keeping control and consent together. A glimpse from military management, command work. gives some insights. The accounts originate from an ethnography conducted in the Swedish Army during domestic command and staff exercises in 1998. but are still representative for the empirical domain. 7. Military Management and Command Work 7.1 Background The military has used and continues to use a great variety of control mechanisms in order to achieve reliability in war conditions and to be understood as a potent, rational and trustworthy nation state institution: internally, and by potential enemies. From early on, this kind of management, incorporating early industrial management practice, has included or fostered a panoptic control ideal [17,18]. Discipline, leadership, bureaucracy, ethos, training, etc. belong to its mechanisms. Traditionally, accounting and classical management control has not had the same elevated position as battlefield management and command in war in the military mind. Now there is a change. Ongoing for centuries, but accelerating during the latest decades, there is a search for a science of command and control, ideally informing design of superior control methods and technologies [19]. This search has two information-related branches. One is the state of art within management and process control, the other is about cognition and decision-making.
P.-A. Persson /Putting CSCW in a Business Context
245
Figure 2. May 1998, workspaces and cooperation, applying procedure (left), situated work (right)
The introduction of the program planning budgetary system (PPBS) during the 1960s [20] was a logical continuation of a long development of scientific management. The PPBS has spread from the US, piloting the current trend of Total Quality Management (TQM)-inspired best practice. Detailed control is not a military idea. Webster, recalling Giddens' theories, draws attention to the surveillance in modern society, not only in military management but in society and business in general [21]. There is a dichotomy between control and coordination ideals for war and peace-time respectively, between generals' decisions in battle, inspired by intuition, and rational action within bureaucratic operations and in accounting. It exemplifies Weber's dialectic between charisma and routinization [22], or simply stated, between mind and matter. This dialectic appears in the design of coordination tools, MIS, and in other kinds of information systems. Control systems design is often influenced by the Operational Research (OR) tradition, implying the modelling of ideal control systems for rapid processes and automatic decision-making rather than being concrete studies of work in context (Figure 2). This bias reflects the corporate interests and the problems to actually carry out studies under war conditions. Since the computer was introduced, its use for managerial control purposes has both reflected the long tradition of strategic management [23], and the successive conceptualizations of 'information systems' [24]. The difficulties to establish a cause-effect relationship between control technology and operational efficiency prevail. The military still searches for rational solutions, leading to very reliable control. The modern military bureaucracy within 'risk society' applies tight control mechanisms similar to other businesses, but with a particular interest in technological frontiers, trying to reduce uncertainty, achieving predictability, precision and speed. Among them, CSCW has a few potential positions. However, what surrounds CSCW and have to be considered are large structures, 'systems of systems', and 'networked' organizations. 7.2 The Case of the Common Tactical Picture Management principles influenced by cognitive science and Simon's view on rational decision-making has led to a dominant top-down commander-centric (one of many categories of decision-makers) view on management (command work), and subsequently a search for coordination mechanisms and decision supporting technologies along the chain of command. The latter aim at attention management, reliable performance and secured command authority across widely distributed units. There is a common belief that it is possible to externalize and distribute knowledge about complex situations among various members in an organization and get an adequate understanding about what is represented, which is often called "situation awareness". Knowledge
246
P.-A. Persson / Putting CSCW in a Business Context
representation of static elements and dynamic transitions in the battlefield is thus central in the requirements lists [25]. Real-time assessment and automated resolution of uncertainty in the bottom-up data stream from surveillance systems and as reports are recent topics, nourished by promises for relief with the help of technology, including a mixture of AI-related approaches. Kahan et al. discuss the concept of the commander's dynamic image, being representative among the command and control issues [26]. It is not a depiction but a mental model of the whole situation. It is to include the understanding of the history of the battlefield situation and projected futures. It helps the commander understand what action needs to be taken. The image must also be 'shared' by staff members. A lot of communication goes on in order to achieve this. Kahan et al. stated that commanders seek options and assessments appropriate to their image rather than data. When they request "data" they do so for reasons related to images and image sharing. "Data" here means something specific. Information is then given meaning by the image that frames it, working as a context. There are some related constructs. Whitaker and Kuperman used the term "Shared information space", connoting a medium that could be used dynamically to help people share their view of the world with others through joint manipulation of each person's personal models of the situation [25]. Whitaker and Kuperman considered it a key concept in the field of CSCW, its implementation being one cornerstone of advanced GDSS. In order to equalize access to critical information across the widely-distributed set of military actors, and the sharing of critical data, all actors must be able to orient themselves and their actions to a common 'picture' (ibid.). The shared 'information pool' is called the Common Battlespace Picture (CBP) (Table 1): Table 1. The characteristics of the Common Battlespace Picture Mutual accessibility Mutual interpretability
Mutual meaningfulness
By all relevant actors An effective CBP must make provision for individual actors' frames of reference (differing knowledge schemas and terminology) with respect to the information delivered to them. (Does this mean an automatic adaptation to each and anyone, and the use of different media and formats?) "the right information in the right time"
Mutual manipulability
By the one who thinks there is something that everyone should know.
Rather than being a plain picture it is a fused depiction of battlefield status available to a given warfighter, without assuming it to denote a literal concentration of data at any one physical site. During the last years, huge efforts have been directed to produce these artifacts and the tools to create them as the "silver bullets" against confusion, mental overload and force degradation. One of the less technical and not too detailed current US definitions of the Common Tactical Picture (CTP) is: The CTP is the current, anticipated, projected, and planned disposition of hostile, neutral, and friendly forces that includes amplifying data (...) for a single operation. Real-time, near real-time, and non-real-time data from national, theater, and tactical sensors (...) feeds the CTP via available communications links (...) provided by the Service Components and other organizations...this collection of data is combined with any final amplifying data (planning, weather, etc.) to produce a Common Tactical Picture (CTP). [27, p. 4]
The Common Operational Picture (COP, for the next command level) is constructed from one or more CTPs. The Swedish military's common view is that a decisive precondition for command and control is that forces automatically and literally get a common image/picture of the current situation. Its production and distribution is to be supported by the command support MIS. The image is usually conceived as symbol layers on an electronic map, representing additional information (terrain, planned action, enemy etc.). Previously, this need was sometimes satisfied with the help of a local HQ television picture. The search for a suitable technical solution goes on. The Army has a prototype construction where situation data (position and status) about units are stored in networked distributed databases, possible to present as symbols
P.-A. Persson / Putting CSCW in a Business Context
247
on electronic maps. Changes in the databases are replicated on command or automatically across the command structure. Full functionality implies very complex and distributed database management, and interruption-free communications. Operators/users, once the updating is complete, can choose how much (or little) of the "image data" should be visible as symbols on their screen. Individual pictures thus can look different and be interpreted accordingly. Little is said explicitly about CSCW in these sources and not much in the daily discourse. The Swedish COP/CTP-concept is supposed to support various instantiations of control and command work, distributed or in the same HQ (subsection). Technically the production and realtime updating of an image requires much computer power and integrated reporting and positiongenerating subsystems. Theoretically it seems to rest on the conviction that it is possible to capture knowledge about a battlefield and situation, pack it into an image and then transmit it to viewers who get the intended knowledge once their attention is awakened. What dominates here is Computer Supported Command Work and ad hoc rather than conscious CSCW. Several other, although less complicated IT artifacts, coexisted with the MIS and its components. Field data illustrate operators' work with map-oriented pictures, instantiations of the CTP. People used traditional maps, dedicated simple common tables and templates, electronic messaging systems, and some MS Office products (Figure 1). They communicated, outlined and sketched future actions, fed in new information (symbols), measured and calculated action. Mental processes and individual activities certainly are initiated and affected but it seems to be very complicated to define down-to-earth substantial control operations. It is likely that systems design using models of mental processes as blueprints is vulnerable and hazardous. 7.3 The Case of the Actualities' Table Much work remained invisible but the ethnography revealed other situated design initiatives leading to standard PC-based constructs. "Drivers" behind design efforts were to get overview and support for resource management, to communicate within and between teams, and planning requirements. One of them was the Actualities' Table, demonstrating other aspects of CSCW. Traditionally staff work means accounting practices, resource administration, auditory tasks, making inventory lists etc. where various tables have been designed, standardized, and used for a long time. They have been pinned up in workspaces in large format or used as paper artefacts, often being distributed. One of them, the original "Actualities' Table" (Figure 3), shows total force structure, chain of command, environmental factors (upper half) and immediately available
Figure 3. The basic 1980 Actualities' Table providing resource overview
248
P. -A. Persson / Putting CSCW in a Business Context
Figure 4. Early version of the modified Actualities' Table
resources, "freedom of action" (lower half), and was to be updated based on regular reports and impressions from the field. The assessment and definition of "Battle capacity" (centre, right) has been, and still is, a crucial part of the work. Various soft and hard data are heuristically fused into an aggregate single figure (the scale being 1 to 5 is used). The figure '3' signifies the lowest acceptable capacity, signifying some deficiencies. In the regular reports and in work, this figure is usually accompanied by written "commander's comments" clarifying details. When initiated early in 1998 the new Actualities' Tables was mentioned in the Standing Operating Procedures (SOP) simply as "templates" and "forms". It evolved during a series of shorter exercises, implying a refinement of its paper-based predecessor. Templates, matrices and forms have belonged to the basic IT artefacts used for control and resource management. Very few outside the core design group realized what the new 'template' really was until it was actually used. It was based on simple traditional work practices and tools, now redesigned in officeware (MSExcel). An early (January 1998) version (Figure 4) shows how the new kind of table was expanded within its basic design. The table was accessible via the surrounding MIS, possible to project on large screens, mail, print out, pin up etc. It actually replicated some functions from the MIS but with less effort. For instance, updating, distribution, storage and retrieval of data between and in the work groups were and required less effort. Through this, subsections in the HQ tailored their own related artefacts, communicated for control and coordination and got an overview. This self-help spreadsheet artefact and its mode of representation of the organization and its operations admitted staff to create any work procedure out of very simple components: write, draw, read, talk; in short - social interaction and communication. It was used in every phase of its growth but did not become 'ready'. Compared to the COP (CTP, CBP), the table-system gave another kind of 'picture' of the current situation. It replicated a strategy well known in accounting, one that admits overview, comparisons and illustrating relations between resources and action. It was not, however, primarily a tool for accounting, on the contrary. It supported coordination and control through the use of aggregations and "economic abstractions". It had considerable social value, allowed 'selective awareness' and was controllable by people in their work.
P.-A. Persson / Putting CSCW in a Business Context
249
Figure 5. The linked structure of function tables
What evolved was a cluster of spreadsheet applications within a hypertext structure (Figure 5). A mix of formalized data and informal registrations and notes could be used. Various tables were linked via macro commands, subordinated to an upper "Gate Table" interface and stored in the HQ servers. The allocation and control of read-write authorities remained unsolved. Special rules had to be implemented. What makes this application especially interesting was that people hardly realized their achievements, being merely invisible even for themselves. The application "died" when the 1998 series of exercises ended. Only later were its benefits realized. 8. Discovering a Work-Related Design Tradition A parallel design initiative occurred within the artillery section where people created their Actualities' Table. The accounts from an artillery officer who had co-developed this table together with the Chief of Artillery told a complementary story. Previously he had experienced failures emanating from earlier advanced design attempts. In 1998, they had further developed his own "vintage" control artifact or tool, the 'Pinboard' (Figure 6). Its basic physical design was a piece of soft-board the size of about 40 x 60 cm with marked columns and rows. It was the answer to demands for a robust tool to use inside narrow armoured vehicles, his workspace when
Figure 6. Sketch of the 'Pinboard' providing an overview for control of artillery activities. 'Arrows' signify pins.
250
P. -A. Persson / Putting CSCW in a Business Context
he in a previous brigade position coordinated the actions of artillery units via the tactical radio networks.Pins were positioned in the fields of the board, representing the activities and situation of sub-units (artillery battalions), and re-positioned according to orders and reports. Notations (ammunition, activities, time) supplemented the pins. The Pinboard helped him getting an overview of artillery activities by representing the current tactical situation with the help of these pins and notations. The updated 1998 more detailed spreadsheet version (still figures and letters, no pins, but printable) and the email system allowed him to work similarly. He could distribute orders, receive updated situation overviews without getting stuck in the evolving MIS when mapping his field of operations. The common denominator between these tables is their roles and capacities as representations of complex social action; past, present, and planned. Theoretically, they are diagrammatic representations where precision is one aspect, but relations and overview more important [28], supporting the capacity for making inferences and calculations, such as on maps and in spreadsheet representations: A representation consists of both data structures and programs operating on them to make new inferences. The data structures we consider are node-link structures that include schemas employing attribute valuepairs. Actions modify data structures, that is, they make and record inferences. Although the data structures we shall postulate are stored externally, on paper, the productions [of programs] that operate on them are in the problem solver's memory, (ibid., p. 67)
9. Interpretations, Rationality and Boundary Management In the military organization a certain rationality guides the design of IT artifacts. It feeds the idea that the COP/commanders' image is the most important single item/application, relying on the conviction that it is technically feasible within a management framework. Because other needs are not satisfied, complementary initiatives are initiated, as during the fieldwork in 1998. Probably both action rationality and decision rationality were desired control outcomes. We do not know whether self-help systems are rational in the same sense as the COP, but they demonstrate a certain action rationality [4]. Something had to be done when the local MIS, itself being a temporary substitute context, did not provide adequate support for resource management but instead meant constraints. An organization seen as an open system has to demonstrate variety vs. a changing environment. The primary task of management is to handle the boundary conditions through managing the co-variation of internal and external processes. The boundary management becomes a vital issue. In order to succeed, flexibility, commitment and initiative are needed [29], by necessity admitted by its information systems. The "economic" representations in the tables allowed autonomy and contributed to efficient work practices and in some aspects countered breakdowns, helping those involved in boundary management, assuring the survival of the whole organization. The applied rationality and its product were, however, considered less desirable deviations from the mainstream design work of the MIS: they drew on limited design and development resources which should be kept together. Concentration of force for maximum output is a classical military strategy. We see the confrontation between rationalities. A team involved in boundary management experiences needs that have to be recognized. Boundary management, by its very nature, means to systematically apply a variety of rationalities to achieve political goals, managerial control or both. The boundary conditions correspond to Schmidt's and Simone's criteria for cooperative work. In order to minimize the risk of breakdowns, and if they occur, work procedures and artifacts must allow smooth recovery through repair work also implying articulation work [14]. People were simultaneously designers and workers themselves, searching for sustainable methods. Any representation of and for human actions must be meaningful, but there is no straight line between design and being
P. -A. Persson / Putting CSCW in a Business Context
251
meaningful, except that participation and local use facilitate meaning. In these cases, meaning, power, and norms were juxtaposed. The norms span both design ideals and control, each aspect linked to various power centres. 10. Design of Collaborative Mechanisms It is necessary to establish the kind of cooperation that Selznick recommended, to unite control and consent, bringing the adaptive social and the rational economy-organization together. Wall and Mosher provide examples of representations aimed at descriptions of work practices for shared understandings between workers and designers: storyboards, project walls, 2D or 3D concept models, visualization and construction games, even Post-it notes [30]. They claimed that representations can make explicit aspects of work that have become transparent to workers. I conclude that such discoveries can make them more confident in what they actually can do. Hirschheim and Smithson stated that what is technically elegant may have limited social value and that the opposite may be the case as well [31]. What we can learn is that artifacts for understanding and communication, respectively, can require different design, that communication of work content and design principles may require other capacities than artifacts for use in work, especially CSCW, but that in both situations simple items can do the job. The old Pinboard and the new Actualities' Table illustrate simple design but qualified work. Both were locally controlled, actively constructed and meaningful within their context. They were responses to artifacts that were not perceived as meaningful. Teege means that flexibility and tailorability are crucial factors for the success of CSCW systems [32]. The purpose of tailoring is to compensate for the contrast between organizational work (executed by the social adaptive system, Selznick) with its situated, collaborative and changing nature, and the formal theories and models of work (in the economy-organization) that underpin and constitute any information system [33]. Models expressing formal theories and actions, they concluded, often break down because work is highly context-dependent and dominated by situated actions. Instead of providing full-fledged applications, components of applications should be provided and then support tailoring. Use means redesign and reinterpretation, and exploitation of 'trafficable' rationalities, even using them symbiotically. Berg defines information technologies in workplaces as a type of reading and writing artifacts [1]. They receive and transform input, producing output ("There is a constant traffic of inscriptions back and forth, and in this interrelation lies the transformative power that IT enthusiasts so gladly attribute to the 'intelligent software agent'," p. 374). This perspective allows a blurring of the boundaries between computer-based and paper-based artifacts, the latter also being transforming. Both kinds of artifacts actively mediate human action, ideally contributing to the cooperation Selznick described, preparing the ground for control and consent. In Berg's terms, applications (potentially) hold the work (group) together and make it perform as a whole. Berg suggests a decentralising perspective, not putting the human actor in the centre, but instead looking at the relations between artifact and humans, sometimes leading to synergy. Hypothetically, collaborative mechanisms, including technologies and artifacts, should be possible to adapt to different rationalities. 11. Conclusions, Future Work and some Aspirations This research has demonstrated what CSCW can be, as a research approach and as work. Ethnography, as the military cases demonstrate, can contribute to both understanding and to work design. According to Suchman, ethnography, which was first used in the service of
252
P. -A. Persson / Putting CSCW in a Business Context
powerful actors wanting to manage others, has become a "militant source of radical interventions and critique on the part of post-colonial efforts" [16, p. 50]. Whether or not this declaration opens business gates is debatable. Difficulties may evolve because not everything in organizations, especially not in the military, is open for investigation. There are a few different rationales for cooperative work in its modern sense. CSCW as a research field has to follow on. The first is seeing cooperative work (read: CSCW) as a temporary transition phase preceding structure and institutionalization, consciously uniting control and consent. A second is cooperative work as a permanent but less orderly state, demonstrating failure. The third is the actual state behind an orderly facade of MIS and procedures, necessary to make an organization function smoothly. Fourth, it is the only possible strategy in certain work conditions where organizing in the normal way is impossible. A related case is when "cooperative work" masks control ambitions. Finally, it is an overt or covert political strategy to avoid too tight managerial control. It might be that cooperative work is a more humane way of organizing than TQM, not less demanding but more rewarding. And as regards the question Why less paper? (Berg), one answer may be that augmented control is desired, for some reasons: discipline, surveillance, costs, process visibility. On the whole, because information systems as social systems evolve over time, the proper occasion for evaluation is difficult to establish. In order to study how meaning, power, and norms are constantly created in organizations, researchers have to be close at hand. Research teams must understand politics, managerial practices, accounting, cultural issues, pragmatically forming their work. A qualitative approach is considered relevant to investigate also modem management control [11] where formal procedures have both political significance and symbolic properties [2]. It may not be evident where the core business occurs: politics and rationalities may intervene. Batteau sees organization as an ongoing struggle to impose order, for strategic ends, seldom living up to the facade of the order it projects [34]. In this environment the "metonymic equivalence of command and rationality within an organization turns resistance into irrationality" (p. 731). Cost-accounting and budget techniques are central in modem management, involving different rationalities and contradictions: budgets and financial reports carry considerable credibility ('a totemic status') but are open for what is called 'cooking the books'. Losses, incomes, profits can be moved or hidden [34]. The way all these practices imply development and use of IT-artifacts are similar to what goes on in the military command work. Except for the fact that tables are for economy-control they are resources for CSCW, ad hoc or deliberately designed. Therefore accounting practices should belong to the research domain of cooperative systems but may, as in the military, be treated as a kind of management undergrowth, a kind of maquis compared to more prestigious work. Hypothetically, the real function of taken-forgranted accounting practices is to create the disciplinating conditions whereupon most other control mechanisms are built, interdependently used for the achievement of control and consent. One of the paradoxes within accounting is that computers, often regarded as the rational technology above all, are abundant and may be the tools for both very precise and credible control and for manipulation. Often cooperative work makes the difference, supported by computers. Accounting itself, the product of cooperative work, often inspires decision support systems for single users. Accounting has been an integrated and integrating portion of culture and organized work for millenia, for long short on theory but now studied from various directions, in the same manner as computer use and information systems design: economics, semiotics, linguistics, and structuralism are just a few of these [18]. Cultures are intentionally produced and used, but, because "opposition and negation are necessary elements of the organizational landscape" [34, p.737], context and consequences have to be studied together, something which might be problematic:
P.-A. Persson / Putting CSCW in a Business Context
253
The moral choice that the anthropological profession must make is whether culture will be represented by those socialized into the ethos of instrumental rationality, or whether instrumental rationality will be interrogated by outside, critical voices (ibid.)
Now, because the value of morality can hardly be denied, CSCW researchers can hope for acceptance also of controversial approaches, at least winning support from some actor in an empirical domain. The prerequisite is that CSCW researchers also, in the same way as people in their work, acknowledge that rationality can be manufactured without being immoral. Valuable lessons can be learned through studying their own practices, intimately linked to panoptic control ideals in the research community and in their empirical sites. Many share the fate of being thrown into this special boundary zone, in 'risk society' having to find ways of surviving (or escaping). Possibly some CSCW application can provide us with power and meaning and at least help us 'fix the bits and pieces' together. Acknowledgements The anonymous reviewers inspired me to redesign the original draft, thus guiding me towards new discoveries. The Army, having generously supported the fieldwork, hopefully will agree and accept another approach.
References [1] M. Berg, Accumulating and Coordinating: Occasions for Information Technologies in Medical Work. Computer Supported Cooperative Work: The Journal of Collaborative Computing 8 (1999), pp. 373–401. [2] G. Symon, K. Long and J. Ellis, The Coordination of Work Activities: Cooperation and Conflict in a Hospital Context, Computer Supported Cooperative Work: The Journal of Collaborative Computing 5 (1996), pp. 1–31. [3] N. Hayes, Work-arounds and Boundary Crossing in a High Tech Optronics Company: The Role of Cooperative Workflow Technologies. Computer Supported Cooperative Work: The Journal of Collaborative Computing 9 (2000), pp. 435–455. [4] N. Brunsson, The Organization of Hypocrisy, Talk, Decisions and Actions in Organizations. John Wiley & Sons Chichester, New York, 1989. [5] N. Garnham, Information Society as Theory or Ideology: A Critical Perspective on Technology, Education and Employment in the Information Age. Information, Communication and Society 2 (Summer 2000). [6] C. G. A. Bryant, and D. Jary, Eds, Anthony Giddens: A Global Social Theorist. In: C. G. A. Bryant, and D. Jary,(eds), The Contemporary Giddens, Social Theory in a Globalizing Age, Palgrave, Basingstoke, Hampshire and New York, 2001, pp. 3–39. [7] W. Orlikowski, The Duality of Technology. In: C. G. A. Bryant and D. Jary (eds.), The Contemporary Giddens, Social Theory in a Globalizing Age. Palgrave, 2001, pp. 62-96. [8] B. A. Nardi and Y. Engestrom, A Web on the Wind. Computer Supported Cooperative Work: The Journal of Collaborative Computing 8 (1999), pp. 1-8. [9] F. Kensing and J. Blomberg, Participatory Design: Issues and Concerns. Computer Supported Cooperative Work: The Journal of Collaborative Computing 7(1998), pp. 167–185. [10] P. Selznick, Foundations of the Theory of Organizations. In: F. E. Emery (ed.), Systems Thinking. Penguin Books Ltd., Harmondsworth, Middlesex, England, 1969, pp. 261-280. [11]A. Bourgignon,. New Management Control: Is it really new? Is it really safe? 5th International Management Control Systems Research Conference, Royal Holloway, University of London. 4–6 July 2001. [12]V. Labouret, Premise for a simultaneous measure of paradigmatically different budget behaviours. 5th International Management Control Systems Research Conference, Royal Holloway, University of London, 4–6 July, 2001. [13]U. Schaffer, J. Weber, C. Prenzler, Characterising and Developing Controller Tasks - A German Perspective. 5th International Management Control Systems Research Conference, Royal Holloway, University of London, 4–6 July, 2001. [14]K. Schmidt and C. Simone, Coordination Mechanisms: Towards a Conceptual Foundation of CSCW Systems Design. Computer Supported Cooperative Work: The Journal of Collaborative Computing 5 (1996), pp. 155– 200. [15]M. Kyng, Users and computers: A contextual approach to design of computer artefacts. Scandinavian Journal of Information Systems 10 (1998), pp. 7–44.
254
P.-A. Persson / Putting CSCW in a Business Context
[16]L. Suchman, Strengthening our Collective Resources: A comment on Morten Kyng's 'A contextual approach to the design of computer artifacts', Scandinavian Journal of Information Systems 1/2 (1998), pp. 45–52. [17]K. W. Hoskin and R. H. Macve. The Genesis of Accountability: The West Point Connections. Accounting, Organizations and Society 1 (1988), pp. 37–73. [18]N. Macintosh, Accounting, Accountants, and Accountability: Poststructuralist Positions. BAS, Handelshogskolan, Goteborg, Sweden, 2001. [19]H. W. Sorenson, A Discipline for Command and Control. In S. E. Stuart and H. A. Levis (ed.) Science of Command and Control: Coping with Complexity. AFCEA Int. Press, Fairfax, Virginia, 1989, pp. 7–15. [20]M. Ways, The Road to 1977. In: F. E. Emery (ed.), Systems Thinking. Penguin Books Ltd., Harmondsworth, Middlesex, England, 1969, pp. 372-388. [21] F. Webster, Theories of The Information Society. Routledge, London and New York, 1995. [22] R. Schroeder, Max Weber and the Sociology of Culture. Sage London, Newbury Park, New Delhi, 1992. [23] K. W. Hoskin, R. Macve, and J. Stone, The Historical Genesis of Modem Business and Military Strategy. Fifth Interdisciplinary Perspectives on Accounting Conference (IPA97), Manchester, 7-9 July, 1997. [24]D. Cecez-Kecmanovic, The discipline of information systems: Boundaries crossed, boundaries pushed. In: M Sheehan, S. Ramsay and J. Patrick (eds.), Transcending Boundaries: Integrating People, Processes and Systems Conference, Griffith University, Brisbane, Australia, 2000. [25]R. W. Whitaker and G. G. Kuperman, Cognitive Engineering for Information Dominance: A Human Factors Perspective. Crew Systems Directorate, Human Engineering Division, Wright-Patterson AFB, Dayton, Ohio, OH 45433-7258, 1996. [26] J. P. Kahan, R. D. Worley and C. Stasz, Understanding Commanders' Information Needs. RAND Corporation. R-3761-A, Santa Monica, CA, 1989. [27]Defense Information Infrastructure (DII); Common Operating Environment (COE); Common Operational Picture (COP); Technical Requirements Specification (TRS). Commander, Space and Naval Warfare Systems Command (PD-15 E), San Diego, CA, Draft February 2001. 26 October 2001: http://diicoe.disa.mil/coe/aog twg/twg/coptwg/coptwg page.html. [28] J. H. Larkin and H. A. Simon. Why a Diagram is (Sometimes) Worth Ten Thousand Words. Cognitive Science 11 (1987), pp. 65-99. [29]F. E. Emery, Introduction. In: F. E. Emery (ed). Systems Thinking. Penguin Books Ltd., Harmondsworth, Middlesex, England, 1969. [30]P. Wall and A. Mosher Representations of Work: Bringing Designers and Users Together. PDC'94, Participatory Design Conference. Chapel Hill, NC, USA, 27-28 October 1994. Computer Professionals for Social Responsibility 1994. [31]R. Hirschheim and S. Smithson, Evaluation of Information Systems: a Critical Assessment. In: L. P. Willcocks and S. Lester (eds.), Beyond the IT Productivity Paradox. John Wiley & Sons, 1999, pp. 381–409. [32]G. Teege, Users as Composers: Parts and Features as a Basis for Tailorability in CSCW systems, Computer Supported Cooperative Work: The Journal of Collaborative Computing 9 (2000), pp. 101-122. [33] A. I. Morch and N. D. Mehandijev, Tailoring as Collaboration: The Mediating Role of Multiple Representations Applications Units. Computer Supported Cooperative Work: The Journal of Collaborative Computing 9 (2000), pp. 75–100. [34] A. W. Batteau, Negotiations and Ambiguities in the Cultures of Organization, American Anthropology 4 (2000), pp. 726–748.
Which Kind of Artifacts for Cooperation?
This page intentionally left blank
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press, 2002
257
Coordinative artifacts in architectural practice Kjeld Schmidt1 and Ina Wagner2 IT-University of Denmark, Copenhagen, Denmark. Email:
[email protected] 2 Vienna University of Technology, Vienna, Austria Email:
[email protected] 1
Abstract: CSCW researchers have increasingly come to realize that the material work settings and the artifacts that populate them play a crucial role in the seamless and effective coordination and alignment of cooperative work. However, while the central role of artifacts in cooperative work has been recognized and applauded, the concept of artifact as used in CSCW is highly problematic as it often presumes mentalist notions of artifacts as simple vehicles of 'information'. This paper is an attempt to depart from these notions. Based upon ethnographic studies of architectural work, we attempt to develop an understanding of the coordinative roles of artifacts which accounts for the multiplicity of artifacts and the complex interplay of particular practices and the specific material forms of artifacts.
1
Introduction
For many years a very large part of CSCW research has been focusing on immediate interaction in small groups, typically co-located. The motivation for this focus has generally been to devise technologies that could help cooperating actors to emulate such interaction over physical distance. Whatever the motivation, however, face-to-face interaction was uncritically conceived of as the paradigm of human interaction, compared to which all other forms of human interaction were taken to be impoverished emulations. The obsession with media spaces and the conversation metaphor that characterized CSCW research for many years bears witness to that. This is of course a gross simplification of the general situation, as there were clear exceptions from this paradigm from the very beginning. Most significantly, ethnomethodogically informed ethnographic studies demonstrated that material artifacts play a crucial role in coordinative practices and developed important analytical categories for studies of such practices [cf., e.g., 4; 5; 6]. Other ethnomethodogically informed workplace studies similarly drew attention to how actors skillfully exploit the affordances of the material work setting in order to effortlessly and fluently coordinate and integrate their individual activities [10; 11; 33; 35]. Other in-depth workplace studies pointed the same way [9; 17]. These and other findings were reflected in early attempts at developing conceptualizations of cooperative work [24-26], which did not manage to turn the tide, however. In the course of the following years, CSCW researchers have increasingly come to realize that the material work settings and the artifacts that populate them play a crucial role in the seamless and effective coordination and alignment of cooperative work. The shift of focus is quite remarkable and can be gauged simply by 'comparative browsing' of early and recent CSCW and ECSCW proceedings as well as the ten volumes of the CSCW journal.
258
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
While one will have to look hard for exhibits of or even references to artifacts in CSCW publications in the first many years, one will not need to search for many minutes to find an abundance of such analyses and exhibits in recent CSCW papers. Obviously, the role of artifacts in cooperative work has become topical. More than that, whereas social psychology, group sociology, communication theory, conversation analysis, etc. played prominent roles in the early years of CSCW research, the field now exhibits an sharper analytical and conceptual attention to the role of artifacts in coordinative practices. Although ethnomethodologically informed ethnographic work, as noted above, pioneered the study of the uses of artifacts in cooperative work, the shift is obvious here as well. While the procedures and conceptualizations of conversation analysis, that were so important in early ethnomethodological studies of cooperative activities, are of course 'still going strong', it is worth noticing that studies of conversational interaction are now being framed in broader and more inclusive analyses that place stronger emphasis on the material settings of cooperative work [12; 33; 34]. Moreover, there are now some very systematic attempts to address the affordances of material as opposed to digital artifacts [29]. The shift of focus can also be seen in that conceptual frameworks such as 'activity theory', 'distributed cognition', and 'actor-network theory' have developed significant followings in the CSCW community in recent years. These frameworks surely have achieved widespread circulation because they are seen to meet the demand for a conceptual foundation for CSCW research and design. But they have also gained ground because they, by contrast to social psychology etc., are seen to accord artifacts and their use a crucial role in human action and interaction. However, the increasing focus on coordinative artifacts has not been accompanied by increasing conceptual clarity. In fact, as far as the 'imported' frameworks are concerned, fundamental conceptual problems have been contracted, so to speak, as part of the bargain. As this is not the place for a thorough discussion of these frameworks, a few, desperately brief comments, merely indicating the nature of these problems, will have suffice. Activity theory emerged in opposition to and as a break with the fundamental presumption of behaviorist psychology, viz. that human cognition is to be understood in terms of generic abilities. Against these presumptions, L. S. Vygotsky, suggested a conception of human action that was heavily influenced by Marxist theory, arguing that cognitive phenomena such as logical reasoning are grounded in historically evolving and culturally specific material practices. Thus, to Vygotsky and his followers, the skills involved in the production and use of tools, in the techniques of reading and writing, arithmetic, etc. are of central concern to psychology. Vygotsky's ambition was undermined, however, by his concept of 'psychological tools': :
1. In the behavior of man we encounter quite a number of artificial devices for mastering his own mental processes. By analogy with technical devices these devices can justifiably and conventionally be called psychological tools or instruments [...]. 3. Psychological tools are artificial formations. By their nature they are social and not organic or individual devices. They are directed toward the mastery of [mental] processes — one's own or someone else's — just as technical devices are directed toward the mastery of processes of nature. [...] 4. The following may serve as examples of psychological tools and their complex systems: language, different forms of numeration and counting, mnemotechnic techniques, algebraic symbolism, works of art, writing, schemes, diagrams, maps, blueprints, all sorts of conventional signs, etc. 5. By being included in the process of behavior, the psychological tool modifies the entire course and structure of mental functions by determining the structure of the new instrumental act, just as the technical tool modifies the process of natural adaptation by determining the form of labor operations.' [39]
On the surface, Vygotsky's concept of 'psychological tools' is an awkward term for what we now would call sign systems. However, on closer inspection it turns out to be quite equivocal, as it denotes techniques, practices, skills, signs, notations, as well as inscribed
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
259
artifacts such as diagrams, maps, and blueprints. More than that, the concept is fundamentally problematic, as it reflects the mentalist preconceptions underlying his understanding of sign systems. Not only does the concept reify the skills involved in speech, writing, numeration, counting, etc., in that it suggests that skillful action is somehow 'determined' by certain mental structures — by subsuming mental as well as material phenomena under the category of 'tools', any notion of materiality is eradicated from the concept of tool. 'Mental processes' are reified while material artifacts are spiritualized. In short, no sooner had the use of artifacts been made a central issue (and rightly so), before the materialist notion of artifact was conceptually dissolved. This de-materialization of the concept of artifacts has been continued uncritically in the subsequent activity theory tradition. Not only is the term 'psychological tools' in continued use [cf., e.g., 42; 43]. But it is also evident that activity theory, as an intellectual tradition and as a conceptual framework, makes it difficult to address the role of material artifacts in work systematically. In an introductory paper, Kuutti for instance, mentions 'instruments, signs, procedures, machines, methods, laws, forms of work organization' as examples of 'artifacts' [18, pp. 26]. Similarly, following Engestrom, Kuutti mentions in passing, without further arguments, that 'An object can be a material thing, but it can also be less tangible (such as a plan) or totally intangible (such as a common idea) as long as it can be shared for manipulation and transformation by the participants of the activity.' [18, pp. 27]. Here the concept of artifact has become utterly vacuous, as it simply denotes anything we can give a name, a point Kaptelinin brings home, unwittingly, by stating that 'Activity theory itself is a special kind of artifact' [16, p. 36]. The 'distributed cognition' framework developed by Hutchins and associates can be seen as a further development of the activity theory framework, in that it insists on studying human cognition in terms of historically and culturally localized practices. As opposed to activity theory, however, Hutchins pays detailed attention to trajectories of action 'distributed' over actors and artifacts in what he terms 'a system of distributed cognition.' [15, pp. 16-17]. In doing so, Hutchins directs attention to the specific format of the artifact and its role in human action.1 In spite of this, Hutchins de-materializes artifacts no less than Vygotsky, albeit in a different way. While Vygotsky talked about 'psychological tools' and thus only indirectly dissolved the concept of artifacts, Hutchins does it directly, by conceiving of artifacts merely as vehicles of so-called 'representations' on par with 'internal memories'. Thus, when summarizing their analysis of cooperative work in an airline cockpit, Hutchins and Klausen states: 'We can see that the information moved through the system as a sequence of representational states in representational media. From speech channels to internal memories, back to speech channels, to the physical setting of a device. Its representation in each medium is a transformation of the representation in other media.' [15, pp. 27]2
The notion that an invariant and immaterial being, 'the information', migrates from mind to artifact to mind is extremely problematic. This ghostlike entity that in turn takes residence in people and artifacts somehow manages to maintain its unity and identity. It is not just a mentalist notion, as Vygotsky's notion of 'psychological tools', it is what Taylor and Harris have aptly termed 'telementational' [7; 8; 36]: the transfer of invariant entities from mind to mind or even, according to Hutchins, from 'medium' to 'medium'. The orderly alignment of activities (which we need to investigate in order to be able to support it technologically) 1
This attention to the specific format is especially pronounced in Hutchins' earlier work [e.g., 13, pp. 47 f.] The same kind of analysis can be found in Hutchins' study of maritime navigation: 'The representations of the position of the ship take different forms in the different media as they make their way from the sighting telescopes of the alidades to the chart. [...] Representational states are propagated from one medium to another by bringing the states of the media into coordination with one another.' [14, p. 117]. 2
260
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
is simply taken for granted, like an invisible hand that mysteriously creates order behind the back of the actors. By presuming 'the information' as a unitary entity that propagates in the system while retaining its integrity, Hutchins and associates ignore the practices of producing this continuity and integrity. The fact that the various 'representational media' are of different nature and have different characteristics is far from ignored by Hutchins. In fact, it is one of his key concerns: 'In the cockpit, some of the relevant representational media are located within the individual p[i]lots. Others, such as speech, are located between the pilots, and still others are in the physical structure of the cockpit. Every representational medium has physical properties that determine the availability of representations through space and time and constrain the sorts of cognitive processes required to propagate the representational state into or out of that medium.' [15, pp. 27, 32]
But this recognition of the different characteristics of minds and dials brings us nowhere, as long as the infallible reincarnation of a unitary being, 'the information,' is presumed. By presupposing the order that is to be investigated and understood and in line with the idealistic precepts of cognitive science, the 'distributed cognition' framework tacitly presumes that the artifacts are mere successive representational incarnations of 'the information', or that the materiality of the artifact is immaterial, so to speak. It is misleading to conceive of 'representations' as cognitive or notional entities; they should rather be seen as and investigated as conventionalized practices of using artifacts. The point of this argument is not to belittle the value of the contribution of 'activity theory' or 'distributed cognition' to CSCW, but simply to point out that while the central role of artifacts in cooperative work has been recognized and applauded, the concept of artifact as used in CSCW is murky, ripe with all sorts of mentalist and cognitivist precepts. It is, of course, far beyond the scope of this paper to develop a theory of coordinative artifacts. The purpose of this paper is far more modest. Having pointed out that while the concept of artifact is topical and having argued that the clarity of the concept does not match its popularity, our aim is to try to frame the problem of artifacts in a new way that, as Roy Harris suggests, might be more productive: The view of human communication adopted here is integrational as opposed to telementational. That is to say, communication is envisaged not as a process of transferring thoughts or messages from one individual mind to another, but as consisting in the contextualized integration of human activities by means of signs.' [7. p. 4]
To do so, we will briefly describe the uses of artifacts in the work of architects. Our attempt at analysis of these practices and what they imply will, generally, be postponed to the discussion at the end of the paper. 2
The work
Imagine a typical architectural office (Fig. 1). It consists of several interconnected large rooms, each with several desks, each of these with a workstation. Most of these desks are covered with materials - plans, sketches, notes, photographs, faxes, books, samples. On shelves are large collections of binders for each of the current projects; in the entrance area a collection of scale models, and on the walls 3D visualizations, sketches, photographs, and newspaper clippings from previous and current work. The walls close to people's workspaces too are used as an exhibition space and decorated with materials from ongoing work.
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
261
Fig. 1 The architectural office
The physical layout of the office reflects the character of the architects' work. On the one hand it involves a smaller number of quite large projects, each of which lasts for months if not years and may occupy up to 20 people in the office. On the other hand the office puts in many tenders for competitions for which a design proposal has to be prepared quickly under high pressure. The architects' work is intensely collaborative. In a large building project various people work on different sections of the building and they may be responsible for particular design tasks. Thousands of documents are created in this process. Most of the internal coordination is done personally. People rush around for communicating design changes, reminding someone of important things to account for, offering explanations, helping to solve an adhoc design problem, checking a drawing, etc. A building project also engages many external actors - technical consultants (for construction, electricity, heating and ventilation, the lighting concept, the facade, etc.), a client and eventually one or several users, several local authorities, a general contractor, building companies, and craftspeople. The architects may have to coordinate the effort and consent of between 30 and 50 different people from different institutions and companies, each with their own professional competences and perspectives. Communication with some of these external specialists is interwoven with planning in an ongoing process. Interactions take place in different forms: asking for ad-hoc advice on the phone, exchanging faxes, email and files, and face-to-face meetings. In most projects it is the architect who has the responsibility for the planning process and the overall quality of the design. S/he coordinates the planning process with external consultants, local authorities etc. who each fill in their bits and pieces. However, the division of labor within an office and between architects and other specialists may vary, with variations reflecting different political regimes and cultures. There are many issues of control involved in the division of labor between the architect and all others. Searching for and negotiating technically and economically feasible solutions for a large number of details implies managing a large network of power and dependencies. 2.1
The process
Designing a building involves far more than having a design idea, developing it into a concept and expressing it in a series of sketches and plans. It is the detailed planning of the building's implementation which is at the heart of architects' work. In most countries the planning process is organized into (legally defined) stages with defined products: predesign, design, construction planning, etc. Each stage is concluded with the respective set of CAD drawings. In practice stages overlap a great deal.
262
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
Planning proceeds in many steps, in intense conversations within the team, project meetings with external specialists, and partly also solitary work. The conversations unfold through 'encircling themes' - addressing a particular issue, trying to clarify the 'facts', generating and testing preliminary solutions. Talking e.g., about the lighting design is connected to a 'journey' through different parts of the building. Many topics are addressed at the same time, as each has implications for many others. While some of these are discussed in detail, others are left open. As there is often a rapid switching of activities, a multiplicity of contexts has to be maintained. Ongoing work in the office is shaped by the different levels of skill and competence within the team on the one hand, by the natural divisions of a building project into parts (levels, functional parts, infrastructure, design-intensive details of various kinds) on the other hand. At the same time the complexity of the process encourages fluent transitions, and often several people (or no one at all) feel responsible for the same task. The process is individual, team-based and multi-disciplinary, enlisting multiple professional competencies and perspectives, at the same time. In this process, a principally unlimited solution space becomes more and more focused to be finally fixed in plans representing the artifact-to-be-built. The density of multi-disciplinary interaction and exchange varies from stage to stage and is not the same for all projects. All these exchanges require a considerable amount of mediating and translating which is partly supported by visual and technical conventions and standards. There is also a critical time aspect involved in design. Negotiations with relevant actors not only involve multiple complex issues but also connect to time consuming procedures, e.g., those of local authorities which follow their own logic of bureaucratic functioning and political compromising. The total time span from preliminary design to construction may be several years and it may happen that well thought out design decisions then turn out to be too costly or no longer technically feasible. 3
Artifacts and practices
Architects work with a large repertoire of artifacts - from sketches, scale models, images, and samples of material to CAD plans, detail drawings, Excel sheets, and Word documents. Many of these artifacts fulfill coordinative functions: As communication objects or 'persuasive artifacts' some may primarily help create a common understanding of a design idea or task, talk about a design in a rich, metaphorical way, and imagine qualities of space and appearance [40]. Some of them act as reminders of design principles, approach, method, open questions. Others help keeping track of activities and materials and again others represent design decisions at a certain level of detail and technical precision. Corresponding to these various and overlapping functions are sets of interrelated activities. Each artifact is surrounded by particular practices of producing, reading, annotating, modifying, checking, evaluating, etc. For example viewing the central CAD drawing in everyday work often involves making print-outs and/or photocopies in A3 format which are spread out on the table to be discussed, modified on a layer of transparent paper, or annotated with differently colored pens. Copies of these plans may be sent by fax to a consultant for commenting and return with suggestions and calculations.
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
263
The following sections describe a set of particular artifacts that have been collected as part of long-term fieldwork in an architectural office3. The selection has been made to illustrate the diversity of artifacts architects use and the practices that surround them. 3.1
Conceptual visualizations
First objectivities of a design concept often are represented through assemblies of sketches, metaphorical text, association images, physical models, and photographic material. While some architects use sketches and pictorial material for generating and expressing their ideas, others prefer poetry and metaphorical text, again others build their designs on (historical) research, the assembling of facts or 'datascapes' [21]. Again others work with scale models from the start of a project, working out their ideas by experimenting with different spatial configurations.4
The 'Big wall', impregnated with color and light The facade as screen, colorful patchwork (Charters) - a shimmering surface, bright and transparent, as seen from a distance, its structure revealing itself when approaching. The movie theatres stones that dip into water - above the surface of a rough, rocky quality, below precious stones that glitter in water - silver, gold, ruby, emerald. The facade as cutting edge between rough concrete and color.
Fig. 2 A collage of 3D visualization, association image, sketch and metaphorical text
The collage of 3D visualization in Fig. 2 (which was generated from a first, rough scale model), text, and association image represents some of the central features of the design concept for Pleasure Dome. 3D visualization and the image of 'Charters' visualize the idea of a 'big wall, entrenched with color and light'. Sketch and the metaphorical descriptions capture the evolving idea of the building's fa9ade as a cutting edge between the rough qual-
3
We here report on fieldwork carried out in 'Architekturburo Rudiger Lainer'. The artifacts that have been selected for this paper have their origin in several building projects and an urban planning study. We refer in particular to Pleasure Dome, an entertainment center in the Gasometer area in Vienna. The projects have been described in [44]. 4 So are e.g., Frank O. Gehry's (handmade) models digitized and then rationalized to achieve repetition without sacrificing form. In the case of the Walt Disney Concert Hall in LA the curved outer surface is covered by a 'skin' of Italian lime sandstone. The computer calculated the most economic way of cutting and producing the complex curvatures. A physical model was computer milled, compared to the original cardboard model and adjusted when necessary [20].
264
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
ity of concrete and the lucidity and colorfulness of the glass skin on the one hand, the movie theatres as 'stones that dip into water' on the other hand. The scale model was built at the very start of the project, and used within the office as well as in meetings with external specialist for visualizing the complex roof situation and internal space of the building. Later a much more accomplished version of this model was used in convincing the key user of the architects' color concept for the movie theaters to be painted in 'silver, gold, ruby, emerald'. Characteristic of these artifacts is their conceptual, and metaphorical nature. Sketches are quite good at capturing the mixture of symbolic richness and abstraction that allows expressing the qualities of space, light, atmosphere, and materials. Also, abstract 3D visualizations of spaces, places and artifacts may be used for conveying a concept, metaphor or shared cultural symbol. Abstract here does not mean the strive for purity (as in an abstract painting); on the contrary, visualizations like the 3D images produced in Pleasure Dome are highly theatrical. They use the language of "artistic impurity, hybridity, and heterogeneity" [22] for communicating certain ideas and qualities of an object. Another feature of these informal representational artifacts is their openness to extensions, modifications, and novel interpretations. 3.2
'Conceptual sheet'
Drawing, sketching, and assembling materials are activities that are often intermingled with talk. This particular example of a 'conceptual sheet' has been taken from an urban planning study. The architects produced it as part of a first planning session. It contains several elements: The artifact to the left contains, inter alia: •
A first work plan - things to do. phases, how to proceed. • Specification of visual material that should be collected or created (pictures, collages, association images, shadow plans, etc.) - how to represent the design of the urban area. • Metaphors - how to talk about the urban area. • A specification of methods - to define spatial qualities, to 'intensify rules', etc. • Explanatory sketches. • References to material to look for. • Names of responsible people.
Fig. 3 Work plan for an urban planning project
The architects use this type of artifacts in various ways. In this particular project members of the team placed copies of the sheet on their desks, using it as a reminder of design principles and the overall work to do. It served as a template for project meetings. In one of
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
265
those meetings the sheet was annotated and enriched. The sketches are pointers to a series of more detailed drawings exemplifying 'rules'. Finally, the sheet also represents the structure of the deliverable - a project report with different types of visualizations of the urban design. 3.3
The central CAD drawing
In a large building project, different people work on different parts of the building and on different problems. Typically, such a building is made up of 15-20 sections and about 30 plans altogether, including 11-12 floor plans, have to be drafted and coordinated. All plans are drawn with a CAD tool and stored on the central server, using a structured file system with different subdirectories for each project period and with predefined file-naming conventions. CAD plans (Fig. 4) assume a central coordinating role in the process of planning. They are true 'boundary objects' [30; 31], acted upon by all responsible actors and connected to specified procedures of approval and inclusion. First versions are created at a very early stage and they are gradually detailed and modified.
Fig. 4 CAD plan with layers
The CAD plan is the artifact in which all the design decisions that have been worked out in various forms - sketches, calculations, technical descriptions, product specifications, etc. are recorded and specified. This involves the work not only of the team of architects but of many external specialists. Within the office, people work on different parts and layers of the central CAD plans. Someone responsible for specific tasks such as 'fire escapes' may work concurrently on parts/layers used by others. This requires constant monitoring of concurrency and makes version control difficult. CAD plans cross organizational and professional boundaries many times. The construction engineer, for example, will view, comment and eventually correct the drawings at different stages of the planning process. S/he will receive the relevant layers of the CAD drawing and work on them. Other specialists may receive a print-out and produce their own drawings, which the architects will view, eventually discussing modifications and alternatives. They then may copy these drawings into their CAD plans or draft their version of the specialist's suggestion. Again others will receive a photocopy of one of the plans and return it with comments, calculations, sketches, etc. It is the architects who monitor and control this process of viewing, detailing, and adding to.
266
3. 4
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
Layered artifacts
Architects use a variety of techniques for communicating things that need to be taken account of or changed. Among those are: making annotations on a document, e.g., putting a red circle around a problem, adding details (correct measures, material), marking a part of a drawing with a post-it with some instructions for changes, corrections (e.g., in pencil directly on a plan), sketching either directly on a plan copy or on transparent tracing paper. In the left-hand part of Fig. 5 is a A3 copy of a CAD plan showing the upper level of the shopping mall in Pleasure Dome. This drawing was used in a meeting of the architect with the lighting designer. The lighting designer got to know the building, 'walking through', pointing to particular places and elements, while the architect was thinking aloud, describing the space, listening to questions and suggestions, simultaneously sketching the lighting concept. While talking, the architect developed a notation, using different colors for different types of lighting. This notation was then used in all documents concerning lighting. A common understanding of the concept, including solutions to some practical-technical problems, was achieved. The colored photocopies of the lighting design were then used as part of presentations to different audiences.
Fig. 5 Layered artifacts
In the top right-hand corner is a drawing for a competition on which the architects toy with different combinations of volumes and voids (in orange and yellow). Below, a blank transparent sheet of tracing paper is placed over a printed plan, and 'anchored' by entering some positional markers. The tracing paper is then used to experiment with design ideas. Layered artifacts facilitate coordination between activities (and the people who are responsible for them). They, for example, provide a collective or individual space for experimentation and change. The CAD drawing itself is a layered artifact, which builds on a particular mix of codes for functions and materials and has been tailored to a particular division of
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
267
labor. An architect who is experimenting with how to conduct a shaft through an open space may not only produce a series of sketches (some of the on tracing paper) but define a special layer for the drawing (e.g., Mike's layer). This example also shows that layers may denote ownership and/or professional competence (e.g., the construction engineer's). 3. 5
Ordering systems: the list of components and detail drawings
Fig. 6 A detail drawing
As planning progresses, more and more details have to be specified and filled in. A large building contains hundreds of details, which can either be left open, to be decided upon later by the construction company and/ or craftspeople, or carefully designed. Much of the quality and specificity of a building depends on these details (Fig. 6). Most details are drawn by hand, the main reason being that the computer system requires a level of precision which does not take account of the 'inexactness' of the building materials. Also, detail drawings are of a scale of 1:5 or even 1:1, and cannot be fitted into a CAD construction drawing. There are two types of details. Components (such as the fa9ade elements) are made of different parts, and the architect may wish to design a specific assembly of parts and materials. On the other hand, a building consists of a large number of joints between building elements and materials, which also may be specifically designed. The plan for a catalogue of components for Pleasure Dome was discussed in an internal project meeting: R: Planning of details - who has got an overview? - G. is in charge. R: To have a list of details would be important, including, what is relevant when? - I would like to have a 'total list' [Gesamtliste] and one 'actual list'. I also would like to have references to 'detail principles' included, e.g., for stairs - Hutte Klostemeuburg, for railings Absberggasse [references to previous projects]. We need such a list for achieving clarity concerning the details, e.g., everywhere closed metal sheets for the stairs outside - how does this fit with all the other stairs? - Or, we do have so many balustrades, some with glass, others ... The point is to coordinate the details in one's head
The main purpose was to generate a complete list of all elements and components, which then would be used for mass and cost calculations and for the call for tender. At the same time, the list should help to ensure conceptual consistency, i.e., that the same design principles and materials were used in different parts of the building. G. (who was in charge) started building a Word document (Fig. 7) that lists all the elements and components ('Aufbauten') to be used in the building - (inner and outer) walls, ceilings, floors, roofs, stairwells, balustrades, etc. This Word list, which was compiled in the office,
268
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
went through several cycles of discussion and annotations, involving construction engineer and building engineer. Small sketches were made, showing the design principles for some components. In parallel, the architects began drawing details.
Fig. 7 The list of components - draft and final document
All detail drawings are listed in an additional 'coordinative artifact' - the Excel sheet 'Detailiibersicht' (overview of details, Fig. 8). Each detail drawing has been assigned a 3-digit detail number, with the digits referring to: type of detail (e.g., interior glass elements), part of building (e.g., mall), element or component (e.g., door to projection cabin). Details are referred to within CAD drawings by their number and framed. The Excel sheet provides an index to the detail drawings which are kept in a binder. Detail list and detail drawings are used together. The binder is located centrally in the room, which is shared by the people working on the construction plans. Whenever someone needs information about details, s/he walks over to the table, searches for the documents, takes them out for photocopying, and places them back in the binder. The detail list contains information about completion and modification dates and helps maintain an overview of the circulation of detail drawings within the network of people involved in planning and building. People can see on the list which detail needs to be sent to whom, either for comments or for approval. The sheet also tells who received a particular detail drawing. Ordering systems type of artifacts play a large role in the architects' everyday work. They are consulted all the time.
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
269
Fig. 8 List of detail drawings
4
Discussion
Our fieldwork material indicates that artifacts play a wide variety of integrative roles in cooperative work and also helps us understand what makes artifacts particularly amenable to coordination. Here we want to very briefly point out a few characteristics of the artifacts found in this setting.
4.1
The multiplicity of artifacts in architectural practice
First of all, the very fact that the setting is so utterly full of artifacts may seem paradoxical, in as much as architectural work is 'knowledge work' par excellence. Nonetheless, although as different from a factory or a power plant as imaginable, the setting is replete with all sorts of material artifacts such as drawings, binders, photos, plans, lists, models, i.e., artifacts that, by contrast to artifacts one will see in every office (such as walls, doors, desks, and chairs), are specific to the trade of architects, domain specific, that is. These artifacts are to be found on all surfaces in the office, on walls, shelves, and desks. In order to understand the plethora of artifacts, one should take into account that architectural work is different from many other types of work insofar as the field of work does not exist, that is does not exist objectively, in advance but is constructed in and through the process of design and planning and, ultimately, construction. Architectural work proceeds through the architects' producing successive objectivations of the design and interacting with them in a variety of ways. That is, the conspicuous display of architectural artifacts can be seen as the fundamental means of making the not-yet-existing and in-the-process-ofbecoming field of work immediately visible, at-hand, tangible. A comparison with radically different work domains such as process control may help to clarify the point. To the operators of the power plant the plant as a whole, the different functional parts, the valves and pumps, the energy transformation processes, the mass flows, the power grid to which it is connected, etc. can be conceived of as their common field of work. It is there, in an important sense, before they start their shift, and it is still
270
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
there when they go home again. However, due to the sheer scale of the plant as well as due to the intangibility of the processes, the various representations of the plant and processes in the control room are taken to stand proxy for the plant and processes beyond the control room. For all practical purposes, they work with representations. By contrast, the field of work of architects is notional. Not only does the building not exist prior to their work but only as a result of their work; the objectivations of the design do not exist prior to their work either. This is of course an exaggeration, as architects reuse previous designs and have a vast array of preexisting resources at hand, such as catalogues of materials, parts, etc on the market. Anyway, the point we are trying to make is that the artifacts that the motley collection of architectural artifacts play a role quite similar to the nuclear power plant and the control room representations of it to the operators. That is, in the absence of a preexisting, material field of work, architectural representational artifacts constitute the field of work of architectural work. Thus, as objectivations of the construction-in-the-making, architectural artifacts are the immediate object of architects' work. In the absence of the building, they are what is looked upon, inspected, gestured at, discussed, modified, annotated, etc. In short, they provide a rich resource for orderly interaction. These similarities notwhithstanding, the fact that the field of work of architects is notional and the artifacts merely objectivations of things-to-come and as such representational artifacts, has important implications. Representations are not the real thing, of course; they are fundamentally 'under-specified' [32] with respect to that which is represented. Representations are local and temporary constructs [2]. And representations are conventionalized practices based on rules of mapping and translation between representation and the object that is represented. Consequently, the infinity of affordances and cues offered by the preexisting objective field of work of plant operators is not available but has to be painstakingly emulated by a rich variety of sign systems, notations, and other conventional practices. Thus, architects' 'conceptual visualizations' as well as their CAD drawings describe the building-in-design on multiple levels of detail, completeness, and 'technicality', using different visual languages. These rules may have to do with scale, with material (with a notation denoting the different kinds of materials), construction, etc. The plethora of artifacts is an expression of the abstractness of representational artifacts and of their limited scope. To architects this 'under-specification' of representational artifacts is of particular relevance, as one can see from 'conceptual visualizations' such as sketches and association images. Connected with this is the 'openness' of such artifacts to facilitate and accommodate the contributions of others, and thus to stimulate their imagination and to eventually perceive the novel within the familiar, to discover relations between seemingly incongruent objects and notions - to relate the 'unrelatable', and to jointly take a step further in the design process [41]. 4. 2
The coordinative roles of artifacts
This multiplicity of coordinative artifacts in architectural practice amply demonstrates that artifacts have characteristics quite distinct from mental constructs and that the use of artifacts cannot simply be grasped and understood in terms of 'representational media' or 'precomputation'. There is no question, of course, that an artifact can serve as an representation of something else. A signature on a piece of paper can be taken to represent the handover of property from the signatory party to another party. A map can be taken to represent a particular territory to the extent that one can use the map to measure distances between e.g. crossroads and plan a journey in detail. This is trivial. There is also no question that artifacts can be
K. Schmidt and 1. Wagner / Coordinative Artifacts in Architectural Practice
271
used to express ideas and plans and other mental constructs. The sketches and drawings of the architects evidently express ideas the architects had 'in mind'. This is also trivial. But these artifacts do not simply serve as containers of preconceived 'mental constructs'. The relationship between cognition and drawing is infinitely more complex than that [1; 28]. Sketching and drawing are also activities of exploration and testing. In addition to or, rather, by virtue of their representational role, the artifacts we have described serve a coordinative function. As they are being used (or not used) in the cooperative effort, their changing state (or static state) offers cues to other actors as to the intentions of the actor or actors effecting the changes. As we can see from the examples of 'layered artifacts', the simple fact that, say, a particular drawing is open in front of a colleague who has placed a transparent overlay on top of it, that the colleague is bending forwards while sketching some modifications on the overlay, may or may not have implications for a colleague working on an intersecting task. The simple fact that a pile of plans is marginally positioned on a desk, as opposed to centrally, may indicate that the architect at that desk may have finished working with them but is not quite sure that she will not need them again. In general, the state of artifacts in the work setting — especially artifacts and other features of the field of work — provides an infinitely array of signals and cues for cooperating actors to effortlessly apperceive the intentions of colleagues, the challenges and problems they are facing, etc. That is, the coordinative role of artifacts is not incidental. It is because these artifacts serve as objectivations of the design-in-progress that they have coordinative functions: architectural work is for all practical purposes done with, on, by means of these artifacts and is thus made immediately publicly visible to competent members. Furthermore, there is a specific class of coordinative artifacts that have deliberately been designed to serve coordinative purposes, such as the list of details and the list of detail drawings. We have elsewhere suggested to call functional complexes of such artifacts and protocols (comprising various artifacts, classification schemes, notations and other protocols) ordering systems [27]. In the case of architectural work, the binder system, the layer organization of the CAD system, the plan identification system (comprising plans, the code for numbering, and the circulation list), and the system of interconnected lists of components and detail drawings are clear examples of such complexes of artifacts and protocols. Their primary function consists in enabling actors to maintain some kind of order in the vast collection of distributed items required to objectify the in-the-process-of-becoming field of work. That is, their coordinative function is different from that of sketches, drawings, models, etc. They are not objectivations of things-to-come but rather normative constructs governing the distributed activities of the project. 4. 3
The materiality of artifacts
Expressing an idea or a plan in a material artifact involves practices that transcend what can be subsumed under conventional notions of representation etc. Material artifacts are publicly accessible. Their state can be inspected by other members. Where they are located can be observed by and made sense of by members. What others are doing to an artifact can be noted and made sense of. As persistent graphical objects, architectural artifacts and configurations of such artifacts can be visually taken in simultaneously, at a glance. They thus offer modalities of interaction that are fundamentally different from the sequential order of speech and action. Artifacts are tenacious and their state and the configurations in which they have entered may transcend the situation at hand and may be instrumental in imposing order beyond the horizon of the immediate situation. In contrast to a digital text or image, a material artifact bears witness to its history, it shows wear and tear. A print-out of a CAD drawing which has been annotated carries the history of past
272
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
work as well as the marks of the contributions of different people, as identified from their handwriting or from the color of the pencil. The faded color on a binder tells about age, hand-written additions to the label on its back indicate that the classification has had to be extended, etc. The use of coordinative artifacts is connected to their specific physical and graphical form. The way in which elements of text, sketches, and arrows are laid out and organized in the conceptual sheet (Fig. 3) is meaningful and may indicate relevance. A CAD drawing consists of a great number of conventions, notations, and layers, from which the different professions involved in the planning process extract the information they need. When it is plotted out in large format, an experienced architect my identify on the spot an unresolved problem in the myriad of lines and signs. Closer investigations would need to consider in detail the conventions of notation, format, and syntax underlying their form and use, such as the specific techniques involved in working with maps, charts, matrices, trees, or linear text [cf. 7, pp. 91 ff et passim; 23; 37; 38]. We are, of course, not alone in pointing to the materiality of artifacts. Many authors have for instance suggested to use Gibson's [3] concept of affordances. Sellen and Harper have for example pointed out that the affordances of paper versus digital documents are quite different and have indicated some of the affordances of paper [29]. Similarly, Latour [19] not only speaks of the 'immutability' of inscriptions on paper, but also of their mobility, due to the paper's flatness, small scale, and inexpensive reproducibility. This involves and enables a totally different set of practices than those connected with, say, an inscription in marble. While the concept of affordances of artifacts is fundamental to an analysis of the use of material artifacts, it is not sufficient for addressing the very intricate interrelationships and interactions between the materiality of artifacts upon which inscriptions are made and the logic of the sign system that is being applied. Not only are different scripts adapted to the nature of the surface upon which it is being inscribed, as argued by Roy Harris: "The use of wax tablets in ancient Rome, baked clay in Babylon, pattra in India, tortoise shell in China, is not unrelated to the form of some of the scripts developed in those regions" [7, p. 30]. It is untenable to conceive of the 'information', the 'content', the 'message' etc. as invariant, irrespective of the technique of inscription: 'Writing with a reed stylus on wet clay is manifestly such a different enterprise from operating a modern printing press that it becomes legitimate to ask: is there any semiological unity underlying this diversity?' [7, p. 7]. That is, in order to understand the uses of coordinative artifacts in cooperative work we need to investigate in detail the techniques and practices of using ordinary as well as representational artifacts. This work has hardly been undertaken vet. Acknowledgments We are indebted to the staff of Architekturburo Rudiger Lainer in Vienna for giving us access to their work. The research has been supported by the Danish National Centre for Multimedia Research under the DMM project and by the Danish Research Councils' Program on Information Technology under the DIWA project. We would finally like to thank Liam Bannon. Francoise Darses, Gloria Mark. Dave Randall for very useful critical comments.
K. Schmidt and I. Wagner / Coordinative Artifacts in Architectural Practice
273
References [1] Gedenryd, Henrik: How Designers Work, Ph.D. diss., Department of Cognitive Science, Lund University, 18 November, 1998. [2] Gerson, Elihu M., and Susan Leigh Star: 'Analyzing due process in the workplace,' ACM Transactions on Office Information Systems, vol. 4, no. 3, July 1986, pp. 257-270. [3] Gibson, James J.: The Ecological Approach to Visual Perception, Houghton-Mifflin, Boston, 1979. [Reprint published by Lawrence Erlbaum Associates, Hillsdale, New Jersey, 1986]. [4] Harper, Richard H. R., and John A. Hughes: 'What a f—ing system! Send 'em all to the same place and then expect us to stop 'em hitting: Managing technology work in air traffic control,' in G. Button (ed.): Technology in Working Order: Studies of Work, Interaction, and Technology, Routledge, London and New York, 1993, pp.127-144. [5] Harper, Richard H. R., John A. Hughes, and Dan Z. Shapiro: The Functionality of Flight Strips in ATC Work. The report for the Civil Aviation Authority, Lancaster Sociotechnics Group, Department of Sociology, Lancaster University, January, 1989. [6] Harper, Richard H. R., John A. Hughes, and Dan Z. Shapiro: 'Working in harmony: An examination of computer technology in air traffic control,' in ECSCW'89: Proceedings of the First European Conference on Computer Supported Cooperative Work, Gatwick, London, 13-15 September, 1989, 1989, pp. 73-86. [7] Harris, Roy: Signs of Writing, Routledge, London and New York, 1995. [8] Harris, Roy: Rethinking Writing, Indiana University Press, Bloomington and Indianapolis, 2000. [9] Harrison, Steve, and Scott Minneman: 'A bike in hand: A study of 3-D objects in design,' in K. Dorst, H. Christiaans and N. Cross (eds.): The Delft protocols workshop: Analyzing design activity, Delft, The Netherlands ,20-22 September 1994, 1994, pp. 205 - 218. [10] Heath, Christian C., and Paul Luff: 'Collaborative activity and technological design: Task coordination in London Underground control rooms,' in L. J. Bannon, M. Robinson and K. Schmidt (eds.): ECSCW'91: Proceedings of the Second European Conference on Computer-Supported Cooperative Work, Amsterdam, 24-27 September 1991, Kluwer Academic Publishers, Dordrecht, 1991, pp. 65-80. [11] Heath, Christian C., and Paul Luff: 'Collaboration and control: Crisis management and multimedia technology in London Underground control rooms,' Computer Supported Cooperative Work (CSCW): An International Journal, vol. 1, no. 1-2, 1992, pp. 69-94. [12] Heath, Christian C., Paul Luff, Hideaki Kuzuoka, and Keiichi Yamazaki: 'Creating coherent environments for collaboration,' in W. Prinz at al. (eds.): ECSCW 2001: Proceedings of the Seventh European Conference on Computer Supported Cooperative Work, 16–20 September 2001, Bonn, Germany, Kluwer Academic Publishers, Dordrecht, 2001. - In press. [13] Hutchins, Edwin L.: 'Mediation and automatization,' Quarterly Newsletter of the Laboratory of Comparative Human Cognition [University of California, San Diego], vol. 8, no. 2, April 1986, pp. 47-58. [14] Hutchins, Edwin L.: Cognition in the Wild, The MIT Press, Cambridge, Mass., and London, England, 1995. [15] Hutchins, Edwin L., and Tove Klausen: 'Distributed cognition in an airline cockpit,' in Y. Engestrom and D. Middleton (eds.): Cognition and Communication at Work, Cambridge University Press, Cambridge, 1996, pp. 15-34. [16] Kaptelinin, Victor: 'Computer-mediated activity: Functional organs in social and developmental contexts,' in B. A. Nardi (ed.): Context and Consciousness: Activity Theory and Human-Computer Interaction, The MIT Press, Cambridge, Mass., 1997. [17] Kasbi, Catherine, and Maurice de Montmollin: 'Activity without decision and responsibility: The case of nuclear power plants,' in J. Rasmussen, B. Brehmer and J. Leplat (eds.): Distributed Decision Making. Cognitive Models for Cooperative Work, John Wiley & Sons, Chichester, 1991, pp. 275–283. [18] Kuutti, Kan: 'Activity Theory as a potential framework for Human-Computer Interaction research ' in B. A. Nardi (ed.): Context and Consciousness: Activity Theory and Human-Computer Interaction, The MIT Press, Cambridge, Mass., 1997. [19] Latour, Bruno: 'Visualization and cognition: Thinking with eyes and hands,' in H. A. Kuklick and E. Long (eds.): Knowledge and Society: Studies in the Sociology of Culture Past and Present. A Research Annual, vol. 6, JAI Press, Greenwich, Conn., 1986, pp. 1-40. [20] LeCuyer, A.: 'Design on the Computer: Frank O. Gehry und Peter Eisenman,' ARCH+, no. 128, September 1995, pp. 26-29. [21] Maas, W., and J. Van Rijs: FRANAX: Excursions on Density, O10 Publishers, Rotterdam, 1998.
274
K. Schmidt and 1. Wagner / Coordinative Artifacts in Architectural Practice
[22] Mitchell, W. J.. Picture Theory: Essays on Verbal and Visual Representation, The University of Chicago Press, Chicago, 1994, [23] Olson, David R.: The World on Paper: The Conceptual and Cognitive Implications of Writing and Reading, Cambridge University Press, Cambridge, 1994. [24] Robinson, Mike: 'Design for unanticipated use...,' in G. De Michelis, C. Simone and K. Schmidt (eds.): ECSCW'93: Proceedings of the Third European Conference on Computer-Supported Cooperative Work, Mi lan, Kluwer Academic Publishers, 1993, pp. 187–202. [25] Schmidt, Kjeld (ed.): Developing CSC W Systems: Design Concepts. Report of CoTech WG4, February 1993, Riso National Laboratory, Roskilde, Denmark, 1993. [26] Schmidt, Kjeld: Modes and Mechanisms of Interaction in Cooperative Work, Rise National Laboratory, P.O. Box 49, DK-4000 Roskilde, Denmark, 1994Riso-R-666(EN)]. [27] Schmidt, Kjeld, and Ina Wagner: 'Ordering systems in architectural design and planning: A discussion of classification systems and practices,' in G. C. Bowker, L. Gasser and B. Turner (eds.): Workshop on Infrastructures for Distributed Collective Practice, San Diego, 6-9 February 2002. [28] Sch6n, Donald A.: The Reflective Practitioner: How Professionals Think in Action, MIT Press, Cambridge, Mass., 1983. [29] Sellen, Abigail, and Richard H. R. Harper: The Myth of the Paperless Office, MIT Press, Cambridge, Mass., 2001. [30] Star, Susan Leigh: 'The structure of ill-structured solutions: Boundary objects and heterogeneous distributed problem solving,' in L. Gasser and M. Huhns (eds.): Distributed Artificial Intelligence, vol. 2, Pitman, London, 1989, pp. 37-54. [31] Star, Susan Leigh, and James R. Griesemer: 'Institutional ecology, 'translations' and boundary objects: Amateurs and professionals in Berkeley's Museum of Vertebrate Zoology, 1907–39,' Social Studies of Science, vol. 19, 1989, pp. 387-420. [32] Suchman, Lucy A.: Plans and Situated Actions: The Problem of Human-Machine Communication, Cambridge University Press, Cambridge, 1987. [33] Suchman, Lucy A.: 'Technologies of accountability: On lizards and airplanes,' in G. Button (ed.): Technology in Working Order. Studies of work, Interaction, and Technology, Routledge, London and New York, 1993, pp. 113-126. [34] Suchman, Lucy A.: 'Constituting shared workspaces,' in Y. Engestrom and D. Middleton (eds.): Cognition and Communication at Work, Cambridge University Press, Cambridge, 19%, pp. 35–60. [35] Suchman, Lucy A., and Randall H. Trigg: 'Understanding practice: Video as a medium for reflection and design,' in J. Greenbaum and M. Kyng (eds.): Design at Work: Cooperative Design of Computer Systems. Lawrence Erlbaum, Hillsdale, New Jersey, 1991, pp. 65–89. [36] Taylor, Talbot J.: Mutual Misunderstanding: Scepticism and the Theorizing of Language and Interpretation, Duke University Press, Durham and London, 1992. [37] Tufte, Edward R.: The Visual Display of Quantitative Information, Graphics Press, Cheshire, Connecticut, 1983. [38] Tufte, Edward R.: Envisioning Information, Graphics Press, Cheshire, Connecticut, 1990. [39] Vygotsky, Lev Semonovich: 'The instrumental method in psychology' (1930); in L. S. Vygotsky: The Collected Works of L. S. Vygotsky. Volume 3: Problems of the Theory and History of Psychology, Plenum Press, New York and London, 1997, pp. 85–89. - Theses of a talk read in 1930 at the N. K. Krupskaya Academy of Communist Education. [40] Wagner, Ina: 'Persuasive artefacts in architectural design and planning,' in Proceedings of CoDesigning 2000, Nottingham, 11–13 September 2000, 2000, pp. 379-390. [41] Wagner, Ina, and Rainer Lainer: 'Open planning: inspirational objects, themes, placeholders, and persuasive artefacts,' in Proceedings Colloque Architecture des systemes urbains, Universite de Technologic de Compiegne, 5 July 2001, 2001. [42] Wertsch, James V.: Vygotsky and the Social Formation of Mind, Harvard University Press, Cambridge, Mass., and London, 1985. [43] Wertsch, James V.: Voices of the Mind: A Sociocultural Approach to Mediated Action, Harvard University Press, Cambridge, Mass., 1991. [44] Zschokke, W.: Rudiger Lainer: Urbanism, Buildings, Projects, 1984-1999. Birkhauser, Basel/Boston, 1999.
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press, 2002
275
Co-Constructing Collaborative ComputerBased Artifacts and Mentefacts Peter MAMBREY, Bettina TORPEL Fraunhofer Institute for Applied Information Technology CSCW Unit, Evaluation and Design of Collaborative Systems Department Schloss Birlinghoven, D-53754 Sankt Augustin, Germany
[email protected],
[email protected]
Abstract. In this paper we critically look at our own empirical work done during the past ten years done in our research institute in the field of Participatory Design for Computer Supported Cooperative Work. The identified shortcomings and experiences have guided our search for better methods in the fields evaluation and design of collaborative systems. We have looked for approaches for co-constructing change processes and future work in communities of practice instead of representing the existing status quo of work and work practice. The co-construction we are envisioning comprises communication processes on issues such as work practice, characteristics of the organization and computer applications as technological infrastructure. We will introduce four procedures supposedly evoking a variety of possible future work settings in the minds of the different participants as an integral part of the design process. Shared artifacts and mentefacts enable the discussions on design issues. The outlined approach is expected to meet the requirements of an appropriate Participatory Design approach for particular communities of the "New Economy". Keywords: CSCW, Participatory Design, Co-construction, artifacts, mentefacts.
Overview In this contribution, we will first argue that a new participation orientation has emerged in the New Economy (section 1). We will then familiarize the reader with research and design approaches we used in past Participatory Design projects conducted in our research institution (section 2). After an ex-post evaluation we will self-critically conclude that they bear specific weaknesses regarding a) participation orientation, b) their possibilities to handle the gaps between description, analysis and construction and c) their orientation toward the status quo vs. toward the future. We will then introduce an approach for discoursively co-constructing collaborative systems exemplified by four procedures (section 3). In particular, we will depict their potentials to overcome the shortcomings of the former approaches.
276
P. Mambrey and B. Torpel / Co-Constructing Collaborative Computer-Based Artifacts
1 The Transformation of Work Practice Suggests New Approaches for Design For a long time, it has been diagnosed that certain perspectives and modes of practice seem to radically exclude each other. Charles P. Snow, for example, as an early observer, states about the relation between the humanities and the sciences: "Literary intellectuals at one pole - at the other scientists, and as the most representative, the physical scientists. Between the two a gulf of mutual incomprehension - sometimes (particularly among the young) hostility and dislike, but most of all lack of understanding" (Snow, 1959/1993). The same kind of seemingly unbridgeable gap has existed between areas like technology production vs. technology use (Suchman 1994b), engineering vs. social science and the outside perspective of "the third person" (she/he) vs. the subjective perspective of "the first person" (I). The mutual incomprehension and ignorance is rooted in the practices of the proponents of the "poles" of the contradictions. Due to established rituals, socialization sequences and modes of dividing labor, common practice does not provide reasons for getting acquainted with the practices of other communities. The communities and practices of many engineers and other design specialists still provide few reasons for developing an understanding of the work practices of other people, including potential users of one's own products. Complementary to this, common practices of many non-engineering professions do not provide reasons for exploring modelling and formalization techniques - not even if it is to determine one's own technological infrastructure and even if one's professional knowledge comprises skills highly relevant for determining a supporting technological infrastructure. Even if we had an ideal design situation in terms of (disciplinary) cultures, another problem for the development of computer applications would remain: abysses also separate the different activities comprised in developing computer applications. It has been one of the most powerful myths of computer science that there is an automatical and smooth transition between activities like collecting data about the potential use situation, analyzing these data and constructive activities. The one necessary sequence of activities is as much an ideological construction as the assumption that the result of one design activity seamlessly translates into content for another activity. Specifically, the description of a given work situation does not seamlessly result in - or translate into - criteria for and catalogues of shortcomings. The same is true for the transitions between identifying shortcomings and generating features of an improved situation or between generating ideas for functionalities and the means and structures for realizing them (e. g. tools, procedures, hardware and software modules). Instead, the transitions between design activities require decisions. These decisions remain necessarily arbitrary; they are bare of any inherent necessity. Many IT professional mistakenly believe in their own and their profession's capability for seamlessly translating instead of arbitrarily constructing. While constructing conditions for future work they are convinced that they merely translate. As long as constructive activities are not recognized as constructive, they appear to be beyond any negotiability and alterability: existing practices, structures and ideologies literally become codified. We think that, instead of denying arbitrariness, negotiability and alterability the design gaps should be taken into account and their seriousness should be reduced. One possibility to reduce the seriousness is to cooperatively generate multiple possible options, make their qualities explicit and negotiate if they should be used or abandoned. This is the main objective of this paper. It has been a persistent task for application system design, especially for Participatory Design, to act in the face of these gaps. The proceedings of the biannual Participatory Design conferences are written records of a multitude of theoretical and empirical efforts relevant for this task (Cherkasky, Greenbaum, Mambrey & Pors, 2000). Similarly, a significant portion of the work reported on at the Computer Supported Cooperative Work conferences has had its focus on approaches for the socio-technical design of collaborative environments integrating work practice, users and technology.
P. Mambrey and B. Torpel / Co-Constructing Collaborative Computer-Based Artifacts
211
Traditionally, the very possibility of participatory approaches to the development of computer applications was either unknown or rejected by the management for political reasons. Proponents of Participatory Design argued 1. that tools co-determined by workers were productivity-enhancing because they are accepted by the workers and 2. that workers should have the right to design their own work means (see e. g. Greenbaum, 1993). In recent years, work has undergone a change process toward fragmentation (cf. Baukrowitz, Boes & Schmiede, 2000, Torpel, Wulf & Kahler, 2001): • Cooperation takes place in geographically distributed work environments. • The notion of the clear "poles" of capital vs. labor have become inappropriate in the face of new working arrangements like alliances of worker-entrepreneurs ("selfemployed labor", cf. Torpel, 2000). • Unions, especially in Europe, have lost influence partly because they focus on the representation of worker in traditional large organizations. • New management strategies include the delegation of responsibilities to units acting semi-autonomously on the basis of their own budget and an organizational culture stressing a participative and cooperation-oriented atmosphere. Instead of through boards of shop stewards, negotiations between employer and employees take place via personal communication in peer groups based on shared values. The division of labor and coordination and control are not formalized and structured a priori but according to the inclinations and competences of the individual workers. Work is based on a sense of "common ground". This sense is continuously reproduced while working in small projects lasting a limited period of time. • Semi-autonomous, budgeted and/or project work implies short innovation cycles for hard- and software. Working arrangements in the face of trends like these have become diverse and differentiated, relations between freedom and restrictions have become as diverse as the arrangements themselves. Possibly, work practice has been more fragmented anyway, and for a longer time than assumed; current practices in fragmented working environments might have sensitized us for the issue of fragmentation. Within some of the current work environments a necessity for a new kind of participation seems to have developed (cf. Torpel et al., 2001). For fragmented work environments we refer to the involved individuals and groups in terms of communities of practice (Lave & Wenger, 1991) rather than in terms of organizations and their units. One reason for this shift in perspective is the inseparablity of formerly separable spheres like the spheres of work vs. leisure or work vs. political activism (cf. Robinson, 1999). Building and extending appropriate support for communication and cooperation in geographically distributed communities has been an ongoing challenge. Cooperation in many fragmented communities rests on exchange and negotiation. Therefore, particularly these communities seem to require discoursive participatory approaches for the development and improvement of their technological cooperation support.
2 Our Past Contributions to Participatory Design Looking back, we deem the empirical and Participatory Design methods we employed in the past ten years to be in need of improvement. During a long-term research and development project, aimed at the introduction and improvement of a groupware system supporting cooperation in a large geographically distributed organization (Sohlenkamp, Mambrey, Prinz, Fuchs, Syri, Pankoke-Babatz, Klockner & Kolvenbach 2000), we gained insights relevant for our new design approach (see below in section 3). In this project, we largely followed the idea of ethnographic analysis and description and focussed on the
278
P. Mambrey and B. Torpel / Co-Constructing Collaborative Computer-Based Artifacts
work practice within the organization. One of our approaches is exemplified by our analysis of the information flow of a letter resulting in a written scenario (Mambrey & Robinson, 1997). We relied on the individual user as informant (section 2.1). A characteristic example for our second approach is a research process in which we identified a a boundary object. Here we relied on an artifact as an informant for design (section 2.2). A third approach rested on principles of action research instead of ethnography. We installed user advocates (Mambrey, Mark & Pankoke-Babatz, 1998) as representatives of the users (section 2.3). Each approach relied on specific main informants for design: individuals as informants, artifacts as informants and work practices as informants with their messages conveyed by user representatives. The approaches complemented each other and were used parallelly. In the following, we will describe these approaches with the specific kinds of insights they allow for and with their specific shortcomings regarding participatory design. 2. 1 Users as Informants To analyze the path of a letter sent to a German ministry, we first conducted open interviews on how, according to each interviewee, the letter was passed on in the ministry. Based on the descriptions given by employees on unit level, section leader level, and shop steward level, we wrote a meticulous scenario (Mambrey & Robinson, 1997) and discussed it with the interviewees. We rewrote the scenario integrating the comments and corrections until it was accepted as a precise representation of the path of letters seen from the employees' perspectives. The description showed that there are two complementary and inseparably entwined spheres of work in the ministry: hierarchical, strict formal work and officially invisible, informal and "private" cooperative work. Availing both spheres' potentials makes this kind of organization powerful and stable; it allows for subtle subversive practices within a necessary framework of strict accountability. Explicating the inofficial path of the letter provided a "thick description" of a cooperative process. The procedure only focused on the present state of affairs from an outside research perspective instead of collaboratively constructing future options. It also did not provide clues on how to proceed from the account of the present to the construction of future artifacts, practices and structures. Serving as interview partners and cross-readers of the written account of the path of the letter is not a far-reaching participation. The users were not even involved in choosing the research topic and strategy and the criteria of the analysis. 2.2 Artifacts as Informants We had the opportunity to follow the process of writing a politically sensitive speech for a German minister. A German ministry is organized according to the Weberian ideal of a rational organization: strict hierarchy, top-down control, transparency of all processes and artifacts, fixed responsibilities, no redundancy etc. The minister initiated the process of composing the speech and the head of the Office of the Minister wrote a formal request letter. The letter was passed on to a defined next person who passed it on to another defined person, and so on, until the collaborator at a lower department received it. Each person signed and dated the letter initiating its workflow. A draft of the speech and an accompanying letter were added from the collaborator at the lower department. The compound artifact was sent back and forth between a chancery clerk and the collaborator before it started returning along the path of hierarchy. At the end, it contained inscriptions from the different individuals responsible at each department: record approvals, navigation paths, date information and deadlines. We modified its description until the persons involved agreed with it.
P. Mambrey and B. Torpel / Co-Constructing Collaborative Computer-Based Artifacts
279
The letter can be described in terms of the concept of the boundary object. For Star (1989) "boundary objects are those objects that are plastic enough to be adaptable across multiple viewpoints, yet maintain continuity of identity" (p. 37). According to Star & Griesemer (1989), the creation and management of boundary objects is a key process in developing and maintaining coherence across intersecting social worlds. The letter inhabits several intersecting social worlds within the Ministry. It is abstract enough to adapt to local needs since its schema does not overspecify work, yet further details, for example for collaboration, may be added as needed. It has (is) a simple common structure, with different meanings prevailing in different social worlds. Depending on the specific context, it provided different information and suggested different actions. After this research the designers used the metaphor of a medium instead of mechanism for the processes and they modelled them as multi-perspective group work rather than as prescribed workflows. The analysis showed the complexity and subtlety of the work practice and therefore was helpful for design. However, it did not provide ideas of how to design the future work practice potentially assisted by a groupware. Again, the procedure remained restricted to gaining insight to the status quo of a particular practice, the process of composing and approving of a speech. The procedure did not guide actions linking analysis and constructive activities. The practitioners were observed from an outside perspective instead of determining what was analyzed and how and they did not participate in the constructive activities. 2.3 Practices as Informants and User Advocates as Conveyors An action oriented approach for informing designers about the users' needs was installed by creating the role of "user advocates" (Mambrey, Mark & Pankoke-Babatz, 1998). Through their education and experience, user advocates could take the perspectives of both designers and users (the general advantages of approaches where informants are able to switch perspectives are outlined in Ruhleder, 1995). They explored the users' needs in their daily work, including actual system use, and propagated the proposals and requirements of the users to the designers. As advocates of the users they controlled the design decisions as well as the transformation of the requirements into practical/technical measures. They learned about the users' needs through observation and discussion during visits to the worksite, at workshops with the users, and via a support hotline they had installed. The idea of having a consultant with expertise working together with customer groups has its theoretical origin in models of process consulting (Schein, 1988) and in urban city planning advocacy approaches. During this experience, the question of technological conservatism vs. innovationorientation was raised: some designers complained that the requirements and ideas of the users only allowed for the "electrification and digitalization" of the already existing work practice. The approach is future-oriented in the sense that design options are generated in any case but were only introduced in the work environment when they had the approval of the user advocates. The negotiations between designers and user advocates certainly tackled the problems involved in the transitions between collecting data, analysis and construction. A discussion of practices, goals and visions directly involving users was missing. 3 A New Approach to Design: Methods for Co-Constructing a Possible Future Reality 3. 1 The Framework of the Approach As a working group, our self-set goal has been to develop methods which enforce organizational change (if it is beneficial) and enable participants (designers and users) to jointly invent and construct future collaborative work practice and digital cooperation
280
P. Mambrey and B. Torpel / Co-Constructing Collaborative Computer-Based Artifacts
platforms. In accordance with a "communities of practice" view (Lave & Wenger, 1991) of work environments we focus on ways in which knowledge and meaning are constructed and distributed in particular working situations: "Knowledge, in this view, is thought of as the ability to participate meaningfully; learning is seen as the process of becoming a member of the working community of practice; and tools are reconceptualized as resources that facilitate integration and interaction within the group so that it can carry out its business." (Jordan, 1993, p. 186). Our design approach involves processes of "objectifying subjective experiences": Collecting and analyzing subjective accounts and confronting these accounts and results with each other yields intersubjectively validated experiences. Re-constructing already existing facts, artifacts, practices and modes of thinking is supplemented by the discoursive generation/construction of mentefacts (as we call constructions of thought anticipating artifacts), artifacts, practices and modes of thinking. Using approaches that stimulate certain modes of imagining (e. g. visual, metaphoric, textual) helps to create mentefacts that can mentally and discoursively be altered, manipulated, tested and evaluated. Making choices for particular mentefacts, as potential alternatives for the future, is followed by the creation of real artifacts, infrastructures and practices in real-life settings. Change is an integral part of the design process. The ultimate success criterion for the design process is the perceived quality of life of the participants. Following Holzkamp (1983), we assume quality of life as reflecting how much someone can access and influence subjectively relevant conditions. In our case, these conditions would be technological infrastructures, work practices and/or organizational features. When the process was successful it resulted in improved real-life conditions. Otherwise the design activities could be taken up again. Having confronted reality with new concepts and artifacts and having confronted the concepts and artifacts with reality provides a new level for further design processes. The participants of the design communities may be practitioners from the most diverse areas of practice. They follow the superordinate goal of improving their work environments; the special design objectives emerge from the necessities experienced in every-day practice they perceive as in need of improvement and worth improving. What appears to be abstract change potentials from an outside perspective is experienced as a new set of challenges and requirements by individuals and communities. The design activities may become a new practice in its own right. Design groups emerge from practices the participants deem as in need of improvement and which can be improved. For the kind of design process we are proposing it is crucial that design problems and the generated solutions are rooted in their personal experience. The participants influence practice and hence restructure collaborative formations. Even though this might take place within organizations, it is not restricted by organizational boundaries. Shared or similar practice and the pertaining experiences (negative ones such as breakdowns or failures within a particular socio-technical setting, positive ones such as support, success, efficiency or agency) result in more realistic design communities than design groups formed within the boundaries of organizations or organizational units. The design approach comprises and extends activities from areas as diverse as participatory action research, meeting facilitation and creative problem solving techniques. The approach relies on highly motivated participants with shared or converging interests. Even though we believe that thorough design, stressing initial clarification, saves resources, we expect most of the procedures to be time-consuming. The discoursive construction of future options has a playful character but it also requires courage because of its political dimension. Therefore, the approach requires trust and protection - the participants need a situation in which they are free to describe problems and visions. The participants of the design group set themselves rules of discourse and standards for protection (for example: rules about making participants' statements public). In the context of an organization, a protected status must be provided for the group. Participants first agree upon a field for design reflecting every participant's specific needs for clarification and change. Within this field, participants identify problem
P. Mambrey and B. Torpel / Co-Constructing Collaborative Computer-Eased Artifacts
281
situations/constellations. As an ongoing activity, the participants acknowledge current lines of discussion and practice related to their field of design. The concepts of this outside state of the art are compared with individual and shared concepts and results. Starting with "problem statements", the depiction of phenomena in need of improvement from participants' individual perspectives, the participants discoursively reconstruct existing reallife infrastructures, practices and modes of thinking. The reconstruction of the existing is supplemented by the construction of mentefacts: visions, scenarios and fantasies of potential infrastructures, practices and modes of thinking are generated and discussed. Visualization, generating metaphors and writing can support this constructive process. A creative atmosphere provides the possibility to generate a number of possible options. Participants generate as many meaningful mentefacts and artifacts as possible. From these, they can later chose the most promising. This collection is continuously extended. The design group evaluates the created future options according to self-set criteria. The most promising mentefacts guide the construction of new real-life artifacts, structures and practices. Practice reveals the viability of the design results. The design community evaluates its own creations according to the superordinate research/design/practice criterion of the quality of the participants' lives. Depending on the perceived level of clarification and aptness in the face of the former (and new) problem constellations/situations, the participants either quit the design process or engage in further design activities. We think that many strategies might support the collaborative/collective generation of discoursively manipulable mentefacts. Four specific procedures will serve as examples: • visualization: taking and analyzing photographs (section 3.2), • using metaphors: collecting, extending and creating metaphors (section 3.3), • contrasting existing documents, (re-) constructing their meaning and underlying assumptions and consequences (section 3.4) and • using stories: remembering characteristic situations, writing, formalizing/modelling, analyzing and transforming episodes (section 3.5). These four specific design procedures will be outlined in the following. 3.2 Visualization According to Jordan (1993) video-based interaction analysis supplements conventional ethnography. Like ethnography in general, it helps reveal patterns of people's interactions (Suchman & Trigg 1991). In the field of CSCW video data have often been used as a basis for analyzing visual aspects of interaction such as eye gaze, gestures, facial expressions (e.g. Sellen, 1992) or for giving information on the remote collaborators (e.g. Ely, Harrison & Irwin, 1993). These attempts were criticized by Nardi, Kuchinsky, Whittaker, Leichner & Schwarz (1996) as evaluations of "talking heads". They conducted a study which shows that video can be a useful tool to enable analysis, problem solving and coordinating team work. While video and spoken or written text are common as data for social science methods, photographs are not normally used as data. Patton (2000) provided an exception when he used photographs in a study on mobility within cities for informing urban planning. Photographs of artifacts are suited to identify "boundary objects" of communities of practice. In this design procedure, photographs provide impulses for design. The participants first take photographs of relevant artifacts in their work environments. The photographs usually display work means instead of individuals at work. Through the photographs participants form hypotheses and (re-) construct principles of their work. Each individual possibly generates different principles. The principles are collected and serve as a basis for generating design options in group discussions. Advantages and disadvantages of existing practices, artifacts and principles are considered
282
P. Mambrey and B. Torpel / Co-Constructing Collaborative Computer-Based Artifacts
and new, imaginary practices and artifacts (mentefacts) are generated, evaluated, rejected, mentally modified and/or replaced. These mentefacts can be depicted in a variety of modes such as drawings, diagrams, episodes or videos. Even prototypes might be created. The procedure potentially involves all interested individuals. The co-construction of mentefacts and artifacts is future-oriented. Generating options, discussing their qualities and making a consensual choice provide the chance to adequately tackle the problem of the transitions between collecting photographical data, analyzing them and constructing new mentefacts and artifacts. 3. 3 Metaphorical Scenarios Quite often visions cannot be explained by using existing words with a predefined meaning. People are also forced to compose new terms. Metaphors are non-literal means of articulating visions. They integrate known and new aspects; they explain new meanings in terms of similarities. Their structural and conceptual forces lie in their surreal quality and poly-semantic. As part of previous research on the daily work of system designers, we found that metaphors were used on different hierarchy levels of organizations. On the operational level, small groups often used metaphors as a set of cognitive tools for creating new ideas and products. Several metaphors were in use, creating a metaphorical scenario of loosely interrelated elements. In order to explain a metaphor to others, new metaphors were created. In some cases, a differentiation process took place, where existing metaphors were replaced by more "precise" and widely understood new metaphors. So different metaphors coexisted on different levels. This provided the opportunity to imaginatively create mentefacts. Metaphors therefore seem to be suited to enhance and structure joint brainstorming or self-reflection in a community of practice. They can foster discourses on technologies under as well as before construction (Mambrey, Paetau & Tepper, 1995). In this procedure, participants step metaphors that are commonly used and relevant for their communities. Metaphors composed of two words form the starting point for a structured brainstorming procedure. The participants are divided into two groups each of which elicits words related to one of the two words. The group rejoins and forms all possible combinations of the elicited words. This results in new compound metaphors. Interesting combinations are written down and then their perceived meanings are explored. The procedure is suited to relieve the discussion from limiting categories while keeping it focused on the aspects of work relevant for design. The degree of participation-orientation depends on the facilitation style. Creating new metaphors is not in itself future-oriented but can promote future-orientation and help to develop shared ontologies. The procedure itself does not suggest how to handle the gaps between elicitation, analysis and construction. 3.4 Contrastation By using contrastation as a method, we change the focus of analysis. Instead of exploring or constructing meanings of guiding visions and metaphors which might be inherited in the mentefact itself or simply constructed by a community of practice, we analyze ongoing discourses in work settings. This is an ex post analysis which aims at working with documents as informants. The design procedures of the previous sections should be supplemented by procedures analyzing discourses. For this procedure participants first identify written documents which they perceive as fundamental for work in their community. These might be, for example, written presentations on the organization, grant proposals or documentations of work means in use.
P. Mambrey and B. Torpel / Co-Constructing Collaborative Computer-Based Artifacts
283
After reading these documents, the first step in the joint analysis of these documents is to identify aspects that the participants find strange or surprising, for example words that do not fit into the text genre like ostensibly unscientific words in a research paper. Other examples are stunning metaphors, structures, concepts and definitions and, most important, "symptomatic omissions". These significant aspects should be written down. Depending on the participants' backgrounds, perspectives and goals, these preliminary results already provide them insights into foundations of their practice suited to guide constructive design activities. More and deeper insight might be gained from contrasting documents and their peculiarities against each other. Contrastations might be performed between exemplars of different document genres (e. g. speech toward the shareholders of an organization vs. the bylaws of the organization), documents from different communities (e. g. bylaws of two organizations) or already existing texts vs. texts written by the participants (see below, section 3.5). Conscious contrastation always comprises explicating or generating the very dimensions of contrastation. This way participants get to know each others' guiding categories including their political dimension (Suchman, 1994a; 1995). These categories are now amenable to change if they prove to be dysfunctional. These categories are essential for design, yet they often do not become explicated, either because they are not conscious or because they seem to be too obvious and self-evident to be articulated. Persons not involved in the community might be invited to get involved in the process of analyzing documents; they might be able to give valuable feedback on practices and assumptions that are taken for granted and therefore initially not amenable to analysis and change. For improving a community's practice and infrastructure, analyzing and contrasting work-related documents is a strategy to gain clarification on the aspects in need of improvement and possible measures for improvement. Visions of the improved situation might be written down in textual form, but other forms (such as diagrams, drawings, videos, prototypes) are equally possible. The special contribution of this procedure lies in its potential to identify, understand and determine peculiarities of communities and their present and future practice (idiosyncrasies, rules sanctions, taboos etc.). The collection and analysis of documents first refer to the past or present. As soon as documents are contrasted and especially when categories are articulated, the procedure allows for future orientation. The generated criteria can serve to evaluate what already exists as well as to guide the construction of new mentefacts and artifacts. The handling of the transitions between data collection, analysis and construction depends on the collaboration style of the design community. The procedure certainly allows for radical participation. 3.5 Creating Possible Futures by Telling and Transforming Stories The following collaborative design approach draws on the Critical Psychology action research programme (e. g. Holzkamp, 1983, chapter 9) and its specification in the story based approach of Memory Work (see e. g. Haug, 1999). The latter is based on creating episodes and appropriating techniques of formalization and depiction to generate future scenarios in appropriate depiction modes. It is supposed to elicit and combine the creative, analytic and constructive potential of the participants to create desirable work environments. The activities of the procedure include the formation of the design group, the choice of objectives, writing and analyzing episodes, creating future scenarios, appropriating modelling techniques and depiction modes and giving promising future scenarios the appropriate written forms. This provides the basis for establishing new work practices, changing organizational structures and creating new computer applications. The new situation might be the starting point for further design processes.
284
P. Mambrey and B. Torpel / Co-Constructing Collaborative Computer-Based Artifacts
As a first step, the participants agree on an aspect of the field of design for which everybody can produce (remember, imagine) a relevant episode. Every participant writes an episode. For their specific kind of episode-based action research (see below), target groups and objectives Haug & Hauser (1992, pp. 135ff) have found it beneficial to describe situations with a defined beginning and end, to write in the third person ("when she once...") instead of in the first person and to limit it to one page. In the design group the episodes are analyzed according to criteria set by the group itself. Among the criteria that appeared to be helpful and revealing for their purposes, Haug & Hauser (1992, ibid.) mention 1. plot of the episode, 2. self construction of the author of the episode within the episode, 3. blank spaces, contradictions and transitions, 4. emotions and 5. peculiar use of language. Minutes are taken, possibly with the preliminary results visible to any participant at any time during the design meetings. Writing episodes helps explicating important dimensions of a field of design as subjectively experienced by the participants. The subjective written accounts make the field of design amenable to analysis. The individual episodes are discussed. The episodes are then compared, among each other or between episodes and other written accounts. In this sense, contrastation (see section 3.4) might become part of this design procedure. Any of the above-mentioned (and other) design procedures - such as the work with visual artifacts, metaphors and existing written material - may certainly become integrated within the work with episodes. An essential step for design purposes in this procedure though is the creation of new text (stories, episodes, descriptions, scenarios) building on the original episodes. The new accounts contain possible future work practices, organizational characteristics and/or technological infrastructures. They build on the results of the joint analysis and potentially replace practices, structures and artifacts that have turned out to be dysfunctional. A second focal activity in this design procedure is the work with a variety of modes of depicting content. Participants get acquainted with a variety of genres (e. g. short story, poem, report) and modelling techniques (e. g. design scenarios, flowcharts, UML) that they consider suited for expressing scenarios and that in software technology are considered means for creating a functional specification. It has been unusual for most "mere" users to learn to express themselves by means of IT related formalization techniques. Most of these techniques are easy to comprehend and use. Every depiction mode, genre or modelling scheme stresses certain aspects of the modelled phenomena (real-world phenomena or products of the imagination) and abstracts from (neglects) others. In this sense modelling/formalization procedures are partial in the double meaning of the word: they only consider particular aspects of phenomena and, by assuming that certain aspects are more relevant than others, they decide about inclusion or exclusion of perspectives/interests which potentially has political implications (Suchman, 1994a; 1995). The most important activity in this design procedure integrates the aforementioned activities of composing and depicting. The participants represent the content of the episodes in terms of different depicting modes, genres and/or formalizations. Highlighting and possibly further developing aspects at the expense of others that might be highlighted by other procedures is a valuable step in a design procedure. None of the results needs to be final, new contents, forms and scenarios can be created or used at any time. It has to be discussed whether and how to combine/integrate the future scenarios. The future scenarios considered to be the most desirable by the group members are expressed by means of the most appropriate formalization techniques. Summarizingly. we expect the presented design procedure to support community members to find out about and articulate their discontent, requirements and wishes in order to then create mentefacts and artifacts. Living in our kind of society, we often suffer from the (societal) conditions we live in and yet we often subconsciously contribute to their further objectification and hence continuation. Analyzing each others' episodes provides good chances to not only reveal unquestioned subliminal detrimental assumptions but also to replace them by more suitable assumptions. The episode-based design method in our
P. Mambrey and B. Torpel / Co-Constructing Collaborative Computer-Based Artifacts
285
view provides the opportunity to consciously reflect the perspectives and interests instead of subconsciously just objectifying them in the evolving artifacts. Collectively analyzing and constructing scenarios provides the chance that future artifacts, practices and structures are intersubjectively validated. The participatory negotiative character of the process means that all participants can determine the shape of the future real-world phenomena and that only agreed-upon scenarios become implemented. 4 Future Perspectives As a contribution for discourse-oriented co-construction of socio-technical infrastructures in fragmented work environments, we introduced an approach exemplified by four procedures. To which extent these procedures fulfil the criteria of participation-orientation, future-orientation and successfully handling the unavoidable gaps between the activities of recording, analyzing and constructing, remains an empirical question. It also remains to be explored in which areas of practice they are most helpful. The procedures contain recontextualized elements of areas as diverse as meeting facilitation, creative problem solving and action research. Experiences from these and other fields provide a potential for the design of computer applications that remains to be further utilized. In joint efforts, participants of design processes and researchers will be able to generate the appropriate procedures for their environments and objectives, thus prospectively meeting the differentiated needs of a whole range of communities. In the practice and research community of Participatory Design a repertoire of methods has been built and extended (cf. e. g. the proceedings of the biannual conferences on Participatory Design). These methods largely apply to traditional large organizations. We would like to continue this effort in a changing economy. Acknowledgements We are greatly indebted to Steffen Budweg, Hans-Jorg Burtschick, Wolfgang Grather, Tom Gross, Uta Pankoke, Patricia Pawlyk, Meik Poschen and Alexander VoB for their valuable comments on earlier versions of this paper. References [1] Bly, S., S. Harrison & S. Irwin (1993), Media Spaces: Bringing People Together in a Video, Audio, and Compting Environment. Communications of The ACM, 36, pp. 28-45. [2] Baukrowitz, A., A. Boes & R. Schmiede (2000), Arbeitsbeziehungen in der IT-Industrie. In: FIFF Kommunikation – Wozu die Arbeit, 4/2000, 16-20. [3] Cherkasky, T., J. Greenbaum, P. Mambrey & J. K. Pors, (2000) (Eds.), Proceedings of the Participatory Design Conference, Nov. 28 - Dec. 1, 2000, in New York, NY, USA. Palo Alto, CA: CPSR Press. [4] Greenbaum, J. (1993), PD: a personal statement. In: S. Kuhn, M. Muller (Eds.) Special Issue on Participatory Design. Communications of the ACM, 36, 4, p. 47. [5] Haug, F. (1999), Vorlesungen zur Einfuhrung in die Erinnerungsarbeit. Berlin: Argument. [6] Haug, F. & Hauser, K. (1992), Marxistische Theorien und feministischer Standpunkt. In: G.-A. Knapp & A. Wetterer (Eds.), TraditionenBruche - Entwicklungen feministischer Theorie. Freiburg: Kore, 115–149. [7] Holzkamp, K. (1983), Grundlegung der Psychologie. Frankfurt a. M.: Campus. Jordan, B. (1993), Ethnographic Workplace Studies and CSCW. Working Paper for the 12th Interdisciplinary Workshop on Informatics and Psychology "Design of CSCW and Groupware Systems" Scharding, Austria, June 1993. [8] Lave, J. & E. Wenger (1991) Situated learning - Legitimate peripheral participation. Cambridge University Press New York N.Y. [9] Mambrey, P., M. Paetau & A. Tepper (1995), Technikentwicklung durch Leitbilder. Neue Steuerungsund Bewertungsinstrumente.
286
P. Mambrev and B. Torpel / Co-Constructing Collaborative Computer-Based Artifacts
[10] Mambrey. P. & M. Robinson (1997), Understanding the Role of Documents in a Hierarchical Flow of Work. In: Hayne, Stephen C. / Wolfgang Prinz (eds.): The Integration Challenge - Group "97. Proceedings of the International ACM SIGGROUP Conference on Supporting Group Work Nov. 16-19. Phoenix. Arizona. ACM New York, NY 1997, pp. 119–127. [ 1 1 ] Mambrey. P., G. Mark & U. Pankoke-Babatz (1998), User Advocacy in Participatory Design: Designers Experiences with a New Communication Channel. In: Computer Supported Cooperative Work. The Journal of Collaborative Computing, 7, 3-4, 291-313. [12] Nardi. B. A.. A. Kuchinsky, S. Whittaker, R. Leichner & H. Schwarz (1996), Video-as-Data: Technical and Social Aspects of a Collaborative Multimedia Application. In: Computer Supported Cooperative Work (CSCW). 4. 73–100. [13] Patton. J. W. (2000), Picturing Commutes: Informant Photography and Urban Design. In: Cherkasky, T.. J. Greenbaum, P. Mambrey & J. K. Pors (Eds.): PDC 2000 - Proceedings of the Participatory Design Conference, New York, NY, USA, 28 November - 1 December 2000, pp. 318–320. [14] Robinson, M. (1998), The One that Got Away: Comments on Users and Computers. In: Scandinavian Journal of Information Systems, 10, 1&2, 61–66. [15] Ruhleder, K. (1995). Involving the ,,Users": Blurred Roles and Co-Design Over Time. SIGOIS Bulletin. 16(2), pp. 1415. [16] Schein, E. H. (1998). Process Consultation: It's Role in Organizational Development. Addison-Wesley. Reading MA. [ 17] Sellen, A. (1992). Speech Patterns in Video-Mediated Conversations. In: Proceedings CHI'92. Monterey. 3-7 May 1992, pp. 49-59. [18] Snow, C. P. (1959/1993). The Two Cultures. Reprint 1993 by Cambridge University Press. [19] Sohlenkamp, M., P. Mambrey, W. Prinz, L. Fuchs, A. Syri. U. Pankoke-Babatz, K. Klockner. S. Kolvenbach (2000). Supporting the Distributed German Government with POLITeam. Multimedia Tools and Applications. 12, 1, S. 39-58. [20] Star, S. L. (1989) The Structure of Ill-Structured Solutions: Boundary Objects and Heterogeneous Distributed Problem Solving. In: Les Gasser, Michael N. Huhus (eds.): Distributed Artificial Intelligence. Volume II. Pitman: London, pp. 37-54. [21] Suchrnan. L. (1994a). Do categories have politics? In: Computer Supported Cooperative Work. 2. 177– 190. [22] Suchman, L. (1994b). Working relations of technology production and use. Computer Supported Cooperative Work. 2, 21–39. [23] Suchman. L. (1995). Speech acts and voices: Response to Winograd et al. In: Computer Supported Cooperative Work. 3. 85-95. [24] Suchman, L. & R. H. Trigg (1991), Understanding Practice: Video as a Medium for Reflection and Design. In: J. Greenbaum & M. Kyng (eds.): Design at Work: Cooperative Design of Computer Systems. Lawrence Erlbaum Assoc. Hillsdale. N.J., pp. 65-90. [25] Torpel, B. (2000). Self-employed Labor meets Codetermination - Participatory Design in Network Organizations. In: T. Cherkasky. J. Greenbaum, P. Mambrey & J. K. Pors (Eds.): Proceedings of the Participatory Design Conference. Nov. 28 - Dec. 1. 2000. in New York. NY. USA. Palo Alto. CA: CPSR Press. 184–191. [26] Torpel, B., V. Wulf, & H. Kahler, (2002), Participatory Organizational and Technological Innovation in Fragmented Work Environments. In: R. Klischewski. C. Floyd & Y. Dittrich (Eds i: Social Thinking Software Practice. Cambridge: MIT Press. (In Press)
Panel Abstracts
This page intentionally left blank
Cooperative Systems Design M. Blay-Fornarino et al. (Eds.) IOS Press, 2002
289
What's New in Knowledge Management? Panel moderator: Liam J. Bannon University of Limerick, Ireland Panelists include: Giorgio De Michelis (University of Milano Bicocca, Italy), Sebastiano Bagnara (Milano Politecnico & IRSO, Italy), Carla Simone (University of Milano Bicocca, Italy) Introduction Knowledge management (KM) is one of the hottest terms in the business community today, being equally prominent in the academic management literature, business journalism, and within the corporate environment. Not surprisingly, whenever a new business concept generates such excitement, there are both prophets and cynics around to sanctify or demonize the whole phenomenon, just as we had with the business process reengineering (BPR) movement some years ago. The purpose of this particular panel is to critically engage with the KM concept, and examine, from somewhat different disciplinary perspectives and from practical experiences, what promise it holds for a novel understanding of how organizations function. Given our specific interest in cooperation at this Conference, the relevance of research studies on patterns of interaction and communication between people, and on the importance of different technical infrastructures to support coordination and cooperation within organizations will be highlighted. For some, KM is not new, but rather a re-labelling of older issues concerning the value of expertise and information-sharing within organizations - an acknowledgment that the human capital of a company represents a key asset. For others, KM is seen as reaction to the excesses of business process re-engineering, with emphasis put on the value of human experience and organizational memory, as a riposte to the "don't automate, obliterate" rhetoric of BPR. What is especially interesting is to see how the pendulum seems to swing over time between approaches that focus on technology as the key source of innovation and growth in organizations to approaches where the human or social factor is seen as the crucial ingredient for success. With the scaling down of claims for AI and 'expert' systems, are we seeing a switch to the human factor again? Or are such binary oppositions (human-machine) too simplistic, and do we need more nuanced and articulated frameworks for interpreting organizational life? The panel will debate a number of issues concerning the KM approach, including the following. Does KM require a clear definition of terms such as data, information & knowledge? Has KM any coherent conceptual basis, or is it a chimera? What has conceptual work within such fields as CSCW and Information Management to contribute to KM?
290
L.
J.
Bannon / What's New in Knowledge Management?
What role does technology play in building effective KM within organizations? How should workers react to KM systems that attempt to make their knowledge explicit? How should employees respond to organizational attempts to map social and expertise networks within, and across organizations? Position Statement: Giorgio De Michelis Knowledge has always been important within organisations and, therefore, there has always been 'knowledge management'. What's new now in Knowledge Management? Is it yet another buzzword for managerial rhetoric? Is it only a quantitative problem (more information to be stored and retrieved; more people accessing it, etc.)?Or, does it indicate the emergence of something really new that deeply impacts the behaviour of many organisations? Let me develop some ideas along the line proposed by the third option above. In the past, putting knowledge to work required the combination of two decoupled processes (using knowledge at work and collecting knowledge in some form of storage, respectively). Specialists used knowledge at work. They sometimes detached themselves from work, in order to update and renovate their knowledge. Archivists collected and maintained knowledge, making it available to those who needed it. Innovation in work practice was slow and costly, because it required the detachment of specialists from the on-going work process. Now, in many cases, people need to access new knowledge while working, on-line, without detaching from work, even without internalising it. Others' knowledge is as relevant as their own knowledge for accomplishing their task. This new situation requires much more than efficient tools for storingand searching large amounts of information: rather, it needs the creation of systems capable, on the one hand, of bringing forth to their users the knowledge they need in a particular moment depending on their 'situation', on the other, to capture as much knowledge as possible in their practice. Only such systems merit, I think, to be called Knowledge Management systems. Position Statement: Sebastiano Bagnara Knowledge management consists of a strategy that, through an integrated system of actions and tools, aims at making always available and easily accessible and usable to the organisation the knowledge possessed by the people that directly and indirectly (e.g., consultants, and, more crucial in the present days, the customers) belong to it. In essence, KM is needed because the organisational knowledge hardly (if ever) coincides with the sum of the knowledge (implicit or explicit) possessed by the people that are in relations with the organisation: the whole knowledge possessed by the people is diverse, and much larger. Sometimes, individuals possess useful knowledge that the organisation is not aware of; sometimes, the opposite is true. In their daily actions and decision making, individuals or groups of people in an organisation may not have access to knowledge already available in the organisation or may create knowledge that will remain in their own heads. All these imbalances in knowledge creation, access, distribution and use have to be managed. KM handles these imbalances by categorising the needed and useful knowledge in the specific context of a given organisation, making
L.J. Bannon / What's New in Knowledge Management?
291
it easy to be accessed, produced, accumulated and shared. It implies also an active action to discard and make people forget previously accumulated knowledge that is of no more use and may turn out to be noxious. Knowledge management is also about deliberate forgetting. By this reasoning, KM is always about knowledge production, selection and organisation. As such it is a strategy that combines organisational actions and technological tools. However, it also necessarily implies a sometimes explicit, or more often, an implicit psychological agreement among the organisation and the individuals: the will to make themselves available to each other and share their own knowledge. This willingness cannot be established case by case: it has to be negotiated, accepted and practised for their mutual benefit and convenience. In this sense, KM is social and psychological in nature. The social component has been very often overlooked in most KM strategies. In conclusion, KM is a change strategy that involves at least three main dimensions: 1. It singles out the knowledge available in the organisation and in individuals. It takes decisions about their usefulness, and in doing so, it may be forced to discard knowledge. It is an organisational cognitive design strategy. 2. It designs how to dynamically accumulate, distribute, update, and access knowledge as a function of changing people, context and objectives. It is an organisational and technological design strategy. 3. It monitors, negotiates and takes care of the psychological contract by which people and organisations in their actual behaviours exhibit their mutual willingness to collaborate. It pays attention to how smooth this knowledge sharing process is, but it pays even more attention to the conflicts that may arise, mainly when individual knowledge becomes obsolete or even dangerous. Thus, KM is also a psycho-social design strategy.
-92
Cooperative Systems Design M. Blay-Fnrnarino el al. (Eds.) IOS Press. 2002
Understanding the roles of artifacts in cooperative work Panel moderator: Kjeld Schmidt, IT-University of Denmark
Over the years it has become evident to many researchers in the research field of ComputerSupported Cooperative Work that material artifacts play a crucial role in the ability of cooperating actors to align and integrate their distributed activities. CSCW researchers are increasingly beginning to address the crucial coordinative role of material artifacts, as evidenced by a number of workplace studies specifically investigating the central role of material artifacts in collaborative practices. At the same time, the notion of the functional equivalence of virtual and material artifacts is being questioned in the area of human-computer interaction as well, as researchers focus on the affordances of paper as opposed to digital documents. These findings are strongly corroborated by studies of the problems encountered in the uses of experimental CSCW technologies such as 'media spaces', collaborative virtual environments, and collaborative digital spaces. In the same vein, researchers are exploring novel ways of combining virtual and material artifacts under headlines such as 'tangible media', 'ubiquitous computing', 'pervasive computing', etc.. In short, having been treated as a side issue in HCI and CSCW for many years, the problem of the role of material artifacts in human action and interaction is coming to the fore. What emerges is that orderly action and interaction among actors engaged in a cooperative effort is grounded in the material nature of action and interaction, not only in the narrow sense of bodily conduct but in the broader and more fundamental sense of embodied action in a world of material artifacts and processes in material environments. Actions are materially situated and only make sense to others as materially situated. However, the coordinative role of material artifacts and environments have not been investigated systematically, nor have the implications been worked out. The panel will focus on the role of artifacts in the coordination of cooperative work: How can we conceive of artifacts? What are their roles in the coordination of cooperative work? Due to which characteristics do they play those roles? Conceptualizations such as artifacts as 'external representations' and the 'affordances' of artifacts will be explored. Of central concern are key issues such as material versus digital artifacts, the relationship between the material characteristics of artifacts and sign systems, etc.
293
Author Index Aanestad, M. Bannon, L.J. Cahier, J.-P. Cathelain, S. Cerratto, T. Crevits, I. Debernard, S. Fuks, H. Gonzalez, V. Habermann, F. Halloran, J. Hanseth, O. Harper, R. Herrmann, T. Hoffmann, M. Hornecker, E. Kunau, G. Lambie, T. Loiselet, A. Long, J.
38 289 226 173 139 173 173 88 23 209 123 38 3 53 53 71 53 107 157 107
Loser, K.-U. Mambrey, P. Mark, G. Nunnari, F. Pacaux-Lemoine, M.-P. Persson, P.-A. Poulain, T. Raposo, A.B. Rodriguez, H. Rogers, Y. Sarini, M. Scaife, M. Scheer, A.-W. Schmidt, K. Seel, C. Simone, C. Thomas, O. Torpel, B. Wagner, I. Zacklad, M.
53 275 23 7 157 239 173 88 139 123 23,191 123 209 257,292 209 7,23,191 209 275 257 226