No title

Synthese (2011) 181:413–431 DOI 10.1007/s11229-010-9716-4

Truthmakers: a tale of two explanatory projects Peter Schulte

Received: 19 January 2009 / Accepted: 2 February 2010 / Published online: 26 February 2010 © Springer Science+Business Media B.V. 2010

Abstract Truthmakers are supposed to explain the truth of propositions, but it is unclear what kind of explanation truthmakers can provide. In this paper, I argue that ‘truthmaker explanations’ conflate two different explanatory projects. The first project is essentially concerned with truth, while the second project is concerned with reductive explanation. It is the latter project, I maintain, which is really central to truthmaking theory. On this basis, a general account of truthmaking can be formulated, which, when combined with a specific theory of reduction (the ‘conceptual entailment approach’), yields a new analysis of truthmaking. This analysis is intuitively appealing and avoids the problem of necessary truths, which poses a serious obstacle for standard accounts. Keywords

Truthmakers · Explanation · Ontology · Truth · Reduction

Truthmakers play a major role in contemporary metaphysics. On the one hand, traditional metaphysical debates are now often framed in terms of truthmakers: questions like “what are the truthmakers for modal propositions?”, “what makes propositions about the past and the future true?”, or “which truthmakers can be found for propositions about laws of nature?” are ubiquitous in the current philosophical literature. On the other hand, the foundations of truthmaker theory are at the center of an increasingly intense debate,1 and the notion of truthmaking itself is subjected to closer scrutiny. This paper is a contribution to the latter discussion.

1 Several monographs and anthologies have been published recently that are devoted to the topic of truthmaking, cf. Armstrong (2004), Beebee and Dodd (2005), Merricks (2007), Monnoyer (2007), Lowe and Rami (2008).

P. Schulte (B) University of Erlangen-Nürnberg, Erlangen, Germany e-mail: [email protected]

123

414

Synthese (2011) 181:413–431

Most theorists think of making true as an explanatory relation: truthmakers are supposed to explain the truth of propositions. But it is unclear what kind of explanation truthmakers provide—and critics have questioned whether they succeed in providing any explanation at all. In this paper, I attempt to answer these questions (or, at least, indicate how they might be answered). After laying out truthmaker theory and its explanatory claims in Sects. 1 and 2, I argue in Sect. 3 that standard truthmaker theory conflates two explanatory projects: the project of explaining truths by facts, and the project of explaining higher-level facts by lower-level facts. This dual character of truthmaker explanations has led to considerable confusion about their nature, which can be avoided by carefully distinguishing between both projects. I then go on to argue that the second project (which is central to truthmaking theory) can be understood in terms of reductive explanation, and proceed to formulate a new general account of truthmaking based on this insight. In Sect. 4, I combine the general account with a specific theory of reductive explanation, thereby developing a new analysis2 of truthmaking (a variant of the general account). This illustrates how, according to my picture, analyses of truthmaking are to be constructed. But the resulting truthmaker definition is also attractive in its own right, since—as I argue in the last section—it enables us to solve a serious problem that has plagued earlier accounts.

1 From the truthmaker intuition to truthmaker theory Truthmaker theorists usually motivate their position by appealing to a powerful intuition: the truthmaker intuition, as I will call it. Rodriguez-Pereyra formulates it in a clear and compelling way: [T]he insight behind the idea of truthmakers is that truth is grounded. In other words, truth is not primitive. If a certain proposition is true, then it owes its truth to something else: its truth is not a primitive, brute, ultimate fact. The truth of a proposition thus depends on what reality, and in particular its subject matter, is like.3 This idea—simply put, the idea that “truth depends on being”—is the foundation of truthmaker theory. It is spelled out differently, however, by different theorists, resulting in a variety of truthmaker principles. The complex debate that has arisen from this diversity focuses on three key issues. First, there is a dispute about truth-bearers: Which entities should be considered the ‘bearers’ of truth (and falsity), the things made true by truthmakers? Most theorists opt for propositions (like e.g. Rodriguez-Pereyra in the passage just quoted), others favor sentences or utterances. Since this question does not bear on the problems discussed

2 When I speak of “the analysis of the truthmaking relation”, I do not mean a conceptual analysis of “making true”; instead, I use the term “analysis” in the sense of “theoretical account”. 3 Rodriguez-Pereyra (2005, p. 21).

123

Synthese (2011) 181:413–431

415

here, I will simply side with the majority.4 (I will also adopt the common convention to abbreviate “the proposition that p” with “p”.) A second question concerns the scope of the truthmaker principle: which propositions require truthmakers for their truth, i.e. which propositions are true if and only if they have a truthmaker? Truthmaker maximalists think that every true proposition must have a truthmaker and thus assign universal scope to the truthmaker principle,5 but other theorists disagree: They restrict the principle to contingent, synthetic or positive propositions. Again, this issue is not directly relevant to my main argument, so I will set it aside for the most part of the paper. (In Sect. 5, though, I will briefly come back to it and argue for a restriction of the truthmaker principle to synthetic truths.) The third, most fundamental controversy that I want to consider here concerns the analysis of the truthmaking relation. It is clear that making true is a cross-categorical relation,6 holding between truthmaking entities and true propositions, but its nature is controversial. The dominant approach, “truthmaker necessitarianism”,7 treats making true as a relation that involves necessitation. Truthmaker necessitarians hold that an entity x can only be a truthmaker for p if the existence of x necessitates (i.e. ensures, guarantees) the truth of p. They subscribe, therefore, to the following principle (with different restrictions on scope): (TMN) p is true iff there is an x such that the existence of x necessitates the truth of p. Or, equivalently: (TMN*) p is true iff there is an x such that necessarily, if x exists, p is true. Among the proponents of truthmaker necessitarianism are the founders of contemporary truthmaking theory—D.M. Armstrong, C.B. Martin, Kevin Mulligan, Peter Simons and Barry Smith—together with several other philosophers.8 There are two points to note here. First, while all truthmaker necessitarians, by definition, think that necessitation is a necessary condition for truthmaking, they are not committed to the view that it is also a sufficient condition. They may (and many of them do) deny that every x whose existence necessitates the truth of p is a truthmaker for p.9 This position is perfectly compatible with (TMN). However, it is obvious that truthmaker theorists who deny that necessitation is sufficient for truthmaking cannot identify these two relations. To give a full analysis of “making true”, they have to specify additional conditions that an entity has to satisfy in order to count as a truthmaker 4 There is also a second, more substantial reason for preferring propositions to the other candidates. As

Trenton Merricks has convincingly argued, we would have to reformulate the usual truthmaking definitions and principles if we chose utterances or other contingent entities as truthbearers, cf. Merricks (2007, p. 6ff). 5 The most prominent proponent of maximalism is Armstrong, cf. his (2002, p. 32; 2004, p. 7). 6 Cf. Armstrong (2004, p. 5). 7 I adopt Armstrong’s terminology here, cf. his (2004, p. 5). Parsons uses the term “essentialism” instead of “necessitarianism”, but his characterisation of this position is identical, cf. his (1999, p. 328). 8 E.g., Fox (1987), Molnar (2000) and Rodriguez-Pereyra (2005). 9 Cf. Smith (1999), Simons (2000), and Rodriguez-Pereyra (2005, p. 18).

123

416

Synthese (2011) 181:413–431

for a some proposition p, a task which is by no means easy. (I will come back to this issue later in the paper.) Secondly, truthmaker necessitarians disagree about which kinds of entities are best suited to play the truthmaking role. Some, like Armstrong, favor states of affairs or facts, others, like Mulligan, Simons and Smith, prefer tropes or moments (i.e. particularized qualities).10 In the following, I will often ignore the second option and treat facts as the paradigmatic truthmakers, but my arguments can easily be rephrased in terms of tropes. (To avoid cumbersome formulations, I will often use “[p]” as an abbreviation for “the fact that p”.) As I have noted above, truthmaker necessitarianism is the dominant view in the field. But it is not undisputed. Lewis (1998/1999, 2001a), Parsons (1999) and others have argued that an entity does not have to necessitate the truth of a proposition in order to make it true, and they have formulated alternative, non-necessitarian truthmaker principles. I think that my arguments can be extended to these theories of truthmaking, but I won’t argue for this claim here. Instead, I will be focusing exclusively on truthmaker necessitarianism. 2 Truthmakers and explanation According to the truthmaker intuition presented at the beginning of the last section, a proposition p is made true by a certain entity x iff its truth depends on the existence of x, or—to use another popular formulation—iff p is true because x exists. These explications make it clear that truthmaking is supposed to be an explanatory notion. But, as I have indicated in the introduction, the explanatory claims of truthmaker theory are controversial. What kind of explanation do truthmakers provide? Everyone agrees that the explanation cannot be a causal one—facts (or other entities) do not cause propositions to be true. But I think that this negative answer cannot suffice. A positive account of the nature of truthmaker explanations is needed, and this is what truthmaker theorists have so far failed to provide. Of course, some theorists reject the suggestion that such an account is required; they propose to treat the “because” of truthmaking as an unanalyzed primitive.11 This, however, seems unsatisfactory to me, especially in the face of current skepticism about truthmaker explanations.12 How can we do better? As a first step, I think we have to get clear about the explananda of truthmaker explanations. On the face of it, this could not be easier: The standard formulations of the truthmaker intuition clearly suggest that truthmakers have to explain why certain propositions are true, i.e. that the explanandum of a truthmaker explanation is the truth of a certain proposition. But appearances are deceptive here. In fact, many truthmaker explanations have explananda with two different components, and this dual character is the key to a better understanding of their nature—or so I will argue. 10 Cf. Armstrong (1997), Mulligan et al (1984). 11 Cf. McFetridge (1977/1990), Liggins (2005). In a similar vein, Rodriguez-Pereyra (2005) uses “in virtue

of” as a primitive notion. 12 Daly (2005), e.g., presents a forceful attack on the explanatory claims of truthmaker theorists.

123

Synthese (2011) 181:413–431

417

My argument starts with the observation that there are two kinds of truthmaker explanations. First, consider the following examples: (1) Aristotle exists is true because Aristotle exists. (2) The rose is red is true because the fact [The rose is red] exists. These explanations seem rather trivial, at least prima facie. The truth of x exists is explained by the existence of x, and the truth of p by the existence of the fact that p. I will call these explanations “simple truthmaker explanations”. (By this, I do not mean to prejudge whether (1) and (2) are genuine explanations, a question I take up in Sect. 3. For now, I use the term “explanation” in a wide sense that covers all attempts to explain something, whether successful or not.) But there is also a second class of truthmaker explanations that is frequently discussed in the literature. Again, let us look at two examples: (3) If Lauren turned around, she would have a sensory impression of a bookshelf is true because the fact [There is a bookshelf behind Lauren] exists (and certain background conditions hold). (4) If Humphrey were confronted with fire, he would run away is true because the fact [Humphrey is in brain state N] exists (and certain background conditions hold). In (3) and (4), the truth of a counterfactual conditional is explained by the existence of some non-modal fact. In contrast with (1) and (2), these explanations are clearly informative. I will call them “substantial truthmaker explanations”. (To be precise, the truthmaker for the counterfactual in (3) is not just the fact that there is a bookshelf behind Lauren, but the complex fact that results when [There is a bookshelf behind Lauren] is conjoined with the relevant background conditions. The same holds, of course, for [Humphrey is in brain state N] in the case of (4). For the sake of brevity, I will omit reference to background conditions in the following, but it should always be kept in mind that, strictly speaking, they are part of the complete truthmaker.) Before moving on, let me emphasize that substantial truthmaker explanations are of major importance for truthmaker theory. First, note that explanations like (3) or (4) are the basis of truthmaker arguments, which give truthmaker theory its metaphysical bite. Armstrong and Martin, for example, use (3) and (4) in their truthmaker arguments against phenomenalism and Rylean behaviorism, respectively.13 Consider, e.g., a phenomenalist like Ayer, who claims (roughly) that he can accept statements about unobserved objects because he can translate them into counterfactuals about possible sense impressions.14 Armstrong and Martin challenge Ayer to provide truthmakers for his counterfactuals; but, as it turns out, he has nothing to offer. (He could, of course, assume the existence of primitive facts about possible sense impressions, but this would be implausible—and it would clearly run against the empiricist 13 Cf. Armstrong (1989, p. 8ff), and his (2004, p. 1ff). Parsons (1999) and Lewis (2001a) defend their

non-necessitarian truthmaker principles by arguing that they can be used in similar arguments. cf. also Sider (2001, p. 40). 14 Cf. Ayer (1946/1952, p. 140f). I am simplifying Ayer’s account here: he actually talks about “sense

contents”, not about “sense impressions”.

123

418

Synthese (2011) 181:413–431

spirit of his theory.) Realists like Armstrong and Martin, by contrast, can easily provide the relevant truthmakers—as (3) demonstrates for one particular counterfactual. Rylean behaviorism is countered along the same lines, with (4) exemplifying the kind of truthmaker explanation that the behaviorist is not able to give. More generally, if we look at the application of truthmaker theory in various fields of ontology, substantial truthmaker explanations always seem to take center stage. When metaphysicians search for the truthmakers of, e.g., modal, nomological or tensed propositions, they are not looking for simple truthmaker explanations. These are easily provided by the schema “p is true because [p] exists”. Instead, what metaphysicians are looking for (and debating about) are explanations that are informative and illuminating, i.e. substantial truthmaker explanations. 3 Truthmakers, truth and reductive explanation Simple and substantial truthmaker explanations are obviously different. But what exactly does this difference amount to? In order to answer this question, I suggest that we have to examine the role of the truth-predicate in both kinds of explanations.15 In simple truthmaker explanations, the predicate “is true” plays an essential role. If we remove it from the true statements (1) and (2) by substituting “p” for “p is true”, we get the following dubious claims: (1-) Aristotle exists because Aristotle exists. (2-) The rose is red because the fact [The rose is red] exists. It is virtually undeniable that (1-) is false. Aristotle’s existence explains a lot of things, but it does not explain itself. (2-) may be more controversial, but I think it is basically on a par with (1-): “the rose is red because it is a fact that the rose is red” does not seem to be a genuine explanation. By contrast, if we eliminate the truth-predicate from substantial truthmaker explanations, the resulting propositions are still perfectly alright.16 (3), e.g., leads to the following thesis: (3-)

If Lauren turned around, she would have a sensory impression of a bookshelf because the fact [There is a bookshelf behind Lauren] exists.17

This seems to be a good explanation: What Lauren would see if she turned around is explained by a fact about her surroundings. The use of “because” is entirely justified. The same applies, mutatis mutandis, to the transformed version of (4). Summing up, we can say that substantial truthmaker explanations remain explanatory after the truth-predicate is removed, while simple truthmaker explanations do not. Why is that the case? The answer is not hard to find. As Benjamin Schnieder has pointed out, substantial truthmaker explanations (as I call them) are contracted 15 Is “true” a real predicate? I agree with Künne (2003, pp. 33–92) that it is, and therefore, I use the expression “truth-predicate” here. However, none of my arguments depends on this assumption. 16 Cf. also Lewis (2001b, p. 278f), who makes a similar observation in the context of a different discussion. 17 As indicated above, I omit the reference to background conditions here.

123

Synthese (2011) 181:413–431

419

versions of “series of explanations”.18 They combine two different explanations: the explanation of a truth by a fact, and the explanation of one fact by another fact (as I would put it). (3), e.g., is really a contracted version of these two explanations: (3a) If Lauren turned around, she would have a sensory impression of a bookshelf is true because the fact [If Lauren turned around, she would have a sensory impression of a bookshelf] exists. (3b) The fact [If Lauren turned around, she would have a sensory impression of a bookshelf] exists because the fact [There is a bookshelf behind Lauren] exists. In general, a substantial truthmaker explanation of the form “p is true because [q] exists” is a combination of an explanation of the form “p is true because [p] exists” and an additional explanation of form “[p] exists because [q] exists”.19 The first explanation is simple (i.e. has the same form as (1) and (2)), the second is substantial. Surprisingly, neither Schnieder nor any other truthmaker theorist has explored the consequences that this observation might have for the question that was raised in the introduction—the question about the nature of truthmaker explanations.20 If truthmaker explanations are heterogeneous in the way described above, it is plausible to suppose that they do not share a common nature. Therefore, I suggest that the initial question should be broken up into two questions—one about the nature of (simple) explanations like (1), (2) and (3a), and another one about the nature of (substantial) explanations like (3b). I will consider these questions in turn. Take simple explanations first. What kind of explanation do we give when we say “Aristotle exists is true because Aristotle exists”? Many are tempted to say that we do not give any real explanation at all, but merely a pseudo-explanation. Presumably, this is because simple explanations can be constructed very easily for any explanandum: When the truth is given, the simple explanation for it is immediately obvious. But the question is not whether the statement “Aristotle exists is true because Aristotle exist” is interesting or informative by everyday standards, but whether it is correct. I am inclined to think that it is. It might be trivial (in a sense) that the proposition Aristotle exists is true because Aristotle exists, but that does not make it false. Most theorists seem to share my intuition. They tend to agree that, at least prima facie, simple truthmaker explanations are genuinely explanatory.21 How can we account for these explanations? Since they are essentially concerned with truth, we cannot hope to fully illuminate their nature without clarifying the nature 18 Schnieder (2006, p. 37). 19 When I call substantial truthmaker explanations “combinations” and “contractions” of two explanations,

I simply mean that they are abbreviations of the conjunction these two explanations: The statement “p is true because [q] exists” is an abbreviation of the full explanation, “p is true because [p] exists, and [p] exists because [q] exists”. 20 Schnieder (2006, p. 42), e.g., challenges the truthmaker theorists “to tell us […] what explanatory relation could justify the truth of the explanations they need for their theory to work”. Schnieder does not consider the possibility that there might be two different explanatory relations involved in (what I call) simple and substantial truthmaker explanations; instead, he only focuses on explanations of the simple kind. 21 This holds not only for correspondence theorists, but also for minimalists, cf. Horwich (1990/1998,

p. 104ff), Künne (2003, p. 154ff), Wright (1992, p. 27).

123

420

Synthese (2011) 181:413–431

of truth itself.22 Consequently, our account of simple truthmaker explanations will ultimately depend on the theory of truth we accept. Correspondence theorists will account for them by appealing (in some way or other) to the correspondence relation, while minimalists have (at least) two options: they can either try to explain the underlying intuition away,23 or they can resort to some ‘sparse’ kind of explanation (e.g. “conceptual explanation”24 ). Alas, since a proper evaluation of these options is beyond the scope of this paper, I cannot go into the details here. Therefore, I will leave the task of clarifying the nature of simple truthmaker explanations to theorists of truth. This is not as problematic as it may sound. These explanations, I suggest, are not central to truthmaker theory anyway; much more important are substantial truthmaker explanations or, more precisely, the characteristic part of these explanations that makes no reference to truth. Some critics will wonder whether this position is even coherent: Isn’t truthmaker theory a theory of truth—more specifically, a version of the correspondence theory? Indeed, many truthmaker theorists have assumed that it is. Armstrong, e.g., writes: “Are truthmakers for truths just the ‘correspondents’ envisaged by the Correspondence theory of truth? I think the answer to this is [. . .] ‘Yes’. Truthmaker theory is a correspondence theory [. . .].”25 But I do not think that this claim is justified. First, it has been noted by several authors that the truthmaker principle (in all its versions) is perfectly compatible with a minimalist account of truth. Minimalists can adopt (TMN) and argue that the truth-predicate in these statements just serves its usual ‘minimal’ function, namely, the function of enabling generalizations.26 Since the correspondence theory is a rival to minimalism, it cannot be identified with a thesis that is compatible with the minimalist account of truth. Secondly, truthmaker theory offers no (non-circular) analysis of “being true”. The only candidate would be the analysis of “being true” as “being made true”, but since “being made true” can only be defined with reference to truth (as in definitions like, e.g., “x makes p true iff the existence of x necessitates the truth of p”), such an analysis would be viciously circular.27 So truthmaker theory is not a theory of truth. This shows that is it at least coherent to suppose that the explanations central to truthmaker theory are not those concerned with truth. But is it also plausible? First, recall the discussion at the end of Sect. 2. There, I argued that substantial truthmaker explanations are central to truthmaker arguments and to the search for truthmakers in various domains of ontology. But I think we can go one (small) step further and say that it is the substantial part of these explanations (i.e. the part not concerned with truth) that is doing all the work here, so that in the end, truth drops out of the picture. As John Bigelow observes in The Reality of Numbers: 22 Consider the following analogy: We cannot hope to clarify the relation of brain states to phenomenal

consciousness if we ignore questions about the nature of phenomenal consciousness; these issues are too tightly connected to be pulled apart. The same holds, I suggest, for the relation of facts to truths and questions about the nature of truth. 23 Cf. Horwich (1990/1998, p. 105). 24 Cf. Künne (2003, p. 154ff). 25 Armstrong (2002, p. 30). 26 Cf. Lewis (2001b). 27 Cf. Merricks (2007, p. 15).

123

Synthese (2011) 181:413–431

421

The force behind Truthmaker [i.e. the truthmaker principle] lies deeper than worries about the nature of truth […]. The term ‘truth’ makes its appearance here largely to facilitate generality of exposition. If we focus on just one particular truth, then the guts of Truthmaker can be stated without even using the term ‘truth’, or any equivalent. Consider for instance the truth that the Parthenon is on the Acropolis. Truthmaker requires that there must be some things such that, necessarily, if these things exist then the Parthenon is on the Acropolis. Thus stated, we have made no explicit mention of […] truth.28 Applied to our discussion, the main contention is this. When we consider a particular truth, we can reformulate the question “What makes p true?” as “Why p?”. Instead of asking for an explanation of a proposition’s truth, we can simply ask why a certain fact obtains. The question “What is the truthmaker for Flying pigs are possible?”, e.g., can be reformulated as “Why are flying pigs possible?”, and the same holds for modal propositions generally, as well as for nomological, mathematical and tensed propositions. Accordingly, if a substantial truthmaker explanation is proposed for one of these propositions, the proposal can be phrased as an explanation of the corresponding fact. Of course, truthmaker theorist sometimes also give simple explanations—namely, in cases where they do not think that substantial explanations are available. In such a case, the truth-predicate cannot simply be eliminated. But still, I think that the ontological thesis behind such a simple truthmaker explanation can be restated without making reference to truth: it can be characterized as the view that the facts in question are primitive. Therefore, it seems that all ontological debates about truthmakers can (in principle) be framed in such a way that talk of truth is eliminated. This suggests that the major debates in truthmaker theory are not about truth, but about (what I call) substantial explanations, and that is why clarifying the nature of these explanations should be of primary importance for truthmaker theorists. I will come back to this point once again at the end of this section. Let us now turn to the nature of substantial explanations. Consider again our example (3b): “The fact [If Lauren turned around, she would have a sensory impression of a bookshelf] exists because the fact [There is a bookshelf behind Lauren] exists.” What kind of explanation is this? It is, I suggest, an explanation of a higher-level fact by a lower-level fact. Let me clarify this thesis. The counterfactual “If Lauren turned around, she would have a sensory impression of a bookshelf” could hold for many reasons: because of a hologram projected into space behind Lauren, because of a neurological disorder of Lauren that makes her see bookshelves whenever she turns her head, or—as in our case—because there really is a bookshelf behind her. This means that the modal fact expressed by the counterfactual is multiply realizable: it could be realized by a number of different non-modal facts. The relation of realization is commonly construed as a relation between facts (or properties) of different levels, with the lower-level fact (or property) realizing the higher-level fact (or property). It seems clear that (3b) is a ‘realization explanation’, and ipso facto, it is also an explanation of a higher-level fact by a lower-level fact. 28 Bigelow (1988, p. 127).

123

422

Synthese (2011) 181:413–431

This explanatory pattern is familiar from other areas of philosophy, notably the philosophy of mind, where the explanation of multiply realizable, higher-level mental facts by lower-level physical facts is discussed under the heading of reductive explanation. Simplifying somewhat, my point can thus be put this way: every substantial truthmaker explanation is a combination of (i) a simple truthmaker explanation and (ii) a reductive explanation (roughly in the sense familiar from the philosophy of mind). This thesis is supported by the analysis of the examples I have given above, and it could be further corroborated by a survey of substantial truthmaker explanations discussed in the literature.29 So it seems that metaphysicians who are trying to answer the question “What are the truthmakers of modal (nomological, tensed, …) propositions?”, are best understood as looking for reductive explanations for modal (nomological, tensed, …) facts. At this point, it might be suggested that the truthmaker theorist’s work is done. If substantial explanations are reductive explanations, we have to turn to theories of reduction to get clear about their nature—in the same way that we have to turn to theories of truth to elucidate the nature of simple truthmaker explanations.30 There is certainly something right about this suggestion. In fact, I completely agree that truthmaker theorists should resort to theories of reduction in order to clarify what substantial truthmaker explanations are. But there is a crucial difference to the case of simple truthmaker explanations: while simple explanations are not central to truthmaker theory (as I have argued above), substantial explanations clearly are. So ‘outsorcing’ questions about substantial explanations will not do; instead, it is vital for truthmaker theorists to get a better understanding of how reduction works. This becomes clearer when we consider what is perhaps the most interesting consequence of the arguments presented above. In this paper, I started from the observation that there are two types of truthmaker explanations, simple and substantial ones, and I went on to argue that the latter should be understood (in part) as reductive explanations. As it turns out, these results can be used to formulate a new definition of “making true”31 : (DR) x makes p true iff (i) x is identical with [p] or (ii) x reductively explains [p]. This definition covers simple and substantial explanations with its first and second clause, respectively; and since all cases of truthmaking fit into one of those two categories, the definition seems to be correct (provided, of course, that my account of substantial truthmaker explanations is on the right track). We can see more clearly now why truthmaker theorists should be interested in theories of reduction. For as it stands, (DR) is still rather vague: it does not say anything about the specific conditions that something must satisfy in order to provide a reductive explanation for [p]. These conditions, however, are of central importance for truthmaker theorists, who need a definition of truthmaking that specifies exactly 29 Armstrong (2004, p. 6) e.g., argues that “a sufficiently dense conglomeration of H O molecules” is a 2

truthmaker for the proposition there exists a certain quantity of water. Obviously, this is a standard case of reductive explanation.

30 I would like to thank two anonymous referees for drawing my attention to this problem. 31 I would like to thank an anonymous referee for urging me to make this point explicit.

123

Synthese (2011) 181:413–431

423

which entities qualify as truthmakers for which truths. In order to formulate such a definition, truthmaker theorists have to combine (DR) with a specific account of reductive explanation. This, of course, can be done in a number of ways, with different accounts of reduction yielding different definitions of “making true”. One might therefore say that (DR) is not so much a truthmaker definition, but rather a blueprint for such definitions. It outlines a general strategy for constructing analyses of truthmaking, which consists in taking whatever account of reduction one deems plausible and specifying clause (ii) of (DR) accordingly. However, there is one caveat: (DR) should not be combined with accounts of reduction that are too narrow. Of course, for many truthmaker explanations, any account of reduction will give the right results. Facts about H2 O make propositions about water true, facts about grains of sand are truthmakers for propositions about dunes, and so on—in these cases, it is clear that reductive explanations are involved, and any account of reductive explanation that does not cover them is clearly defective on its own terms. But there are also other instances of truthmaking—among others, our examples (3) and (4), and cases like the following: (5) [a is red] makes a is colored true. (6) [a is red] makes a is red or a is blue true. It is hardly deniable that (5) and (6) are correct. But (DR) can accommodate these statements only if it is accepted that [a is red] provides a reductive explanation for [a is coloured] and for [a is red or a is blue]. This shows that we should combine (DR) with a liberal account of reductive explanation that includes limiting cases like these. (The best alternative, I think, is to reject (DR) and to claim that there are several different kinds of substantial truthmaker explanations: some that involve reductive explanations, and others than involve different (non-causal) kinds of explanations, e.g. ‘conceptual’ or ‘logical explanations’. However, this strategy has the disadvantage of disuniting truthmaker theory even further, possibly to the point of making “truthmaking” and “truthmaker explanation” dubious umbrella terms. A more unified account is therefore clearly preferable.) So far, I have laid out (and argued for) a general strategy for constructing truthmaker definitions. In the next section, I will implement this strategy by developing a specific definition of “making true”, based on a particular approach to reductive explanation. This serves, first, to illustrate the strategy sketched above, but it also shows that there is at least one account of reductive explanation that is wide enough to cover all relevant cases (like, e.g. the statements (5) and (6) above). Finally, I think that the emerging account of truthmaking is interesting in its own right, as the subsequent discussion will (hopefully) make clear. 4 A new analysis of truthmaking Reductive explanation is a controversial and difficult topic, to say the least. Nevertheless, a promising account of it has emerged in recent decades, developed and defended by Horgan (1984), Lewis (1994/1999), Jackson (1994/1998) and Chalmers (1996), among others. This account ties reductive explanation to conceptual entailment. Roughly speaking, its core thesis can be summarized thus: The fact that q is

123

424

Synthese (2011) 181:413–431

reductively explained by the fact that p only if q is conceptually entailed by p, i.e. only if the conditional If p, then q is analytic.32 According to this approach (which I will call the “conceptual entailment approach” or CE-approach, for short), the reductive explanation of higher-level facts by lower-level facts requires that we show how truths about higher-level facts can be logically deduced from truths about lower-level facts and conceptual truths. Consider, for example, what it takes to reductively explain the fact that H2 O freezes at 0◦ C. First, we can deduce from the molecular structure of H2 O and the relevant laws of nature that H2 O-molecules form stable hydrogen bonds at 0◦ C, thereby generating a crystal structure, i.e. a structure where the place of each H2 O-molecule is fixed relative to all the other molecules. But if a (sufficiently dense) object is such that the position of each of its parts is fixed relative to its other parts (i.e. if the arrangement of its parts can only be changed by considerable force), it follows—by the meaning of “solidity”—that this object is solid. Therefore, we can conclude that samples of H2 O become solid at 0◦ C, and since “becoming solid because of a drop in temperature” is synonymous with “freezing”, we are justified to draw the further conclusion that H2 O freezes at 0◦ C. Proponents of the CE-approach hold that deductions of the same form are necessary if we want to give reductive explanations of biological, psychological or sociological facts in terms of physical facts. To put it more vividly, the central idea is this: everything that is reductively explainable in physical terms is such that it could be ‘read off’ of a purely physical description of the world by a Laplacean Demon (i.e. a being with unlimited cognitive capacities that possesses all of our human concepts).33 By saying that, I have not specified sufficient conditions for reductive explanation, and I will not attempt to do so here. For our purposes, the intuitive idea that reductive explanations show how “the large” depends on “the small and many” (as David Lewis puts it34 ) will suffice. Thus, according to CE-approach, a reductive explanation consists in an explanation of one set of facts by a more fundamental (or more finegrained) set of facts, where the complete description of the latter conceptually entails the description of the former. Many objections have been raised against the CE-approach. Some philosophers question the idea of conceptual truth in general,35 others doubt the ability of the

32 This is not entirely true. As Chalmers and Jackson have noted, it is necessary in many cases to conjoin the complete physical description P with (i) a proposition T stating that the description is complete (i.e. a ‘that’s all’ clause, cf. Chalmers and Jackson 2001, p. 317) and (ii) a piece of indexical information I (“a ‘you are here marker’ added to the objective map given by [the microphysical description]”, cf. Chalmers and Jackson 2001, p. 318). So, strictly speaking, something is reductively explainable if it is conceptually entailed by the conjunction of P, T and I. 33 Note that the examples discussed here can also be used to underscore the connection between reductive explanation and truthmaking: it is quite plausible to say that facts about the formation of hydrogen bonds between H2 O-molecules constitute the truthmaker for H2 O freezes at 0◦ C, and that physical facts make propositions about biological, psychological and sociological matters true, provided that there is a reductive explanation of these higher-level facts in terms of physical facts. 34 Lewis (1994/1999, p. 294). 35 Among these critics are Williamson (2006) and, of course, the adherents of Quine (1953/1980).

123

Synthese (2011) 181:413–431

425

CE-approach to deal with scientific a posteriori identities (like “water = H2 O”).36 I cannot discuss all these criticisms here, so I simply refer to the reader to the works cited above, especially to Jackson (1998) and Chalmers and Jackson (2001), who develop ingenious and (in my opinion) quite plausible rebuttals to most standard objection. In any case, it seems undeniable to me that the CE-approach is a powerful and wellmotivated proposal that should be taken seriously. Therefore, I will not dwell on its problems here; instead, I want to focus on the consequences that follow for the truthmaker principle and the analysis of truthmaking if we assume that the CE-approach is correct.37 When we combine (DR) with the CE-approach to reductive explanation, we get the following definition of “making true”: (DCE1)

x makes p true iff (i) x is identical with [p] or (ii) x exists conceptually entails p.

Since [p] exists obviously entails p, we can simplify the definition by dropping the redundant first clause, thus arriving at (DCE2): (DCE2)

x makes p true iff x exists conceptually entails p.

This seems to be an elegant and promising definition. However, there is a problem with the way it is formulated: “x exists” does not designate a unique proposition. If x is Aristotle, the term “x exists” is indeterminate between Aristotle exists, The teacher of Alexander exists, The most famous disciple of Plato exists and a lot of other propositions. This is more than just a formal problem, because we cannot simply allow that Aristotle makes p true iff any of these existential propositions entail p. If we allowed that, Aristotle would also make propositions like Alexander exists or Alexander has a teacher true (since they are conceptually entailed by The teacher of Alexander exists), and that would be a clearly disastrous result.38 Fortunately, it is a result that can be evaded. In order to do so, I suggest that we introduce the notion of a basic description. The basic description of an entity x is the complete description of x in terms of the ‘lowest’, most fundamental level of description.39 If physicalism is true, basic descriptions will be microphysical descriptions; if it is false and, say, property dualism happens to be true instead, basic descriptions will 36 Cf. Block and Stalnaker (1999). 37 My proposal to combine truthmaker theory with the CE-approach of reductive explanation should not be

confused with Schnieder’s (2006) attempt to analyze (elements of) truthmaking explanations as “conceptual explanations”. According to Schnieder’s (2006, p. 33) construal of conceptual explanations, p conceptually explains q only if q contains “complex or elaborated concepts” that are explained in recourse to the “more primitive concepts” involved in p. By contrast, conceptual primitiveness is not relevant for reductive explanations: facts about the hydrogen bonds between H2 O-molecules provide a reductive explanation for the fact that H2 O freezes at 0◦ C, although the concepts “hydrogen bond” and “molecule” are surely not more primitive than the concept “freezing”. 38 Merricks (2007, p. 13) makes essentially the same point against a formulation of the truthmaker principle by Bigelow. 39 What if there is no fundamental level, as Schaffer (2003) suggests? In that case, I think we would have to

relativize truthmaking to levels. But this modification raises complicated issues that I cannot pursue further here.

123

426

Synthese (2011) 181:413–431

sometimes also contain constituents that refer to irreducible non-physical qualia.40 With the notion of a basic description in hand, we can give the following definition: x makes p true iff there is a description δ such that δ is the basic description of x and (δ) exists conceptually entails p.

(DCE3)

(I write “(δ) exists” instead of “δ exists” to indicate that the relevant proposition is not the description the F exists, but the F exists.) On the basis of (DCE3), we can also formulate a new version of the truthmaker principle, which reads as follows (with different restrictions on scope): p is true iff there is an x and a description δ such that δ is the basic description of x and (δ) exists conceptually entails p.

(TMCE)

Or, equivalently: (TMCE*)

p is true iff there is an x and a description δ such that δ is the basic description of x and if (δ) exists, then p is analytically true.

It should be noted that (TMCE) entails that for every true proposition p in its scope, there has to be something whose existence necessitates that p is true, so the new principle—while clearly different from (TMN)—is still a version of truthmaker necessitarianism. However, I think that this new account of truthmaking has (at least) one decisive advantage over classical necessitarianism: it enables us to solve a problem that has plagued truthmaker theory for a long time—the problem of necessary truths. 5 Solving the problem of necessary truths The simplest necessitarian analysis of the truthmaking relation identifies it with the relation of necessitation.41 According to this analysis, the definition of “making true” reads as follows: (DN1)

x makes p true iff the existence of x necessitates the truth of p.

Or, equivalently: (DN1*)

x makes p true iff necessarily: if x exists, p is true.

As I have mentioned in Sect. 1, many truthmaker necessitarians reject (DN1). They do so for different reasons, but one of the main reasons is the fact that (DN1) generates strongly counterintuitive results when it is applied to necessary truths.42 Consider, e.g., 2 + 2 = 4. What makes this proposition true, according to (DN1)? The answer is: everything. My left foot, the Eiffel Tower, the planet Mars and innumerable other things (and facts) necessitate the truth of 2 + 2 = 4: for necessarily, 40 When I speak of “descriptions” here, I do not mean linguistic expressions or utterances; instead, I use the term to refer to constituents of propositions, i.e. to abstract objects. 41 A proponent of this analysis is Fox (1987, p. 189). Armstrong (1997, 2002, 2004) is not clear on this

issue; some passages suggest that he accepts (DN1), others seem to imply that he does not. 42 This problem was first formulated by Restall (1996).

123

Synthese (2011) 181:413–431

427

if one of these entities exists, 2 + 2 = 4 is true. But intuitively, my left foot is not a truthmaker for 2 + 2 = 4, and neither is Mars or the Eiffel Tower. The existence of these things in no way explains (‘accounts for’, ‘grounds’) the truth of mathematical propositions. Let us look at another example: the pair of propositions God exists and God does not exist. Since, by definition, God exists necessarily if he exists at all, either God exists or God does not exist is necessarily true. Assume, for the sake of the argument, that the former is the case. Under these circumstances, (DN1) would imply that my left foot is the truthmaker for God exists! This is clearly unacceptable. Different solutions to the problem of necessary truths have been proposed in the literature. Some theorists simply suggest that the truthmaker definition should be restricted to contingent truths: x makes p true iff (i) p is contingent and (ii) the existence of x necessitates the truth of p.43 The problem is that such a restriction is ad hoc. Why should necessary truths be exempted from the truthmaker definition? Proponents of (DN2) will probably say that necessary truths do not need truthmakers, but this is not enough. They have to explain why the necessity of a proposition gives us a reason to reject the demand for a truthmaker. The whole truthmaking approach is based on the intuition that propositions “owe their truth to something else” (cf. Sect. 1), and an unmotivated restriction of the truthmaker definition would clearly be at odds with this intuition. In the absence of an independent justification of (DN2), therefore, the restriction strategy must be rejected. The other strategies involve more substantial modifications. Some theorists propose to replace the necessitation relation with relevant entailment (roughly: “x makes p true iff the existence of x relevantly entails the truth of p”)44 or a primitive ‘in virtue of’ relation (“x makes p true iff p is true in virtue of x”),45 others add a primitive ‘aboutness condition’ (“x makes p true iff the existence of x necessitates the truth of p and p is about x”)46 or something similar.47 All of these accounts have turned out to be problematic—some because they introduce new primitives (like the last two proposals), others for different reasons.48 I cannot lay out the details here, but it is safe to say that no proposal has gained universal acceptance (or anything close to it). Thus it is reasonable to look for promising alternatives. (DN2)

43 Cf. Jackson (1994/1998, p. 163f). 44 Cf. Restall (1996). 45 Cf. Rodriguez-Pereyra (2005, p. 18). 46 Cf. Künne (2003, p. 158). 47 Smith (1999), e.g., introduces a projection relation to make the notion of “aboutness” more precise. I do not think, however, that his proposal is successful, mainly for the reasons that Gregory (2001) has given. 48 The relevant entailment account, e.g., has very general problems with substantial truthmaker explanations. As Rodriguez-Pereyra (2006, p. 976) points out, a is red does not relevantly entail a is colored, although it is plausible to say that the fact [a is red] is a truthmaker for a is colored. To avoid these difficulties, proponents of the relevant entailment account of truthmaking could move from relevant logical entailment to relevant conceptual entailment—but in that case, their view would become a version of the account that I am defending, not a rival to it. And, even more worryingly, it would not be easy to motivate the revised account if I am right that classical conceptual entailment suffices to solve the problem of necessary truths.

123

428

Synthese (2011) 181:413–431

Here is the point where our newly developed truthmaker definition comes in: (DCE3)

x makes p true iff there is a description δ such that δ is the basic description of x and (δ)exists conceptually entails p.

As it seems, this definition solves the problem of necessary truths effortlessly: neither God exists nor God does not exist is made true by my left foot, because neither of these propositions is conceptually entailed by (γ ) exists, where γ is a basic (microphysical) description of my foot. The same holds, presumably, for 2 + 2 = 4. There is an obvious objection to (DCE3), though. Consider analytic truths like Mothers are female. Since analytic truths are conceptually entailed by every other proposition, (DCE3) implies that my left foot is a truthmaker for Mothers are female, which is almost as counterintuitive as the claim that my left foot is a truthmaker for 2 + 2 = 4. Indeed, if the logicists are right and mathematical truths are reducible to logical truths, (DCE3) also implies, like (DN1), that my left foot is a truthmaker for 2 + 2 = 4. However, all it takes to avoid the problem of analytic truths is a simple modification of our original definition: (DCE4)

x makes p true iff (i) p is synthetic and (ii) there is a description δ such that δ is the basic description of x and (δ) exists conceptually entails p.

At first glance, this seems preposterous. I have argued against (DN2) that a restriction of the truthmaker definition to contingent propositions is ad hoc and unmotivated, and that it undermines the basic motivation behind truthmaker theory. So, why should (DCE4) be any better? I think that (DCE4) is indeed better, and the reason is that the restriction to synthetic truths can be motivated independently. More precisely, (DCE4) can be defended by drawing on the idea that analytic truths are “true merely in virtue of meaning”. This idea has come under attack recently,49 but it is quite intuitive, and I am rather confident that some version of it can be defended.50 If I am right about this, we can say the following: Synthetic propositions are true because of the way (a certain part of) the world is, i.e. their truth is explained by truthmakers. Analytic propositions, by contrast, are true independently of the world. Their truth has a different explanation. It would not be quite right to say that they are “true merely in virtue of meaning” or that their truth “can be explained by their meaning alone”, since we talk about propositions, not sentences, and propositions do not have meanings—they are meanings. Still, I think that the same basic idea can also be formulated for propositions, e.g. in roughly the following way: analytic propositions are true in virtue of ‘what they are’, i.e. in virtue of the components they contain and the way these components are put together. They are true in virtue of their ‘intrinsic nature’, and not in virtue of something external—something ‘in the world’ that makes them true. According to this account, the proposition Mothers are female does not need a truthmaker, while God exists clearly does. Whether 2 + 2 = 4 needs a truthmaker 49 Cf. Boghossian’s (1996) arguments against the metaphysical interpretation of this phrase (which is the one that I am concerned with). I set these arguments aside here, together with the general attacks on the notion of analyticity by Quine (1953/1980) and Williamson (2006), but I agree that all these objections have to be addressed. I plan to do this in another paper. 50 Some recent defenses include Russell (2008) and Hofmann and Horvath (2008).

123

Synthese (2011) 181:413–431

429

depends on the question whether mathematical truths are analytic or synthetic, i.e. whether the logicist reduction of mathematical truths to logical truths is successful or not. This is as it should be. Intuitively, one of the main attractions of logicism is that it promises to get rid of mathematical truthmakers sui generis (ontologically fundamental mathematical objects, properties and facts) without getting rid of mathematical truths. These considerations show that there are good reasons to restrict the truthmaker definition to synthetic truths (reasons, by the way, that are also accepted by other truthmaker theorists51 ), and that the move from (DCE3) to (DCE4) is not an unmotivated, ad hoc measure designed to avoid specific counterexamples. Let us consider a final objection (adapted from Restall 1996). It runs as follows: Since the proposition Electrons exist is synthetic, the conjunctive proposition Electrons exist & mothers are female is also synthetic. Furthermore, the latter proposition is conceptually entailed by the former, since Electrons exist entails itself and also every conceptual truth. According to definition (DCE4), this implies that electrons are truthmakers for Electrons exist & mothers are female. The problem arises if we combine this result with the conjunction principle (a principle widely accepted among truthmaker theorists): (C)

If x is a truthmaker for p&q, then x is a truthmaker for p and x is a truthmaker for q.

From (C) and the proposition that electrons are truthmakers for Electrons exist & mothers are female, we can derive that electrons are also truthmakers for Mothers are female—which is exactly the result we have tried to avoid. How can we reply to this objection? We could simply reject (C), of course, but this reaction might seem ad hoc. I think there is a better answer available: Modifying a suggestion from D.M. Armstrong,52 we can restrict (C) to purely synthetic truths, where purely synthetic propositions are those that do not contain any analytic propositions as logical constituents: (C*)

If x is a truthmaker for p&q, and p&q is purely synthetic, then x is a truthmaker for p and x is a truthmaker for q.

I think that (C*) is a version of the conjunction principle which is intuitively appealing and which does not commit us to the conclusion that electrons are truthmakers for Mothers are female (because the conjunction Electrons exist & mothers are female is not purely synthetic). So, I conclude that definition (DCE4) can be successfully defended, and that we are entitled claim that this definition solves the problem of necessary truths.

51 Simons (2007), e.g., also makes the point that truths that are “true in virtue of meaning” do not need

truthmakers. 52 Cf. Armstrong (2004, p. 11), where Armstrong defends a restriction of the conjunction principle to

purely contingent truths.

123

430

Synthese (2011) 181:413–431

6 Conclusion The first part of my paper concerned the nature of truthmaker explanations. I argued that these explanations have a ‘simple’ and (sometimes) a ‘substantial’ part. The simple part has the form “p is true because [p] exists”, and I suggested that analyzing these explanations is a task that can be left to theorists of truth. What is central to truthmaking theory, I argued, is the substantial part, which has the form “[p] exists, because [q] exists”. According to my first main thesis, these explanations should be construed as reductive explanations. On this basis, I proposed a general definition of truthmaking, (DR), that defines “making true” in terms of reductive explanation. This general definition constituted the starting point for the second part of the paper, where I explored the implications of combining (DR) with the conceptual entailment approach to reductive explanation, and arrived at a new analysis of the truthmaking relation, (DCE3). Finally, I argued that this analysis avoids one major problem of classical truthmaker necessitarianism, the problem of necessary truths. I do not think that my account solves all the problems that have been raised for truthmaker theory.53 Indeed, I think that there is still a lot of work to be done. But I hope that the account of truthmaking explanation that I have given here, and the corresponding formulations of the truthmaker definition and the truthmaker principle, can provide a solid foundation for further theoretical progress. Acknowledgements I would like to thank Vuko Andri´c, Ansgar Beckermann, Lars Dänzer, Frank Hofmann, Romy Jaster, Tim Kraft, Christian Nimtz, Adolf Rami, Torsten Wilholt, Julian Zurek and two anonymous referees for this journal for helpful comments on previous versions of this paper and for stimulating discussions of the issues involved.

References Armstrong, D. M. (1989). C.B. Martin, counterfactuals, causality, and conditionals. In J. Heil (Ed.), Cause, mind, and reality (pp. 7–15). Dordrecht: Kluwer. Armstrong, D. M. (1997). A World of States of affairs. Cambridge: Cambridge University Press. Armstrong , D. M. (2002). Truths and truthmakers. In R. Schantz (Ed.), What is truth? (pp. 27–37). Berlin: Walter de Gruyter. Armstrong, D. M. (2004). Truths and truthmakers. Cambridge: Cambridge University Press. Ayer, A. J. (1946/1952). Language, truth and logic (2nd ed.). New York: Dover. Beebee, H. & Dodd, J. (Eds.). (2005). Truthmakers. The contemporary debate. Oxford: Clarendon Press. Bigelow, J. (1988). The reality of numbers. Oxford: Oxford University Press. Block, N., & Stalnaker, R. (1999). Conceptual analysis, dualism, and the explanatory gap. Philosophical Review, 108, 1–46. Boghossian, P. (1996). Analyticity reconsidered. Nous, 30, 360–391. Chalmers, D. (1996). The conscious mind: In search of a fundamental theory. Oxford: Oxford University Press. Chalmers, D., & Jackson, F. (2001). Conceptual analysis and reductive explanation. Philosophical Review, 110, 315–360. 53 Since the new analysis is a version of necessitarianism, it commits us to the existence of special truthmakers for inessential predications (i.e. facts or tropes), and has problems with negative propositions (especially negative existentials like there are no unicorns). These difficulties suggest that (DCE4) might be in need of further refinement.

123

Synthese (2011) 181:413–431

431

Daly, C. (2005). So where’s the explanation?. In H. Beebee & J. Dodd (Eds.), Truthmakers. The contemporary debate (pp. 85–103). Oxford: Clarendon Press. Fox, J. (1987). Truthmaker. Australasian Journal of Philosophy, 65, 188–207. Gregory, D. (2001). Smith on truthmakers. Australasian Journal of Philosophy, 79, 422–427. Hofmann, F., & Horvath, J. (2008). In defence of metaphysical analyticity. Ratio, 21, 300–313. Horgan, T. (1984). Supervenience and cosmic hermeneutics. Southern Journal of Philosophy Supplement, 22, 19–38. Horwich, P. (1990/1998). Truth (2nd ed.). Oxford: Oxford University Press. Jackson, F. (1994/1998). Armchair metaphysics. In F. Jackson Mind, method and conditionals (pp. 154–176). New York: Routledge. Jackson, F. (1998). From metaphysics to ethics. A defense of conceptual analysis. Oxford: Clarendon Press. Künne, W. (2003). Conceptions of truth. Oxford: Clarendon Press. Lewis, D. (1994/1999). Reduction of mind. In D. Lewis (Ed.), Papers in metaphysics and epistemology (pp. 291–324). Cambridge: Cambridge University Press. Lewis, D. (1998/1999). A World of Truthmakers? In D. Lewis (Ed.), Papers in metaphysics and epistemology (pp. 315–220). Cambridge: Cambridge University Press. Lewis, D. (1999). Papers in metaphysics and epistemology. Cambridge: Cambridge University Press. Lewis, D. (2001a). Truthmaking and difference-making. Nous, 35, 602–615. Lewis, D. (2001b). Forget about the correspondence theory of truth. Analysis, 61, 275–280. Liggins, D. (2005). Truthmakers and explanation. In H. Beebee & J. Dodd (Eds.), Truthmakers. The contemporary debate (pp. 105–115). Oxford: Clarendon Press. Lowe, E. J. & Rami, A. (Eds.). (2008). Truth and truth-making. Chesham: Acumen. McFetridge, I. (1977/1990). Truth, correspondence, explanation and knowledge. In I. McFetridge (Ed.), Logical necessity and other essays (pp. 29–52). London: Aristotelian Society. Merricks, T. (2007). Truth and ontology. Oxford: Oxford University Press. Molnar, G. (2000). Truthmakers for negative truths. Australasian Journal of Philosophy, 78, 72–86. Monnoyer, J.-M. (2007). Metaphysics and truthmakers. Frankfurt: OntosUniversity Press. Mulligan, K., Simons, P., & Smith, B. (1984). Truth-makers. Philosophy and Phenomenological Research, 44, 278–321. Parsons, J. (1999). There is no ‘truthmaker’ argument against nominalism. Australasian Journal of Philosophy, 77, 325–334. Quine, W. V. O. (1953/1980). Two dogmas of empiricism. In W. V. O. Quine (Ed.), From a logical point of view (2nd ed., pp. 20–46). Cambridge, MA: Harvard University Press. Restall, G. (1996). Truthmakers, entailment and necessity. Australasian Journal of Philosophy, 74, 331–340. Rodriguez-Pereyra, G. (2005). Why truthmakers. In H. Beebee & J. Dodd (Eds.), Truthmakers. The contemporary debate (pp. 17–31). Oxford: Clarendon Press. Rodriguez-Pereyra, G. (2006). Truthmaking, entailment, and the conjunction thesis. Mind, 115, 957–982. Russell, G. (2008). Truth in virtue of meaning. Oxford: Oxford University Press. Schaffer, J. (2003). Is there a fundamental level?. Nous, 37, 498–517. Schnieder, B. (2006). Truth-making without truth-makers. Synthese, 152, 21–46. Sider, T. (2001). Four-dimensionalism. Oxford: Oxford University Press. Simons, P. (2000). Truth-maker optimalism. Logique Et Analyse, 169/170, 17–41. Simons, P. (2007). Truth in virtue of meaning. In J.-M. Monnoyer (Ed.), Metaphysics and truthmakers (pp. 67–78). Frankfurt: Ontos. Smith, B. (1999). Truthmaker realism. Australasian Journal of Philosophy, 77, 274–291. Williamson, T. (2006). Conceptual truth. Aristotelian Society Suppl, 80, 1–41. Wright, C. (1992). Truth and objectivity. Cambridge, MA: Harvard University Press.

123

This page intentionally left blank z

Synthese (2011) 181:433–450 DOI 10.1007/s11229-010-9717-3

No two entities without identity Benjamin C. Jantzen

Received: 18 April 2009 / Accepted: 10 February 2010 / Published online: 22 February 2010 © Springer Science+Business Media B.V. 2010

Abstract In a naïve realist approach to reading an ontology off the models of a physical theory, the invariance of a given theory under permutations of its propertybearing objects entails the existence of distinct possible worlds from amongst which the theory cannot choose. A brand of Ontic Structural Realism (OSR) attempts to avoid this consequence by denying that objects possess primitive identity, and thus worlds with property values permuted amongst those objects are really one and the same world. Assuming that any successful ontology of objects is able to describe a universe containing a determinate number of them, I argue that no version of OSR which both retains objects and understands ‘structure’ in terms of relations can be successful. This follows from the fact that no set of relational facts is sufficient to fix the cardinality of the collection of objects implied by those facts. More broadly, I offer reasons to believe that no version of OSR is compatible with the existence of objects, no matter how ontologically derivative they are taken to be. Any such account would have to attribute a definite cardinality to a collection of objects while denying that those objects are possessed of a primitive identity. With no compelling reason to abandon the classical conception of cardinality, such an attribution is incoherent. Keywords Cardinality · Identity · Indistinguishability · Hole argument · OSR · Quantum particles · Space time points · Structural realism 1 Introduction Scientific realism is the view that our successful scientific theories provide partial inventories of what exists, whether observable or not. It is commonplace to cast debate over the viability of this view in terms of two opposing arguments: the ‘no miracles’ B. C. Jantzen (B) Carnegie Mellon University, Pittsburgh, USA e-mail: [email protected]

123

434

Synthese (2011) 181:433–450

argument in favor of realism (e.g. Putnam 1979) and a ‘pessimistic meta-induction’ leveled against it (e.g. Laudan 1981). In the modern literature beginning with Worrall (1989), various notions of ‘structural realism’ have been offered as alternatives that avoid offending against either argument. The metaphysically mildest and historically prior version of structural realism amounts to an epistemic claim: our scientific theories only tell us about the structure of what exists, not about the entities that bear this structure or their intrinsic natures. Increasingly in vogue in both the philosophy of physics and broader philosophy of science literature is a metaphysically more radical structuralism known as ‘Ontic Structural Realism’ or OSR. Championed most notably by Ladyman with French (French and Ladyman 2003) and Ross (Ladyman et al. 2007), OSR promises all the advantages of its epistemic forbear but is supposed to rid us of the unsettling possibility of unknowable objects. As exciting a prospect as this is, the OSR account remains mostly in slogan form. Ladyman—who coined the term ‘Ontic Structural Realism’—describes it broadly as “…any form of structural realism based on an ontological or metaphysical thesis that inflates the ontological priority of structure and relations” (Ladyman 2007). To make such a position concrete requires providing explicit accounts of both ‘structure’ and of ‘ontological priority’. There are an overwhelming number of ways in which this might be done and so one cannot hope to say anything decisive about all conceivable renderings of OSR. Instead, I focus here on one sort of approach, the failure of which has broader implications for theories of identity and individuation unrelated to the debate over OSR. I have in mind the species of OSR that privileges structure as ontologically prior while nonetheless embracing the existence of objects as genuinely existent but supervenient1 ‘placeholders’ in that structure. In this paper, I advance both a narrow and a broad claim concerning OSR theories of this stripe. In a narrow sense, I show that attempts to make the OSR position explicit by cashing out ‘structure’ in terms of relations cannot account for the emergence of objects. Specifically, I show in Sect. 4 that specification of a relational structure fails to fix the cardinality of the set of objects which are supposed to exist as ‘placeholders’ on this structure—no determinate number of objects can be said to arise from such a relational structure. In arguing against the structure-as-relations approach, I will adduce reasons to suspect a more general problem with this brand of OSR. It is essential to such an approach that it be possible for a determinate number of indistinguishable objects to exist that do not also stand in determinate identity relations with one another—it must be possible to have definite cardinality without primitive identity. In Sect. 5, I review why this is not possible under the classical account of cardinality. In Sect. 6, I consider some proposed alternatives to the classical definitions of cardinality that are supposed to have been purged of any dependence on identity. I assess 1 I am using the notion of supervenience loosely, since objects are clearly not themselves properties. The

idea here is that there can be no difference in the set of objects that exist in a world without a difference in the set of relations which give rise to them. Given the evident similarities, ‘supervene’ seemed the most appropriate term for the manner in which objects and the relations they instantiate are connected in the OSR account.

123

Synthese (2011) 181:433–450

435

each of these with respect to three questions: (i) is the proposal internally consistent, (ii) is the proposed definition genuinely independent of identity, and (iii) is there reason to think that the proposed definition of ‘cardinality’ has anything to do with the concept captured by the classical definitions? The variant definitions of cardinality I consider are constructed within a diversity of formalisms, and so it is important to stress from the outset that my discussion cannot be framed in one particular approach such as Zermelo–Fraenkel set theory. To avoid ambiguity, I indicate which formalism is in play whenever a technical result is cited. It is equally important to note that considerations pertaining to question (iii) are formalism independent. Multiple classical definitions are thought to capture, at least in part, a common concept of ‘how many’. Deciding whether a proposed definition is also a representation of the same concept of cardinality is not a question that can be answered within any particular formalism. Instead, appeal must be made to the concept itself, its relations to other concepts, and the interpretation of other parts of the formalism in which the new definition is constructed. Ultimately, all of the proposals considered are shown to fail with respect to one to one or more of the above questions. Furthermore, I suggest that the search for an alternative account of cardinality is ill-motivated. Without grounds to replace the classical notion of cardinality, the assertion of a determinate number of objects that do not also stand in relations of identity with one another is incoherent. This then is my broad claim: it is inconsistent to assert that exactly two (or three, or n) things exist—however derivatively—while simultaneously denying that each is identical with itself and not identical with the others. In the spirit of sloganeering, one might say there cannot be two entities without identity. 2 Motivating OSR Ontic Structural Realism is motivated by a persistent difficulty with traditional realism that is easiest to illustrate with a simplified, exaggerated version of the traditional realist position—call it ‘naïve realism’. The naïve realist attempts to read his ontology off the models2 of a theory by mapping parts of formal representations to postulated entities one to one. For example, suppose some theory of classical dynamics is modeled by trajectories in a two-particle phase space, and consider a single trajectory, q(t), p(t), q  (t), p  (t). Here, the q’s are spatial coordinates and the p’s are 2 Throughout this paper I use the term ‘model’ in two principal senses. In the logical sense it is a collection

of entities, functions, and relations that can be represented by some sentence or theory in a formal language. Such a collection is a model of a given sentence or theory just if the interpretation of that sentence or theory in terms of the entities and relations in question is true (see e.g. van Dalen 2004, p. 71). So in this sense of model, Peano arithmetic is a model of Peano’s axioms in a first order language. In the second sense, a model of a theory is an abstract structure which satisfies the equations or constraints of a physical theory along with whatever boundary conditions are pertinent. In this sense, the solutions of a set of differential equations constitute models of those equations and a balanced reaction is a model of the rules of inorganic chemistry. For the realist, these models correspond directly to the world in some sense. I do not mean to use the term ‘model’ in the sense of a physical or abstract imitation of a portion of the world, such as a scale model of a car or the molecular orbital model of chemical bonding. Which sense of the word I intend should be clear from context.

123

436

Synthese (2011) 181:433–450

momenta. In this case, the naïve realist interprets this expression as representing a world in which one particle—let’s call it ‘left’—is at position q with momentum p at some moment of time t, while the other—call it ‘right’—is at q  with momentum p  . The trajectory q  (t), p  (t), q(t), p(t) lying elsewhere in the same phase space carries a distinct interpretation, namely one in which the left particle occupies position q  with momentum p  while the right particle is at q with momentum p. In other words, this second trajectory represents a world in which the properties of the left and right particles have been exchanged. It is a basic principle of naïve realism that distinct representations entail distinct physical possibilities. So it matters which of the two permuted trajectories is the correct one. Trouble arises for the naive realist when a theory displays certain formal symmetries. Suppose for instance that the dynamical theory considered above is invariant under exchange of particle coordinates such that, if the trajectory described by q(t), p(t), q  (t), p  (t) is a model of the theory, so too is the trajectory q  (t), p  (t), q(t), p(t). This would be the case in classical mechanics if the particles possess the same mass. These two models of the theory provide distinct representations of the history of the world, and thus for the naïve realist there are two possible histories of the world according to this theory: one in which ‘left’ is at q(t) with momentum p(t) while ‘right’ is at q  (t) with p  (t), and another history in which the properties of the particles are permuted. Furthermore, there is no way for the theory to choose between these possibilities.3 Such an inability of a theory to uniquely select from among inequivalent representations of the world has been referred to as ‘indeterminism’4 in the literature, but might better be called ‘incompleteness’. Naïve realism leads to this sort of incompleteness given certain symmetries, namely automorphisms of the model. We can make this problem more concrete by considering an instance in which realism, naïve or otherwise, has struggled to accommodate symmetry in a theory. Perhaps the most prominent example in the philosophical literature comes from spacetime physics, where a symmetry of the General Theory of Relativity (GTR) has featured in debates over realist interpretations of that theory. Specifically, the general covariance of the field equations has been argued to present a problem for the would-be advocate of substantivalism—the view that space-time is a thing which is possessed of non-material properties and which may exist independent of any material things. 3 Of course, in classical mechanics, a trajectory is uniquely picked out once an initial point in the phase space is stipulated. That is, once the world starts on one of the two trajectories, classical mechanics keeps it there. But without this arbitrary initial condition—arbitrary in the sense that the theory lacks the resources to distinguish between the possibilities—the permuted trajectories are both models of the dynamics. 4 See Earman and Norton (1987). If a scientific theory is incomplete in this sense, it fails to pick out a unique

description of the world. I prefer to call such a failure ‘incompleteness’—and such a theory ‘incomplete’— in order to distinguish it from the more common notion of indeterminism. A theory that is incomplete may nonetheless pick out a unique sequence of future states given an initial state. That is, incompleteness does not entail indeterminism. Even in cases such as GR, a failure to specify a unique description may not constitute a failure to specify a full set of physical observables for all time. Whether or not incompleteness for a given theory lines up with indeterminism depends both on the details of the theory and on the strictness of one’s notion of determinism. Butterfield (1989) has argued, for instance, that a particular definition of what I’ve called incompleteness can be constructed for theories formulated on a manifold (like GTR) and which really ought to be viewed as indeterminism in the standard sense.

123

Synthese (2011) 181:433–450

437

There is much to recommend a substantival view of the theory to the scientific realist: GTR admits of solutions devoid of matter, and the ostensibly non-material spacetime component of the theory both acts upon and is acted upon by the material components. If this position is inconsistent, the realist must take quarter in less intuitive territory. A model of GTR consists of two components: a manifold M of spacetime points and a set of one or more fields specifying distributions of properties over these points. Generally, a metric field g specifies the geometrical properties of space (e.g. distance between points). Additionally, a non-vanishing stress-energy tensor T may be introduced that specifies the distribution of matter and energy over M. We may denote a particular model then by a triplet M, g, T . In the simplest substantivalist scheme, it is the manifold M of such a model that is identified with an independently existing space-time.5 This straightforward identification, however, opens the substantivalist to the problems of the naïve realist. It is a basic feature of Einstein’s field equations that they are generally covariant under diffeomorphisms of the underlying manifold M. If M, g, T  is a model of GTR, then so too is M, d ∗ g, d ∗ T  where d is a diffeomorphism of M onto itself and d ∗ g and d ∗ T are the drag-along fields resulting from the diffeomorphism. The famous and well-worked “hole argument” of Earman and Norton (1987) exploits this symmetry to exact a price for adopting the naïve substantival position. It asks us to consider a particular sort of diffeomorphism—which we may call the ‘hole transformation’—that is simply the identity mapping everywhere except within some finite region (open subset) of M (the ‘hole’). So, under the hole transformation d, the drag-along metric d ∗ g differs from g only within the hole region and similarly for d ∗ T . The hole argument then runs as follows (Norton 2004): Given two distributions of metric (g) and matter fields (T) related by a hole diffeomorphism, the manifold substantivalist must maintain that the distributions represent distinct physical possibilities. However, these distinct physical possibilities are observationally equivalent— none of the invariants corresponding to physical observables in GTR differ between the two. Furthermore, the laws of GTR cannot decide between the two solutions given a specification of g and T everywhere outside of the hole. Therefore, the manifold substantivalist is committed to an unwarranted bloating of our physical ontology, one that renders GTR incomplete. The incompleteness ascribed to GTR by the hole argument is just an instance of the general incompleteness plaguing the naïve realist: if components of the models of a theory—manifold points in this case—are taken to represent physical objects, with each distinct combination of components corresponding to a distinct physical possibility, then invariance of the theory with respect to automorphisms6 on the set of those components leads to incompleteness. GTR cannot specify which physically distinct but qualitatively indistinguishable situation obtains. It is this tension between the realist’s urge to read ontologies off of models and his desire to avoid the incompleteness associated with symmetric models that OSR hopes

5 Earman and Norton (1987) have defended this identification as the natural choice for the substantivalist. 6 The diffeomorphisms relevant to the hole argument are automorphisms.

123

438

Synthese (2011) 181:433–450

to resolve. The proponent of OSR denies a premise of the naïve realist. There would be no problem of incompleteness according to OSR if we deny that any theoretical terms directly refer to objects, and instead look to the theory to tell us only what structure obtains. If objects are individuated solely on the basis of their structural roles, then the multiple solutions leading to incompleteness would collapse into redundant descriptions of one and the same physical possibility. What I’m calling OSR can thus be seen as the conjunction of two claims: (i) Theories describe structure in the world, structure being what fundamentally exists, and (ii) objects exist but are individuated solely by the ontologically prior structure in which they are the property-bearers—they lack a primitive identity.7 To return to our concrete example, a number of ontological theories falling under this rubric have been proposed that purport to defeat the hole argument for GTR. These include ‘metric-field substantivalism’ (Hoefer 1996), ‘sophisticated substantivalism’ (Pooley 2005), and the ‘reflexive definition’ of spacetime points (Stachel 2002). While there are important differences amongst these theories, they have in common the OSR strategy for avoiding the incompleteness of naïve realism. For the hole argument to go through, we must accept that there are distinct possible worlds differing only in which point is ascribed which value of a field, an observationally indistinguishable difference. But this metaphysical distinction between possible worlds can only be drawn if spacetime points can be said to have the non-qualitative property of being identical with themselves. Proponents of these OSR theories suggest instead that, while spacetime points are genuine things in the world, the points are bearers only of qualitative properties—they do not possess a primitive identity, and so the incompleteness implied by the hole argument vanishes. Under this view, there is an element of physical reality that corresponds to the point manifold M of a GTR model. However, the points comprising this object supervene on the topological and geometrical relations amongst the points, not the other way around. Ontologically, structure precedes spacetime points. Thus, any mathematical model with the appropriate structure is a representation of the physical object. This structural feature of the equivalence class of GTR models related by diffeomorphism is uniquely picked out by the theory—no incompleteness obtains. Put more generally, OSR resolves the difficulty inherent in naïve realism by asserting that, while the formal components of the models of a theory may stand in identity relations with one another, no such relations obtain for the objects these models describe. To return to the toy example with which I began, it is the case

7 Stachel (2006) decomposes the logical space of possibilities as regards the ontological priority of things

vis-à-vis relations into four types of theory. If I might broaden Stachel’s categories a bit by considering the more general notion of ‘structure’ in place of ‘relations’, then there are two possibilities corresponding to OSR: (i) there is only structure—any reference to a putative object is shorthand for more structure, and (ii) structure is ontologically primary and objects, while they exist, are parasitic on this structure. I am focusing only on those species of OSR that endorse (ii)—those theories that suppose we can have structure and objects too, though structure is ontologically prior. The accounts cited in the text are all instances of such theories. Stachel (2002) in particular defends this sort of account as do (Esfeld 2004; Esfeld and Lam 2008). While it is not entirely clear, it seems plausible—their choice of title notwithstanding—to read (Ladyman et al. 2007) along these lines as well (Hawley 2008).

123

Synthese (2011) 181:433–450

439

that q(t), p(t), q  (t), p  (t) and q  (t), p  (t), q(t), p(t) are distinct because these two ordered 4-tuples are not identical. However, if we deny that particles stand in any identity relations, then there can be no fact as to which particle bears which coordinates and momenta. Both trajectories describe the same singular state of affairs in which there is one instance of a particle at q(t) with momentum p(t) and one instance of a particle at q  (t) with p  (t) at time t, and nothing further can be said regarding which is which.

3 A problem for OSR The sort of OSR theory I have been considering is one in which objects are retained as relata or as the instantiators of structure. Presumably, this means retaining or recovering facts about how many objects exist. I take it as a premise that any successful ontology of objects must be capable of expressing the claim that a determinate number of objects exists in the universe or in some portion of the universe. Let me emphasize that this demand pertains only to ontological claims. Epistemically, it may be the case that, whatever one’s ontology, it is unknowable or unknown how many entities of a kind exist, be they structures, relations, or objects. As an epistemic claim, it is always coherent to assert that there exist an indefinite number of objects, so long as this assertion is understood to mean that one does not know what the fact of the matter is. It is another thing entirely to assert that there is no fact of the matter. While quantum field theory may convince you that at some level of description this is actually the case, I am claiming that any approach to ontology which admits of objects but is committed a priori to there being no fact of the matter how many there are is not a satisfying approach. If one’s ontology admits objects, it ought to admit determinate numbers of them. To avoid incompleteness in the manner I have sketched above, the proponent of an OSR theory can only admit of objects which do not stand in the sort of non-qualitative relations of identity that result in incompleteness for the naïve realist. I have referred in passing to the objects of OSR as lacking a ‘primitive identity’. In order for the critique that follows to gain purchase, this notion needs to be rendered as explicitly as possible. Throughout this paper I use the term ‘primitive identity’ to refer to an irreducible property of self-identity, roughly what Adams calls a ‘thisness’ (Adams 1979). This notion can be understood as the non-qualitative property each entity bears of being identical with itself, or more accurately as the relation in which each entity stands with itself but no others. If a is identical with b in this sense, a and b refer to the same entity. Entities that are not identical are called ‘distinct’. I intend this notion of primitive identity in the thinnest metaphysical sense. By asserting a primitive identity I make no claims about the identification of objects across possible worlds—for my purposes, I only need to consider identity within a given world. For a more thorough discussion of the notion of ‘primitive identity’, see e.g. French and Krause (2006, pp. 11–15). To be as specific as possible, this notion of primitive identity will always be cashed out in mathematical representations as strict numerical identity. For instance, in set-theoretic terms it is the relation corresponding to the diagonal of the domain. See Krause (2008) for a discussion of the distinction between ‘strict’ or ‘true’ identity and, e.g., ‘logical identity’.

123

440

Synthese (2011) 181:433–450

Any plausible version of OSR that admits of objects must endorse the possibility of a determinate number of entities lacking primitive identity. As a matter of course, this possibility is merely asserted. For instance, Hoefer (1996, p. 20) says that “[i]f we have two qualitatively identical objects (whether points or particles) in a physically possible world, they are distinct individuals just because there are two and not one” (emphasis in original). Yet Hoefer, along with most proponents of OSR, neglects to provide an account of this possibility. Without a theory of what it means for there to exist two distinct things, it cannot be baldly asserted that it is possible for them to lack identity relations. In what follows I argue for the implausibility of any such account. To do so, I first consider the proposal to understand structure in terms of relations. I show below that this proposal is inadequate because, in the absence of identity relations, no relational facts are sufficient to fix the cardinality of the objects implied by those facts. If objects are individuated solely by the relational structure they instantiate, it remains indeterminate whether even a plurality of objects exists. Once the problem with the relational approach is made clear, I provide reasons to believe that a similar difficulty obtains for all OSR theories that admit objects. Whatever one takes ‘structure’ to be, I argue that it is incoherent to assert a definite cardinality for a collection of objects while simultaneously denying that primitive identity relations obtain for those objects.

4 Relational structure One way of understanding structure is in terms of relations, the collection of monadic properties, binary, ternary, and n-ary relations that obtain in the world. This is the view of structure Russell had in mind when arguing for his version of epistemic structural realism. As he put it, the notion of structure is applicable only to “…relations and systems of relations” (Russell 1992, p. 249). In the current literature, Stachel (2002) advances an ontology of relations when urging that objects can be ‘reflexively defined’ by the structure in which they occur. On this approach, the idea is to allow for the existence of objects like spacetime points but to insist that this existence and the numerical distinctness of these objects are grounded in ontologically prior relations such as those encoded by the metric field. It is not enough, however, that relational structure entail the existence of some objects. Rather, the existence of a determinate number of objects must follow from a given relational structure. We are left then with the following problem: what can be said about a relational structure so as to fix the cardinality of the set of objects which stand in those relations, assuming those objects do not stand in relations of identity? Certainly it is insufficient to stipulate the arity of each relation in an ensemble of relations. Any such relational structure could be satisfied by one, or two, or any number of objects. For instance, the fact that an ensemble of relations contains a binary and a ternary relation, call them R1 and R2 , is compatible with the existence of only one entity a1 , such that R1 (a1 , a1 ) and R2 (a1 , a1 , a1 ) or of five entities such that R1 (a1 , a2 ) and R2 (a3 , a4 , a5 ). We can make our considerations rather more precise and general by appealing to model theory. In this framework, our question takes the following form: can a the-

123

Synthese (2011) 181:433–450

441

ory (in the logical sense) that stipulates relational structure in some language without identity fix the cardinality of the set of relata in models of that theory? For a first order theory, the answer is plainly no. It is straightforward to prove that, for any theory T expressed in a first-order language without identity, if T has a model of cardinality n then it has a model of every cardinality greater than n.8 The reason is that strict identity cannot be defined in a first order language for which it is not given as a primitive. Without the identity relation, it is possible to add any number of indistinguishable objects to the universe of the model. Of course, if we do add identity as a primitive binary relation and interpret it as strict identity in all models of the theory then there is no such difficulty. But to do so is to insist that elements of the universe of every model possess the sort of primitive identity the OSR proponent is attempting to do without. If the difficulty derives solely from the lack of a primitive identity relation, then it might seem that the problem could be averted by moving to a second-order theory in which identity can be defined. However, it is the case that a binary relation co-extensive with strict identity can be defined in second-order theories only if it is assumed that second order models are ‘full’—that is, only if we insist that all models of our second order theory contain every possible relation that can be defined extensively on the domain.9 While this might be reasonable and quite useful in mathematical contexts, it is eminently absurd if we intend the models of our theory to represent the relations in the world. If we must have full models to define identity, then OSR requires a priori the existence of things in the world corresponding to every possible relation of every arity. Aside from the fact that this forces upon us an inordinately bloated ontology, it makes the specification of ‘structure’ moot. There would be nothing to discover about the world concerning structure since we know in advance it must be the most complex structure possible.

5 Cardinality and identity If we take relational structure to be ontologically basic, then we have seen that there is no sense in which a determinate number of objects may be said to arise from this structure. But there are many richer notions of structure, and one might hope that one or more of these could sustain a version of OSR that admits of objects. The failure of the relational approach however, suggests that this is not the case. The essential problem for this brand of OSR is the deep dependence of the notion of cardinality upon that of identity. In the standard set-theoretic account (in ZFC),

8 Roughly, if T has a model M where |dom(M )| = n, then we can construct an extension M of M 1 1 2 1

by adding one or more elements to the domain that stand in exactly the same properties, relations, and functions as some particular element from dom(M1 ). It can be shown that the resulting model is elementarily equivalent to the original model, and thus all statements true in M1 are also true in M2 . Since in our construction we could add as many of these duplicate elements as we like, there are thus models of T of every cardinality greater than n. For an interesting variant of the relevant proof, see French and Krause (2006, pp. 252–254). 9 Krause (2008, p. 12) points this out as well.

123

442

Synthese (2011) 181:433–450

the cardinal number associated with a set A is just the smallest ordinal number n such that there is a bijection from n to A. Ordinals can only be defined if asymmetric relations can be defined, and the latter require an invocation of identity (French and Krause 2006, p. 284). Even were this not so, even if ordinals were handed to us by God, we could only select a unique such ordinal as the cardinal of a set by appealing to bijection. Yet bijections are just a special class of functions, and the notion of a function depends upon the primitive identity of elements in a set. This is obvious in the case of finite sets for which the associated cardinal number n is just the ordinal {0, 1, . . . , n − 1}. That is, we understand the ‘size’ of a finite set A to correspond to what we get by ‘counting’ the elements of A in the normal, intuitive sense of counting. Counting is really just indexing, affixing to each element of A a unique label as we do when we point to each element and say “one,” “two,” “three,” etc. But indexing requires that we be able to pick out individual elements of A without reference to their properties. So without the primitive distinctiveness of the elements of A, we cannot provide an account of cardinality or the ‘size’ of a finite set. The same applies to sets in toto. That a primitive identity relation amongst elements is so freely appealed to in set theory is not surprising. Classically, it is simply assumed that the elements of a set are possessed of a primitive identity. Cantor puts it this way: “By a [set] we are to understand any collection into a whole M of definite and separate objects m of our intuition or our thought” (Cantor 1915, p. 85). This is also the view of modern set theory endorsed by Krause (1992, footnote 3), who recognizes this feature as incompatible with the OSR agenda. That the notion of cardinality as it applies to physical objects also involves a notion of primitive identity was recognized by Adams (1979) when considering Black’s famous spheres. In that example, intended to undermine the status of Leibniz’s Principle of the Identity of Indiscernibles as a logical truth, Black (1952) asks us to consider a universe that is empty save for two qualitatively identical iron spheres. Adams concedes that one can speak of this universe of two indistinguishable spheres differing solo numero from a universe that contains only one. But, he says, to grant such a distinction “…is to hold that the thisnisses of the two globes are metaphysically primitive” (Adams 1979, p. 16). For the globes in Black’s example to be of a definite number, we must be able to count them. To count them means that they must be distinct, and that they can be brought into correspondence with the natural numbers. Of course, this does not mean that there is a way to settle the epistemological question of which is sphere ‘1’ and which ‘2’. It is merely to say that the pair of spheres can be represented by a set, whose elements can be labeled ‘1’ and ‘2’. To count things, they must be distinct. It is incoherent to assert that there exists a set of definite cardinality composed of indistinguishable elements that lack a primitive identity to ground their distinctness. But if this is true, then it is not possible to make the essential move of OSR and embrace objects while denying that they stand in primitive relations of identity with one another. No matter what structure we have in mind, the problems of automorphism invariance will obtain so long as qualitatively indistinguishable objects are thought to exist and to possess a primitive identity. Yet in denying identity relations amongst the objects, the proponent of OSR blocks any claim to a determinate number of existing objects.

123

Synthese (2011) 181:433–450

443

6 An obvious objection: alternate notions of cardinality One might urge at this point that I have put too much weight on the classical notion of cardinality. Surely if we are going to shift our ontology from objects to relations, some adjustments will have to be made in related concepts. So why not resolve the problem by providing an alternate account of cardinality, one that doesn’t depend on identity? Since there are at least three proposals in the literature that purport to do just that, perhaps we could simply adopt one of these. There are two reasons to reject this approach. First, the available proposals are each inadequate for one reason or another. Second, the search for an alternate definition of cardinality is ill-motivated in the first place. The clarity and foundational role of the classical notion of cardinality throughout metaphysics, mathematics, and the sciences outweighs the metaphysical gains that may follow from replacing it. With respect to alternate notions of cardinality, Tzouvaras (2005) has provided a broad analysis of the possibilities within set theory. His analysis is motivated by the claim that the concepts of cardinality and well-orderability are logically independent, at least in that one might have a set which is not well-orderable but which nonetheless possesses a definite cardinality. His explicit concern then is with well-ordering, but the definition of well-ordering makes use of the relation of identity. Showing that well-orderability and cardinality can be pried apart is clearly a prerequisite for the sort of definition of cardinality we are seeking—Tzouvaras’ search for a definition of cardinality that stands independent of well-orderability represents a first step in the search for a notion of cardinality divorced from identity. To explore the possibilities, Tzouvaras lays down a set of plausible conditions for a notion of cardinality in any model of Zermelo–Fraenkel set theory (ZF). If M is a model of ZF, then he claims that a notion of cardinality is a mapping C ⊂ M such that (Tzouvaras 2005, p. 123): (1) The domain of C is M and the range of C is the set of classical cardinal numbers, Card. (2) C(κ) = κ for every κ ∈ Card. (3) For any disjoint sets x and y, C(x ∪ y) = C(x) + C(y). (4) For any sets x and y, C(x × y) = C(x) · C(y). (5) If f : x → y is an injective mapping, then C(x) ≤ C(y). Conditions (1) and (2) require every notion of cardinality to coincide with classical cardinality with respect to the cardinals themselves—an alternate notion of cardinality cannot assign to a classical cardinal a different cardinal. The remaining three conditions ensure that every notion respects the classical calculus of cardinals, at least in form. Whether or not we find these conditions satisfactory, a key result scuttles the project from the perspective of OSR. At the outset, Tzouvaras shows that if the axiom of choice (AC) is assumed to hold, the only notion of cardinality to meet the above conditions is the classical cardinality. Thus, we must abandon AC. Since there is no prima facie reason the proponent of OSR should be tied to AC, let us assume that this result itself poses no problem for the intended metaphysics. However, even without AC, the only notion of cardinality which meets the above conditions for finite sets is again the classical notion. So for the finite sets it seems these concepts cannot be

123

444

Synthese (2011) 181:433–450

teased apart at all, at least not as long as we stay close enough to classical cardinality to respect the above conditions.10 This result is problematic for OSR. If a non-classical notion of cardinality which makes no use of identity is to be endorsed, we must exclude the possibility of finite sets and thus the finite collections of objects they represent. In adopting one of Tzouvaras’ alternative cardinality notions the proponent of OSR commits himself to the further claim that there exist no finite collections of entities. This is an unreasonable assertion to make a priori. Domenech and Holik (2007) attempt to move further from classical cardinality by leaving classical set theory altogether. Specifically, they attempt to construct a well-motivated definition of cardinality within the formalism of Quasi-Set Theory (Q), stripped of its original primitive cardinality function (see below). Quasi-Set Theory—see e.g. Da Costa and Krause (2007), French and Krause (2006), Krause (1992), and Krause et al. (2005)—is a conservative extension of Zermelo–Fraenkel set theory with urelements (ZFU). The axioms of the theory are given in a first-order language without identity, that is, without a binary predicate in the language that is to be interpreted as strict identity in all models. In addition to the usual logical symbols, the language contains two binary relations: membership ‘∈’ and indistinguishability ‘≡’. It also contains a handful of monadic predicates, the most important of which distinguish between two types of urelement. One sort, which Krause called the ‘M-atoms’ (French and Krause 2006, p. 275), are intended to behave like those of classical ZFU. The other urelements, designated the ‘m-atoms’, are supposed to sustain the relation of indistinguishability but not that of identity. To get an idea how this works, consider first the primitive relation ‘≡’ which is postulated to apply to all objects of the theory (urelements and quasi-sets) and to constitute an equivalence relation. Since it is only an equivalence relation, objects can be indistinguishable while nonetheless failing to be coextensive: if x ≡ y it may still be the case that there exists a qset z such that x ∈ z but ¬(y ∈ z). Thus ‘≡’ is not the identity relation. The theory lacks identity amongst its primitive relations, so the best one can do to represent identity is to define a relation that satisfies the (Hilbert–Bernays) axioms for first-order identity. I’ve already remarked on the failure of first-order identity to capture full identity, but French and Krause implicitly take such a relation to be close enough to count as identity in developing their formalism. The relation they introduce is just extensional identity: x =E y =def (Q(x) ∧ Q(y) ∧ ∀z(z ∈ x ↔ z ∈ y)) ∨ (M(x) ∧M(y) ∧ ∀ Q z(x ∈ z ↔ y ∈ z)) Here, Q(x) is a predicate indicating that x is a quasi-set (qset), M(x) is true if x is an M-atom, and ∀ Q z means that z ranges over all qsets. This relation of extensional 10 In an attempt to separate notion of cardinality from well-orderability in finite sets, Tzouvaras also

considers non-standard models of Peano arithmetic. But even in this case any notion of cardinality applicable to the standard coded subsets of the universe of any model of the theory is identical with classical cardinality and so depends upon identity. Thus, we seem to have much the same problem—to reject notions of cardinality that depend upon well-orderability (and thus identity), we must reject the existence of those sets which most intuitively represent finite collections of objects.

123

Synthese (2011) 181:433–450

445

identity is defined only for those objects of Q that are not m-atoms. An axiom of the theory further guarantees that if x =E y then y can be substituted for free instances of x in any well-formed formula, and so extensional equality is a first-order identity relation for the classical things in Q. No analogous relation is explicitly defined for the m-atoms. The intended interpretation of the theory involves two sorts of objects, one for which an identity relation obtains and is represented by the defined relation ‘=E ’, and one for which no such relation is defined. For the latter, the m-atoms, we cannot say which is which beyond an equivalence class of the indistinguishability relation, ‘≡’. The appeal of this formalism for expressing the thesis of OSR is evident. According to OSR, the world is populated exclusively by relations and by objects that are well-represented by m-atoms. The problem I have framed for the OSR proponent can be stated in the Quasi-Set formalism as that of defining cardinality for qsets which contain only indistinguishable m-atoms. For such qsets—dubbed ‘pure’ qsets (French and Krause 2006, p. 277)— there is ostensibly no identity and so no way of defining an asymmetric relation. Without such a relation, one cannot construct ordinals and thus cannot construct the analogue of classical cardinals. Even worse, one cannot define functions in the usual way since “…a function cannot distinguish between arguments and values if there are m-atoms involved” (French and Krause 2006, p. 281). Nonetheless, Domenech and Holik (2007) argue that a well-motivated definition can be provided for the cardinality of such a qset. If we call the qset in question X then—informally speaking—their approach is to construct qsets from X which resemble classical singletons, argue that these ‘quasi-singletons’ should be assigned a cardinality of 1, and then to count how many such quasi-singletons can be extracted from X. While there isn’t space to recount the entire construction here, it will suffice for our purposes to focus on one particular theorem used and the implicit assumption which underlies it. To begin with, the quasi-singletons are constructed as follows. Given a non-empty base qset X and some x ∈ X, let P(X) denote the power qset of X. (Q contains an axiom for the existence of a power qset, defined in complete analogy to the classical power set of ZFU.) Next, construct the qset of all sub-qsets of X which contain x: Ax =E [a ∈ P(X) : x ∈ a]. The quasi-singleton x is then given by: x =E



a.

a∈A x

To motivate the definition of cardinality they construct, Domenech and Holik attempt to provide a convincing reason to suppose that the qset x should be assigned cardinality 1. To this end, they provide a proof that the only sub-qsets of x are ∅ and itself. This is certainly a property one would expect of a collection containing just one object. The problem is that, for the proof to go through, much more needs to be assumed about the existence of the various sub-qsets of X than is guaranteed by the

123

446

Synthese (2011) 181:433–450

axioms of Q. The authors implicitly assume that, given a qset X, every possible subqset of X exists—P(X) is the ‘real’ power qset of X. But this assumption is sufficient to give us an identity relation on the m-atoms comprising X, at least relative to the power qset of X. Specifically, we can offer the following nominal definition: x ∼ y =def y ∈ x. It is straightforward to prove11 that if X is a ‘pure’ qset (one containing only indistinguishable m-atoms) then given the assumptions of Domenech and Holik, x ∼ y is equivalent to the first-order indiscernibility formula for Q restricted to the power set of X.12 In a language with finitely many predicates, this formula has all the properties of first-order identity (Ketland 2006, p. 307). Put another way, if we restrict the range of our quantifiers to the qsets in P(X) then ‘∼’ is a first-order identity relation for m-atoms just as ‘=E ’ is an identity relation for everything else. With this identity relation, we can recover the cardinalities assigned by Domenech and Holik simply by counting the distinct elements of X (distinct with respect to the identity relation ‘∼’) in the old-fashioned way. Their appealing definition of cardinality was purchased at the price of introducing, whether acknowledged or not, a relation of identity. One might object at this point that the assumption made by Domenech and Holik only provides a first-order identity relation, not strict identity. This is true, but the same could be said for the relation ‘=E ’ which is supposed to capture classical identity for the M-atoms and other classical parts of the theory. These first-order identity relations stand or fall together. Either there is no classical set-theoretic identity relation for either M-atoms or m-atoms or there is a relation ‘close enough’ to count as identity for both. It is worth noting here that the interpretation of Quasi-Set Theory as a whole is ambiguous in this way. We are told that a nominally defined identity relation good enough to capture identity in the classical part of the theory holds for M-atoms but not for m-atoms. Yet an equally good first-order identity relation can be nominally defined for the m-atoms without violating any axioms of Q. This is guaranteed by the fact that the language of Q contains only finitely many predicates. The idea seems to be to ignore this additional first-order identity relation which holds for all models of Q, and instead make reference to m-atoms only up to an equivalence class of ‘≡’. But this looks remarkably like Weyl’s strategy for which Quasi-Set Theory was explicitly posited as an alternative (see French and Krause 2006, pp. 261–264). Whether or not one is troubled by the approach of ignoring one of the identity relations inherent to Q, 11 Proof sketch Suppose y ∈ x. By construction of x it is obvious that ∀ z∈P(X) (x ∈ z → y ∈ z). To show that ∀z∈P(X) (y ∈ z → x ∈ z), suppose there exists a qset z ∈ P(X) such that y ∈ z but x ∈ / z. Then, by construction of y it must be the case that x ∈ / y since x is missing from at least one qset in A y . Obviously x ∈ x, so it must be the case that ¬(x =E y). However, y ∈ x ⇒ x ∈ A y ⇒ y ⊆ x. But by Proposition 4.4 of Domenech and Holik, the only non-empty subsets of x are extensionally equivalent to x, and so we have a contradiction. Thus, y ∈ x ⇒ ∀z∈P(X) (x ∈ z ↔ y ∈ z). Since all members of the pure base qset X are indistinguishable, y ∈ x ⇒ ∀z∈P(X) ( (x ∈ z ↔ y ∈ z) ˆ (x ≡ y)). But the latter is the first-order indistinguishability relation relative to the power set of X, and is thus a first-order identity relation on P(X).   12 It is more complicated to state the relation for mixed qsets, and I do not attempt this here. Since we are interested in the case of indistinguishable objects, a consideration of pure qsets suffices.

123

Synthese (2011) 181:433–450

447

it is the case that Domenech and Holik use the illicit relation to define cardinality, and thus break the important distinction between m-atoms and M-atoms. For this reason, their account does not provide a notion of cardinality divorced from identity. In his original formulation of the theory, Krause (1992) did not make use of the identity relation definable for m-atoms. Instead, he introduced ‘quasi-cardinality’ as a primitive function ‘qc( )’ in the language of Q. The axioms of Q concerning quasicardinality ensure that the introduced function obeys a set of conditions analogous to those of Tzouvaras above. However, since one can imagine models of the theory in which this primitive function is interpreted as something other than cardinality, this approach can only be satisfying if there is reason to favor an interpretation in terms of cardinality. It is hard to see how such an interpretation could be supported. To begin with, this primitive function is supposed to differ radically from classical cardinality in the sense that it obtains for collections of objects that cannot be well-ordered or identified in any way. This exacerbates the need for motivating the intended interpretation. But such motivation appears to be provided mostly by postulating that the range of qc is the class of classical cardinals. Whatever support this fact about qc provides, an interpretation in terms of cardinality still looks unappealing when we note that facts about the quasi-cardinality of a pure qset do not derive from any other features of that qset. No definite quasi-cardinality attaches to a given (pure) qset, irrespective of how that set is defined unless stipulated as an independent axiom—the cardinality of pure qsets it is always an ad hoc addendum in Q.13 The freedom to assign a quasicardinality to a pure qset irrespective of other facts about that qset emphasizes the triviality of such a purported cardinality function and undermines any motivation we might have had for interpreting the function in terms of cardinality in the first place. In sum, producing an ostensibly non-classical notion of cardinality by way of a primitive function in the language is ad hoc and liable to reinterpretation as something other than cardinality. Each of the proposals considered above fails to produce an account that can achieve what OSR needs—an explication of cardinality in the absence of identity relations. In the first instance, the notions of cardinality and well-orderability could not be pulled apart at all for finite sets. In the second proposal, a relation of identity was smuggled in, and explains the otherwise surprising cardinality results. And in the last proposal by French and Krause, the defined predicate for ‘quasi-cardinality’ is ad hoc and, in the cases of interest, makes the quasi-cardinality of a collection of objects a brute fact. Thus, we have little reason to grant that the primitive function qc should be interpreted as a sort of cardinality. None of the objections I have raised is a decisive argument against the possibility of a well-motivated notion of cardinality that makes no reference to identity relations. Such a decisive argument—if possible at all—would require committing to very precise accounts of what one means by a ‘well-motivated’ definition and of just how different from classical cardinality one is willing to get and still call the result a ‘cardinality’. Any such account sufficiently restrictive to permit a proof is likely to be 13 Of course, the axioms concerning cardinality in Q are such that, once a cardinality is established for

one qset, the cardinalities of other qsets which are subsets of this initial set have restrictions imposed on what cardinality can be assigned to them.

123

448

Synthese (2011) 181:433–450

unconvincing. Rather, the objections offered above to a handful of specific proposals are taken to be inductive support for the claim that no convincing alternative is likely. Adopting any alternative to classical cardinality means abandoning both a welldeveloped formalism and the strongly intuitive notion of equinumerosity it captures. What reason might compel us to do so? Quantum mechanics is often taken to offer strong evidence to this effect. Tzouvaras (2005, p. 140) and Domenech and Holik (2007, pp. 855–857) all cite the permutation invariance imposed upon states describing multiple particles of the same fundamental type. The fact that the possible states of, say, more than one electron are restricted to an antisymmetric subspace of the total two-particle Hilbert space is supposed to motivate the claim that electrons are non-individuals—if electrons did not stand in relations of identity with one another, then the notion of a permutation wouldn’t even apply and the restriction of possible states to the antisymmetric subspace would simply be a reflection of this fact. But if electrons lack identity, then the fact that we can nonetheless measure the number of electrons in an atom suggests that we need a new account of cardinality. There are multiple problems with this argument. As French and Krause (2006, Chap. 4) have argued to great effect, a picture of particles lacking primitive identity is not the only consistent interpretation of the physics available. For instance, it is still consistent to view quantum particles as possessing identity and to account for the permutation invariance of quantum states by supposing there is a restriction on the space of possible states for composite systems of particles. Alternatively, one could simply deny that there are any particles. After all, the only uncontroversial claim in this debate is that one can count such things as tracks in a bubble chamber, for which identity is unproblematic. One could just as well deny that the number of these things corresponds to a number of particles. Quantum systems may simply have propensities to produce integer valued effects. To turn a popular metaphor around, consider the money in an electronic bank account. There are no actual individual bits of currency, no coins or bills in such an account. In fact, we can view the account as a mereological simple with no parts at all. Rather, the account possesses an integer-valued property we call the ‘balance’. When this account interacts with other accounts—such as when the account holder pays a bill electronically— only discrete changes can be made with respect to this property. But it makes no sense to ask about the identity relations amongst the components of the account, because it has no components—only a property. Perhaps then electrons really are like money in the bank. Whether or not one favors this particular cartoon interpretation, the fact remains that consistent alternatives exist. The physics alone cannot force us to accept that quantum particles exist in definite numbers and yet lack identity relations. In fact, the argument in favor of a new notion of cardinality can be viewed instead as a reductio against the particles-as-non-individuals view. Since classical cardinality— which reduces to counting for finite sets—is deeply intuitive and plays a foundational role in a great deal of other scientific, mathematical, and philosophical theories, it is the sort of notion one would surrender only in the face of contradiction with another, even more fundamental notion. Certainly the vague notion of entities lacking identity relations is not such a candidate. Thus, when faced with the contradiction of

123

Synthese (2011) 181:433–450

449

cardinality without identity, it is our interpretation of quantum mechanics we should jettison, not the notion of cardinality. If it were true, OSR would rid us of the sort of incompleteness plaguing the naïve realist. However, OSR—at least the species that admits the existence of objects—quite generally requires that a definite cardinality be attributed to a collection of entities without identity. This in turn requires a new notion of cardinality. I have argued that the example of quantum mechanics does not provide sufficient reason, independent of concerns over OSR, to favor the rejection of classical cardinality. That leaves only our original concern over incompleteness. Losing our grip on such a clear notion as cardinality seems a high price to pay for resolving concerns over theoretical incompleteness. In the absence of a compelling reason to abandon it, we should embrace the classical conception of cardinality and its immediate consequence: there can be no two entities without identity.

7 Conclusion The problems OSR is intended to resolve are serious, and the solution on offer is appealing. If objects can be viewed as mere placeholders that supervene on an ontologically prior structure then we would have a metaphysics immune to theoretical incompleteness of the sort underlying the hole argument of GTR and plaguing the naïve realist whenever theories are possessed of the relevant symmetries. The structure that is uniquely specified by even a very symmetric theory could be seen to correspond directly with something in the world. Furthermore, this structure would underwrite the objects which started all the trouble, objects like indistinguishable quantum particles and spacetime points. However, the manner in which objects can be said to supervene on structure is mysterious, and must be clarified before any such OSR account can be compelling. An account in which structure is taken to be relational and objects individuated by their relational role cannot do the job for the simple reason that stipulating a relational structure fails to fix the number of objects that can instantiate the relations—model theory tells us that there is nothing one can say about relational structure that would fix the cardinality of its models unless a primitive identity relation is invoked. But this shortcoming is not a quirk of the relational approach. Rather, it is a problem for any account that does away with primitive identity relations amongst objects while nevertheless maintaining that a determinate number of them exist. The reason is simple: the notion of cardinality depends inextricably on identity. Beyond debates over OSR, this is a lesson that must be heeded in broader considerations of identity, individuality, and indiscernibility. Under the classical conception of cardinality, it is simply incoherent to assert a definite number of objects when those objects do not stand in identity relations with one another, and in this case we have no compelling reasons to abandon the classical view. Acknowledgements I would like to thank Jeremy Avigad, Spencer Breiner, Kohei Kishida, Mara Harrell, John Earman, John Norton, Bryan Roberts, Ed Slowik, and two anonymous referees for helpful comments and criticisms on previous versions of this paper.

123

450

Synthese (2011) 181:433–450

References Adams, R. (1979). Primitive thisness and primitive identity. The Journal of Philosophy, 76, 5–26. Black, M. (1952). The identity of indiscernibles. Mind, 61, 153–164. Butterfield, J. (1989). The hole truth. The British Journal for the Philosophy of Science, 40, 1–28. Cantor, G. (1915). Contributions to the founding of the theory of transfinite numbers (Philip E. B. Jourdain, Trans.). Chicago: The Open Court Publishing Company. Da Costa, N. C. A., & Krause, D. (2007). Logical and philosophical remarks on Quasi-Set Theory. Logic Journal of IGPL 2007, 15, 421–431. Domenech, G., & Holik, F. (2007). A discussion on particle number and quantum indistinguishability. Foundations of Physics, 37, 855–878. Earman, J., & Norton, J. (1987). What price spacetime substantivalism? The hole story. British Journal for the Philosophy of Science, 38, 515–525. Esfeld, M. (2004). Quantum entanglement and a metaphysics of relations. Studies in the History and Philosophy of Modern Physics, 35, 601–617. Esfeld, M., & Lam, V. (2008). Moderate structural realism about space-time. Synthese, 160, 27–46. French, S., & Krause, D. (2006). Identity in physics. Oxford: Clarendon Press. French, S., & Ladyman, J. (2003). Remodelling structural realism: Quantum physics and the metaphysics of structure. Synthese, 136, 31–56. Hawley, K. (2008). Throwing the baby out with the bathwater. Retrieved June 29, 2009, from http:// www.st-andrews.ac.uk/~kjh5/OnlinePapers/EveryThingMustGoReview.pdf Hoefer, C. (1996). The metaphysics of space-time substantivalism. The Journal of Philosophy, 93, 5–27. Ketland, J. (2006). Structuralism and the identity of indiscernibles. Analysis, 66, 303–315. Krause, D. (1992). On a Quasi-Set Theory. Notre Dame Journal of Formal Logic, 33, 402–411. Krause, D. (2008). Logical aspects of quantum (non-)individuality. PhilSci Archive. Retrieved March 14, 2009, from http://philsci-archive.pitt.edu/archive/00004375/01/Qua2(Non)Indiv.pdf Krause, D., Sant’Anna, A., & Sartorelli, A. (2005). A critical study on the concept of identity in Zermelo– Fraenkel-like axioms and its relationship with quantum statistics. Logique Et Analyse, 48, 231–260. Ladyman, J. (2007). Structural realism. Stanford Encyclopedia of Philosophy. Retrieved December 20, 2007, from http://plato.stanford.edu/entries/structural-realism/#OSRSpaPhy Ladyman, J., Ross, D., Spurrett, D., & Collier, J. (2007). Every thing must go: Metaphysics naturalized. Oxford: Oxford University Press. Laudan, L. (1981). A confutation of convergent realism. Philosophy of Science, 48, 19–49. Norton, J. (2004). The hole argument. Stanford Encyclopedia of Philosophy. Retrieved December 20, 2007, from http://plato.stanford.edu/entries/spacetime-holearg/#HolArgBri Pooley, O. (2005). Points, particles, and structural realism. In D. Rickles, S. French, & J. Saatsi (Eds.), The structural foundations of quantum gravity. Oxford: Oxford University Press. Putnam, H. (1979). Mathematics, matter, and method (2nd ed.). Cambridge: Cambridge University Press. Russell, B. (1992). The analysis of matter (3rd ed.). New York: Routledge. Stachel, J. (2002). ‘The relations between things’ versus ‘the things between relations’: The deeper meaning of the hole argument. In D. B. Malament (Ed.), Reading natural philosophy. Essays in the history and philosophy of science and mathematics (pp. 231–266). Chicago: Open Court. Stachel, J. (2006). Structure, individuality, and quantum gravity. In D. Rickles, S. French, & J. Saatsi (Eds.), Structural foundations of quantum gravity (pp. 53–82). Oxford: Clarendon Press. Tzouvaras, A. (2005). Cardinality without enumeration. Studia Logica, 80, 121–141. van Dalen, D. (2004). Logic and structure (4th ed.). New York: Springer. Worrall, J. (1989). Structural realism: The best of both worlds?. Dialectica, 43, 99–124.

123

Synthese (2011) 181:451–470 DOI 10.1007/s11229-010-9733-3

Practical success and the nature of truth Chase Wrenn

Received: 26 August 2008 / Accepted: 22 February 2010 / Published online: 11 March 2010 © Springer Science+Business Media B.V. 2010

Abstract Philip Kitcher has argued for a causal correspondence view of truth, as against a deflationary view, on the grounds that the former is better poised than the latter to explain systematically successful patterns of action. Though Kitcher is right to focus on systematically successful action, rather than singular practical successes, he is wrong to conclude that causal correspondence theories are capable of explaining systematic success. Rather, I argue, truth bears no explanatory relation to systematic practical success. Consequently, the causal correspondence view is not in a better position to explain success than the deflationary view; theories of truth are the wrong place to look for explanations of systematic practical success. Keywords

Kitcher · Truth · Deflationism · Correspondence · Success · Action

1 Introduction Perhaps the most popular views of truth these days are deflationary views and causal correspondence views. According to deflationists, truth is not a “natural” property with an essence to be understood in causal terms. If it is any sort of property at all, it is a purely formal or “Cambridge” property. Truth, they say, does not have causalexplanatory power. Causal correspondence views are almost exactly the opposite of deflationary ones. They maintain that truth is a natural property with a causal essence. More specifically, causal correspondentists claim that (a) truth’s nature is encapsulated by a Tarski-style, recursive definition employing no primitive semantic notions but reference, and (b) representational tokens refer to their semantic values in virtue of certain complex causal relations obtaining between them. Reference is or supervenes on one or more C. Wrenn (B) Department of Philosophy, University of Alabama, Tuscaloosa, AL, USA e-mail: [email protected]

123

452

Synthese (2011) 181:451–470

causal relations, and truth is a generalization of reference from singular and general terms to whole sentences.1 The causal correspondence view treats truth as a genuine property, eligible to participate in causal explanations. The deflationary view does not. There are apparently correct explanations, though, that seem to cite truth as a causal factor. Some of the most obvious cases are explanations of successful actions, such as: Jack found Jill because his belief that she had gone up the hill was true. Deflationists owe an account of these explanations that does not require treating truth as a genuine, causally relevant2 property. The most obvious deflationary move is to reinterpret the explanations disquotationally: Jack found Jill because she had gone up the hill and he believed she had gone up the hill. Some critics of deflationism maintain that these explanations are inferior to explanations that posit a robust property of truth that helps to cause actions to succeed. Philip Kitcher, for example, has argued that deflationists can offer only “shallow” explanations of certain successful actions, and they cannot explain systematic practical success at all (2004, pp. 204–211). Causal correspondence views, he thinks, offer deeper explanations, and they can explain systematic practical success. So, he thinks, causal correspondence views are superior to deflationary views of truth. In my view, the sort of objection Kitcher raises to deflationism fails to show that causal correspondence theories have the advantage. Even if deflationary explanations of success are inadequate, causal correspondence explanations are no better. This is because the distinctive feature of causal correspondence theories, the positing of content-constituting causal connections between representations and the world, is irrelevant to explaining practical success. Consequently, the causal correspondence view is no better poised to explain practical success than deflationism is. I make my case for this view as follows. In the next section, I distinguish between singular success and systematic success. I also explain why it is important to consider systematic success more carefully than deflationists often have done. Section 3 summarizes Kitcher’s argument to the effect that systematic success is inexplicable given deflationism but not given causal correspondentism. In Sects. 4 and 5, I argue that Kitcher fails to show that causal correspondence explanations are better than deflationary explanations. I do that in Sect. 4 by showing that reference-fixing causal connections do not bear any of the most commonly cited explanatory relations to successful action. In Sect. 5, I cast doubt on the claim that truth has causal powers, even as the causal correspondentists understand truth. An important theme in Sects. 4 and 1 See Field (1972) for a classic elaboration of this view. 2 There is one notion of “causal relevance” according which deflationism is compatible with the idea

that truth is causally relevant to successful action. On that notion, any predicate that figures in a causal explanation of any sort trivially picks out a “causally relevant property,” even if that property has no causal powers, has no essence to be understood in terms of causation, and is in fact a purely formal or Cambridge property. See Damnjanovic (2005) for details. That thin notion of causal relevance is not the one at play in the disagreement between deflationists and causal correspondence theorists.

123

Synthese (2011) 181:451–470

453

5 is that, according to causal correspondence theories, truth is a historical property of beliefs, but the success of our actions depends not on the history of our beliefs but on their intrinsic causal powers. Consequently, positing content-constituting causal connections does not enable causal correspondence theories to explain practical success any better than deflationary theories already can. Section 6 describes two ways in which we might explain systematic practical success without treating truth as a natural property. Before going further, I should mention three preliminary points. First, there are many different deflationary theories of truth, differing in how they describe the function of the truth predicate or concept, in whether they are pitched at the level of propositions, sentences, utterances, or something else, in whether the notion of truth admits of a formally explicit definition, and in several other ways. For the most part, these differences among deflationary theories will not be relevant to what I have to say here. I will assume that deflationists not only deny that truth is a natural property, but hold that (a) our understanding of truth consists in our mastery of the usage of the truth predicate, and (b) the truth predicate is principally a device for disquotation, generalization and re-expression. Claims such as ‘Jack’s beliefs about x are true’ thus amount to claims such as ‘If Jack believes that …x…, then …x…’. Most developed deflationary theories of truth include or presuppose enough of a theory of content to make sense of claims of the form, ‘p, according to representation R’. Different deflationisms might embrace theories that differ in their details. For the purposes of this paper, I take it for granted that some such theory is available to deflationists, though I will leave its precise details unspecified, apart from the fact that it does not treat content as a causal-historical property of representation tokens. Second, there are also a wide variety of causal correspondence theories of truth. At least as many such theories are possible as there are causal theories of reference, and there are many causal theories of reference. My concern here is exclusively with causal theories of reference that are not in the tradition of “success semantics” or some forms of teleosemantics. Success semantics and teleosemantics give the role of a representation in coordinating successful action a part in determining its reference. My concept, snow, for example, refers to snow in virtue of its distinctive role in helping my snow-regarding actions to succeed, a role that it and only it plays with respect to those and only those successes. A theory of truth built around such an account of reference might qualify as a “causal correspondence theory,” but it is not a theory that explains successful action by appeal to the truth of beliefs. To the contrary, it explains the truth of a belief by appeal to the success of actions. The causal correspondence theories of truth that concern me in this paper are backward-looking theories, which make the truth of a representation a function of the way that representation came into being. A fully developed causal correspondence theory will include a theory of content or of truth conditions, and different correspondence theories will differ in the details of those theories. For the purposes of this paper, I assume that such a theory is available to the causal correspondentist, and it has the resources to underwrite assignments of contents to representations. The details of how such a theory would work (apart from the fact that it would construe content as a causal-historical property of a representation) are beyond the scope of this paper.

123

454

Synthese (2011) 181:451–470

Finally, it is important to distinguish Kitcher’s criticism of deflationism from another, related criticism. Kitcher’s criticism is that deflationary theories of truth do not explain systematic success as well as causal correspondence theories do, and he thinks this is a reason to prefer the correspondence view. The other criticism is that deflationary theories must be false because (a) the truth of beliefs makes a causal contribution to the success of actions, (b) such a contribution is impossible if the truth predicate is merely a device for generalization, but (c) deflationism says the truth predicate is merely a device for generalization. My primary aim is to undermine the claim that causal correspondence theories do better at explaining success than deflationary theories, not to undermine the other objection to deflationism. In my view, if this latter objection works against deflationism, then it should work against causal correspondence theories as well. For, as I argue in Sect. 5, truth as construed by the causal correspondence theory is just as causally impotent as deflationary truth.3 2 Singular and systematic success Singular success occurs when a particular action has its intended outcome. Examples might include Jack’s success at finding Jill atop the hill at time t1 , Bill’s success at getting a beer by nodding to the bartender at time t2 , or Ophelia’s success in traveling from the castle to the brook at time t3 . Singular successfulness is a feature of dated, particular, token actions. Systematic success occurs when a family of particular actions, all of which depend on some common stock of beliefs, is such that its members tend to succeed. Examples might include Bill’s general success at getting the food and drinks he wants at bars and restaurants (by drawing on his stored information about how bars and restaurants work), Ophelia’s tendency to find her destinations around Elsinore (by drawing on her beliefs about the area’s geography), and chemical engineers’ success at inventing new compounds with desired properties (by applying various scientific theories, such as those encapsulated by the periodic table). An instance of systematic success is a collection of singular successes that result from actions that depend on a common stock of background beliefs. Kitcher criticizes deflationism for giving too-shallow explanations of singular success and no explanation at all for systematic success. I think both criticisms are mistaken. Nevertheless, some prominent deflationary treatments of truth and practical success have erred by overemphasizing explanations of singular success and misconstruing the relationship between those explanations and explanations of systematic success. Consider Paul Horwich’s account. He considers the hypothetical case in which Bill succeeds in getting a beer by nodding to the bartender, which is an instance of singular success. We might explain Bill’s success as follows: 3 Deflationism and causal correspondentism are not the only theories of truth on the market. There are also phenomenological theories, pluralist theories, identity theories, “indirect” correspondence theories, primitivist theories such as that of Davidson (1990), and all manner of relativisms. This paper, however, concerns only causal correspondentism and deflationism, and the question whether former provides better explanations of practical success than the latter.

123

Synthese (2011) 181:451–470

455

1. Bill wants a beer. 2. Bill believes he can get a beer by nodding to the bartender. 3. Bill is in a situation such that, if he wants a beer and he believes he can get one by nodding to the bartender, then he will nod to the bartender. 4. Bill’s belief that he can get a beer by nodding to the bartender is true. 5. So, Bill nods to the bartender. (from 1, 2, 3) 6. So, Bill gets a beer. (from 4, 5) Horwich points out that there is no need for more than a disquotational understanding of the mention of truth in 4. The explanation works just as well if we replace 4 with: 4*. Bill can get a beer by nodding to the bartender. So, there is no need to construe truth as a robust, causal-explanatory property in order to explain Bill’s success. Here we have an explanation of a singular success that appears to appeal to the truth of a belief about means and ends. Horwich shows that a deflationary view of truth is sufficient in this case. But what about systematic success? For example, what about the success of engineering endeavors employing the periodic table, or those that depend on the theory that nothing can travel faster than light? We want to explain these successes by appealing to the truth of the background theories, but those are not theories about means and ends. Instead, they are theories about such things as chemical valences and the relationships between mass and velocity. Horwich’s strategy is to generalize from the case of singular success explained by true means-ends beliefs to the case of (possibly systematic) success explained by true background beliefs: [Means-end] beliefs are more likely to be true if they are inferred from true premises; and very little of what we believe can be definitively excluded from the prospect of entering into such inferences as a premise. Therefore it is clear, in general, how true beliefs contribute to practical success. Nothing beyond the minimal theory is called for to explain this phenomenon. (Horwich 1998, p. 45) Suppose some set of actions (say, the actions of some engineers) are guided by the theory that nothing travels faster than light. We might say: The actions tend to succeed because the theory that guides them (i.e., the theory that nothing travels faster than light) is true. Horwich would have us understand this explanation as pointing to the fact that the actions are caused by means-ends beliefs, which are inferred from the theory that nothing travels faster than light. If nothing travels faster than light, and one infers that doing X will accomplish Y from the theory that nothing travels faster than light, then it is more likely that doing X will accomplish Y. And if it is likely that doing X will accomplish Y, and one does X in an effort to accomplish Y, one will likely succeed. The actions tend to succeed because each is likely to succeed; each is likely to succeed because it is guided by a means-end belief that is likely to be true; and the means-end

123

456

Synthese (2011) 181:451–470

belief is likely to be true because it is inferred from true background beliefs. None of that requires us to assume more than a deflationary notion of truth. Kitcher criticizes Horwich’s explanation for its “shallowness.” It does not explain why it is likely that doing X will accomplish Y if (a) one infers that doing X will accomplish Y from the theory that nothing travels faster than light and (b) nothing travels faster than light (Kitcher 2004, pp. 204–211). This is a criticism I reject below (see Sect. 6), but there are better reasons to be unsatisfied with Horwich’s approach. First, the approach is too strongly tied to a particular account of the mechanisms that connect belief to action. According to that account, our actions are caused by means-end beliefs, those beliefs are inferred from other beliefs, and the only influence our background beliefs have on our actions is mediated by the inference of meansend beliefs from them. This is a substantive, if sketchy, model of how our beliefs influence our actions. If the view turns out to be mistaken, it could mean trouble for Horwich-style explanations of systematic success. It only makes things worse that there is some reason to doubt that model. There is a growing body of evidence that many of our actions are guided directly by “background” beliefs, by perceptual beliefs, and by offline simulations.4 Even in cases in which such non-means-end beliefs influence our actions more indirectly, it is not at all clear that they do so by providing inputs to a process whereby means-end beliefs are inferred. Presumably, some cases of systematic success are explicable by the truth of non-means-end beliefs, even though those beliefs do not function as an agent’s premises in an inference whose conclusion is a true means-end belief. Taken strictly, Horwich’s proposal is incompatible with that possibility. A second problem with Horwich’s approach arises from an ambiguity in his claim that means-end beliefs are more likely to be true when they are inferred from true background beliefs. The claim could be disambiguated in either of these two ways: 7. Means-end beliefs that are inferred from true background beliefs are more likely to be true than to be false. 8. Means-end beliefs inferred from true background beliefs are more likely to be true than means-end beliefs inferred from false background beliefs. 7 is intuitively plausible. It asserts that the mechanisms whereby we infer our meansend beliefs are generally truth-preserving. But if 8 were false, then the truth of background beliefs would not explain systematic success, even given the truth of 7. Horwich offers no argument for 8, though, and 8 is far from obviously true. Though our inferential mechanisms might be truth-preserving, it does not follow that they are falsehood-preserving. And it might turn out that, given our mechanisms, the right false background belief could make it more likely that we infer true means-end beliefs than a true belief would (Stich 1990). The problem with Horwich’s proposal is not that it presupposes 8 and 8 is false, nor that it turns on an equivocation between 7 and 8. The problem is just that 8 might be false, and Horwich has provided us with no reason for thinking it isn’t. In the absence

4 See, for example, Grush (2004), Clark and Grush (1999), and Clark (2001, Chaps. 5 and 6).

123

Synthese (2011) 181:451–470

457

of support for 8, Horwich’s approach does not adequately explain systematic practical success. Deflationists have little trouble explaining singular successes in terms of true beliefs about means and ends. It is thus no surprise that they might try to piggyback explanations of systematic success on their explanations of singular successes. The trouble is that such an approach risks misconstruing the relationship between a person’s actions and her beliefs that do not concern means and ends, and it relies on the unsupported assumption that true background beliefs are more likely to lead to true means-ends beliefs than false background beliefs are. To avoid these problems, deflationists need to show how to explain successful actions in terms of true beliefs other than beliefs about means and ends. Then they would be able to explain not only systematic success, but those singular successes where one’s actions depend on what one believes but not on explicit beliefs about means and ends. Stephen Leeds (1995) does try to show how deflationists could give such explanations. When we attribute a person’s (singular or systematic) success to the truth of her beliefs, Leeds says, we are pointing out three things. First, she acted on the basis of her believing true action-guiding sentences. Second, she believed those sentences because she believed certain background sentences. Third, the action-guiding sentences were (disquotationally) true because the background sentences were (disquotationally) true. To explain truth by appeal to success, on this account, is to point out that the epistemic order among an agent’s beliefs recapitulates the explanatory order among their contents. Moreover, Leeds contends, this recapitulation is neither surprising nor in need of explanation in terms of a causal correspondence theory of truth. Instead, it is only to be expected given that principles of charity and rationality constrain us to interpret others in ways that maximize the similarity between the inferential relations among their beliefs and the explanatory relations among their beliefs’ contents. (For reasons we need not go into here, Leeds takes belief to be a relation between subjects and sentences in an interpreter’s language.) Leeds’ approach is not without problems, though. Like Horwich’s, it presupposes a model of action according to which some beliefs are “action-guiding,” and those action-guiding beliefs are conclusions of inferences whose premises are more theoretical, background beliefs. If that model is untrue to the psychodynamics of action, then Leeds’ suggestion is of no help to deflationists. A more serious problem for Leeds is that, also like Horwich, he proceeds as though the problem has been solved once we have shown how to account for the role of true background beliefs in singular success. To explain systematic success, on Leeds’ approach, we just point out that some smallish number of background beliefs are implicated in the explanations of a largish number of singular successes. Explaining a case of systematic success only by explaining its constituent singular successes, though, might not be very satisfying. It would be like “explaining” why the ratio of males to females in a population is m:f by explaining why each of the individual males is male and why each of the individual females is female. Such an explanation would seem to miss the point of the question, “Why is the ratio of males to females in this population m:f ?” Similarly, the conjunction of the explanations of a collection of singular successes might not explain their systematicity. When we aim to explain a case of systematic success, we might want to look beyond the success of

123

458

Synthese (2011) 181:451–470

her particular actions and explain the pattern of success her actions appear to exhibit. We might want to explain the apparent regularity that she tends to succeed when her actions are of a certain type. Given the regularity, we might predict that her actions would succeed if they were of that type or that they will succeed in the future when they are of that type. The conjunction of explanations of her singular successes sheds no light on why that would be so. The most obvious move an advocate of Leeds’ proposal might make is to say that the counterfactual and future success are neither surprising nor in need of explanation, given the roles that the principles of charity and rationality play in attributing beliefs to others. We must attribute mental contents in accord with those principles, and, consequently we are bound to interpret the agent counterfactually or in the future so that her behavior exhibits patterns of systematic success. She will or would succeed systematically because we will or would interpret her in ways that make her behavior count as successful. This move is unconvincing, though, because it answers the wrong question. It tells us why, given the agent’s future or counterfactual behavior, that behavior will or would include patterns of systematic success. (Answer: Because we are bound by the principles of charity and rationality to attribute mental contents that make the behavior include such patterns.) It does not tell us why, given that the agent’s beliefs have certain contents, whatever behavior those beliefs guide will or would include patterns of systematic success. That is the question we must answer to explain the apparently projectable regularity that the agent succeeds when she acts on beliefs with certain contents. None of this is to say that things are hopeless for an approach generally in keeping with Leeds’. The problems I have mentioned might be solvable. My point is not that Leeds or Horwich is wrong. My point is that they leave some of the important explanatory work undone by trying to build an explanation of systematically successful action out of explanations of singular success. The case of systematic success is importantly different from the case of singular success, and it is in deflationists’ interest to address it on its own terms. I return to the question of how deflationists might account for systematic practical success in Sect. 6. First, though, I examine Kitcher’s case for thinking the causal correspondence view provides better explanations than those that are available to deflationists.

3 Kitcher’s argument Kitcher stresses the importance of explaining systematic success to the debate between deflationism and causal correspondentism. Even if deflationists could adequately explain every instance of singular success by appeal to the (disquotational) truth of means-end beliefs, their defense would completely miss the point of the causal correspondence criticism: [We] should ask exactly what has been explained. The answer, surely, is that if an agent has a true belief about means-ends relations, then that agent is likely

123

Synthese (2011) 181:451–470

459

to be successful. That isn’t quite the explanandum that realists have taken to be crucial in their defenses of correspondence truth. Correspondence truth has been supposed to be necessary because of the way in which true beliefs about means-ends relations, or behavior that is as if the agent has true beliefs about means-ends relations, result from true beliefs about the objects that figure in the desired goal-state. (Kitcher 2004, p. 204) Kitcher goes on (rightly, in my view) to diagnose the deflationists’ failure to address the right explanandum as a consequence of restricting their attention to singular, rather than systematic, practical success. To remedy the failure, Kitcher takes as his paradigm not a case in which a single action, based on a single, straightforward means-end belief, has its desired outcome, but a case in which a person relies on an extensive system of beliefs to succeed in a variety of tasks on a variety of occasions. The hypothetical case he asks us to consider is that of Ophelia, who has a map of Elsinore and relies on it to find her way around. Ophelia almost never gets lost, and she almost always reaches her desired destinations with little trouble. A natural explanation is that Ophelia succeeds (systematically) because her map is accurate. To avoid complications concerning map reading, I will suppose Ophelia has memorized her map. She thus has a cognitive model of Elsinore’s geography, which I will treat as comprising a certain large set of individual beliefs, whose contents are a certain large set of propositions. The more of those beliefs are true, the more accurate Ophelia’s cognitive model of Elsinore is. I will refer to Ophelia’s cognitive model as M. Suppose Ophelia tends to succeed in her navigations in Elsinore because M is accurate. The deflationary gloss of this explanation is: Ophelia tends to succeed in her navigations because she tends to think things in Elsinore are where they are.5 The causal correspondence version is: Ophelia tends to succeed in her navigations because they depend on M, and M has a certain causal connection to the world (namely, the complex causal relation that causal correspondence theorists identify with the substantive property of being true). Kitcher thinks the deflationary explanation is inferior to the correspondence explanation. The point of both explanations is that Ophelia’s systematic success can be understood by the applicability of this explanatory schema to her successful navigations (Kitcher 2004, pp. 206–208): 9. Ophelia’s cognitive model of Elsinore is M. 10. Ophelia desires that q. (Possible replacements for q include ‘Ophelia reaches the brook’, ‘Ophelia reaches the chapel’, ‘Ophelia reaches the cliffs’, ‘Ophelia reaches the graveyard’, etc.) 5 That is, she tends to succeed because, ordinarily, if x is at y according to M, then x is at y.

123

460

Synthese (2011) 181:451–470

11. If Ophelia’s cognitive model of Elsinore is M and she desires that q, then she does A(q). 12. If M is accurate and Ophelia does A(q), then q. 13. M is accurate. 14. So, Ophelia does A(q). (From 5, 6, 7) 15. So, q. (From 8, 9, 10) The difference between deflationism and the correspondence theory lies in how each construes 13. Deflationists read it as an indirect way of asserting the contents of M. Thus it means something like, “If Ophelia thinks x is at y in Elsinore, then x is at y or not far off.” Causal correspondentists see it as attributing a certain substantive causal relation between M and the objects and properties M refers to. The problem Kitcher sees for deflationism concerns how we make sense of 11 and 12, given the deflationary rendering of 13. Both mention an action, A(q), which Kitcher calls “the action pertinent to q.” The trouble is that there are two different notions of pertinence in play (Kitcher 2004, p. 208). Let us call an action psychologically pertinent to q (relative to Ophelia and M) if and only if it is what Ophelia would do if she desired q and M were her cognitive model of Elsinore. Call an action effectively pertinent to q (relative to Ophelia and M) if Ophelia’s doing it would bring it about that q, if M were accurate. We might define A(q) as the action that is psychologically pertinent to q. In that case, 11 is analytic, but we need an explanation for 12. Alternatively, we might define A(q) as whatever is effectively pertinent to q. In that case, 12 is analytic, but we need an explanation for 11. Either way, Kitcher thinks, deflationists are stuck with an explanatory mystery: Why do psychological and effective pertinence coincide? Here is how he puts the point: Deflationists might think that the schema of explanation [9–15] will go through, even if the notion of truth is interpreted as they prefer. But this fails to address the explanatory mystery. If we define the notion of “pertinence” so that [11] is automatically true, then we need grounds for [12] and conversely. To read [13] in the deflationist’s sense leaves at least one of [11], [12] unexplained. (Kitcher 2004, p. 208) While deflationists are stuck with an explanatory mystery, Kitcher thinks causal correspondentists are not. The causal relation between M and Elsinore is such that M corresponds to the geography of Elsinore, and that correspondence is responsible for the fact that what is psychologically pertinent to M is also effectively pertinent. It explains why actions that fit M psychologically also fit Elsinore practically. Here is how Kitcher makes his point: The parallel between the psychological life of the subject and the effectiveness of the action comes about because of the correspondence between elements in the subject’s representations and elements in the world. Ophelia’s decisions to orient her body in particular ways stem from [her application of M] in light of her preferred destinations. A full psychological understanding of those decisions must probe the ways in which the semantical relations between her tokens and the

123

Synthese (2011) 181:451–470

461

objects in her surroundings are akin to the causal relations between her perceptual states and those objects. Those decisions turn out to be effective … because the causal relations that connect Ophelia’s tokens to entities in the world pick out entities that stand in the spatial relations represented in Ophelia’s thoughts, the very spatial relations represented in [M]. (2004, p. 209) As I see it, Kitcher is making three claims here. First, psychological and effective pertinence coincide because of the correspondence between M and Elsinore. Second, Ophelia’s routes depend on the contents of her beliefs about Elsinore and where she wants to go. Third, Ophelia’s routes succeed because things in Elsinore are where she thinks they are. It is unclear to me whether Kitcher means these points to be independent of one another or, instead, he means the latter two points to support the first one. Because the second and third points seem not to give any particular support to the first, whatever Kitcher might intend, I will address each point independently in the next section. 4 Correspondence and explanation Kitcher’s first point is that the correspondence relation between M and Elsinore explains why psychological and effective pertinence coincide. Even if this claim is true, we must consider whether it is true in a sense that is both unavailable given a deflationary notion of truth and available given a causal correspondence view of truth. Otherwise, it would not give us a reason to favor causal correspondentism over deflationism. This is important because there is a deflationary notion of correspondence available. In this sense, for M to “correspond” to Elsinore is just for things in Elsinore to be where M says they are. The correspondence between M and Elsinore is just the correlation (in the statistical sense) between where M says things are and where they are. Note that it is perfectly in order for deflationists to talk about things being where M says they are. Doing so does not require invoking a substantive concept of truth.6 Kitcher cannot have this sense of correspondence in mind when he says that correspondence explains the coincidence of psychological and effective pertinence. Rather, he must have in mind a notion of correspondence that is unavailable to deflationists and available to causal correspondentists. He must have in mind the causal relation in virtue of which M refers to the objects, properties, and relations that it refers to. That is the relation causal correspondentists posit but deflationists do not, and that is the relation causal correspondentists identify as part of the essence of truth. Let us call that relation M’s content-constituting causal connection. For Kitcher’s first point to tell in favor of causal correspondentism, as against deflationism, M’s content-constituting causal connection must explain why psychological and effective pertinence coincide. There is good reason to think it does not. 6 As an anonymous reviewer has noted, it might be possible to formulate a formally explicit definition of truth along the lines of “R is true if and only if the world is the way it is according to R.” That would be consistent with the basic deflationist position, which denies that truth is a natural property with a causal essence and is neutral with respect to the possibility of explicitly defining truth.

123

462

Synthese (2011) 181:451–470

For one thing to explain another, it must bear some explanatory relation to the other. I do not know what all possible explanatory relations are. Nevertheless, if it can be shown that (a) x is not necessary for y, (b) x is not sufficient for y, (c) x is not an insufficient but necessary part of an unnecessary but (minimal) sufficient condition for y [i.e., an “INUS” condition, see Mackie (1974)], and (d) x does not make y more likely, then we have very good reason to deny that x explains y. M’s content-constituting causal connection bears none of those relations to Ophelia’s success, and so we should doubt that it explains her success. To see that the causal connection is not sufficient for Ophelia’s success, remember that it suffices only to fix the content of M. It is possible for M to have the very same content-constituting causal connection to Elsinore, and so to have the very same content, even if the correlation between where things are and where M says they are is very weak. In that case, Ophelia would not systematically succeed in finding her destinations by relying on M; she might even systematically fail. Because M could have the very same content-constituting connections in cases where Ophelia fails systematically, those connections are not sufficient for her systematic success. Similar considerations show that the content-constituting connection does not make Ophelia’s success more likely, either. The mere existence of a representation with the content that p has no influence on the likelihood that p. Many American girls enjoy drawing pictures of unicorns, but the unicorn-representations they create do not make it any more likely that unicorns exist. The fact that M is connected to the world so as to have the content that the cemetery is beside the brook does not alter the likelihood that the cemetery is beside the brook one bit. So, if Ophelia’s success requires the cemetery to be beside the brook, M’s content-constituting causal connection does nothing to make her success more likely. We can make the point somewhat more directly. For every case in which Ophelia relies on M and succeeds, there is a possible case in which M has the very same content-constituting connection, Ophelia relies on M, and yet Ophelia fails (because Elsinore is not arranged as M says it is). Since we can hold the content-constituting connection constant and find a possible failure for every possible success, the content-constituting connection does not make success more likely.7 To see that the content-constituting connection is not necessary for Ophelia’s success, we can consider an imaginary scenario of the following sort: Ophelia is struck by lightning, and the lightning strike obliterates the part of her brain that constitutes M. Immediately, though, the swamp gases ignite so that a perfect intrinsic duplicate of M forms spontaneously and replaces M. Call the duplicate M*, and call Ophelia “Swamp Ophelia” after the transformation. M∗ has the exact same influence on Swamp Ophelia’s behavior that M has on Ophelia’s behavior. In particular, M∗ has the same influence on Swamp Ophelia’s route 7 This point can be made more precisely: When there are no more cases in which P&Q than cases in which P&∼Q, Pr (Q|P) can be no greater than Pr (∼Q|P). There are no more cases in which M has a given contentconstituting connection and Ophelia succeeds, than there are cases in which M has that content-constituting connection and Ophelia fails. So, the probability that Ophelia succeeds, given M’s content-constituting connection, is no greater than the probability that she fails, given M’s content-constituting connection. This follows from an interpretation of probabilities as relative frequencies among possibilities. It also follows so long as we assume that, whatever probabilities are, relative frequencies track them.

123

Synthese (2011) 181:451–470

463

selections. So, Swamp Ophelia is exactly as successful at reaching her destinations as Ophelia is at reaching hers. But M∗ lacks M’s content-constituting causal connection. Indeed, M∗ has no content-constituting connections at all. So, those connections are not a necessary condition for Ophelia’s success; she could succeed without them. A causal correspondentist might contend that M∗ has no content. On that basis, she might reply that Swamp Ophelia’s movements through Elsinore do not qualify as actions, and so they are not successful actions either. Consequently, Swamp Ophelia does not succeed after all.8 Such a reply is of no help. It requires either a too-narrow conception of action or a too-narrow conception of success. When Swamp Ophelia wants to go to the cemetery, she plots a course through Elsinore and follows that course. She arrives at the cemetery. Because M∗ is intrinsically the same as M, everything seems to Swamp Ophelia just as it seems to Ophelia. She wants to go to a certain place, and she deploys M∗ to guide her bodily movements in trying to get there. It is a Procrustean notion of action that treats Swamp Ophelia’s endogenously motivated and controlled movements through Elsinore as things that happen to her rather than things she does.9 On the other hand, even if we do refuse to call her movements actions, we cannot deny that they systematically culminate in the satisfaction of her desires. Is the systematic satisfaction of Swamp Ophelia’s desires an example of success? On the restrictive conceptions of success and of action, the answer is no, but it is not an example of failure either. Rather, what happens to Swamp Ophelia is outside the logical space of success and failure altogether, and it is a category mistake to call it either. So, if we grant the restrictive conception of action or of success, M’s contentconstituting connection is a necessary condition for Ophelia’s success, but not the sort that would explain why she succeeds rather than fails. Causal correspondentists, however, need the connection to explain why Ophelia’s actions systematically succeed rather than fail, not why her movements fall under the label ‘action’. The key point here is that M and M∗ play precisely the same role, in precisely the same way, in causing the systematic satisfaction of Ophelia’s desires, which is what we want to explain when we explain her success. M∗ does not have M’s content-constituting connection, so the connection is not a necessary condition that causes Ophelia’s systematic success.10 Another objection might be that the Swamp Ophelia case is too fanciful. Sure, M’s content-constituting connection isn’t logically necessary for Ophelia’s success, as the case shows, but it might be nomologically necessary. It might be a matter of natural law that Ophelia could not succeed as she does without a mental representation that has M’s precise content-constituting causal connection to the geography of Elsinore. Such an objection would not work, however. If we suppose that M’s content-constituting connections are necessary for Ophelia’s success, we should ask why that is so. There are two possibilities. Either they are necessary because, as a matter of 8 I thank an anonymous reviewer for suggesting I consider this line of argument. 9 An anonymous reviewer has suggested our actions might be grounded in the phenomenology of our mental states rather than their causal histories. In that case, because M and M∗ are phenomenologically indistinguishable, the bodily movements M∗ guides would qualify as actions and qualify as successful. 10 I thank an anonymous referee for encouraging me to put the point in this way.

123

464

Synthese (2011) 181:451–470

fact, those connections are the only nomologically possible way to bring something with M’s intrinsic features into the world (so an intrinsic duplicate with different or no content is nomologically impossible), or they are necessary because M’s intrinsic properties, the layout of Elsinore, and the rest of Ophelia’s mental states do not suffice for her to succeed, but they do suffice if we add M’s content-constituting causal connection into the mix. The first possibility is irrelevant. First, it is doubtful that a state with M’s intrinsic properties could come into being only by having M’s content-constituting connection to the world, even when we restrict our attention to nomological possibility. Second, even if there is only one nomologically possible way to bring something with M’s intrinsic features into being, it is doubtful that Ophelia needs such a state to succeed anyway. Sometimes false beliefs lead to practical success, and sometimes false beliefs can help us in cases when true beliefs would not (see, e.g., Stich 1990). Third, and perhaps most important, we must be careful about what we treat as causally explanatory of what. In virtue of its intrinsic features, M interacts with Ophelia’s other mental states to cause certain behavior. The precise nature of that behavior depends on M’s intrinsic features. Suppose it is nomologically impossible for a token with M’s intrinsic features to exist without M’s content-constituting causal connections. As it happens, anything with those intrinsic features will have a causal history of a certain sort; that is just the only way to get something with those features into the world. That still does not give causal correspondentists what they need, for what matters in this case is not that those causal-historical properties bestow a certain content on M, but that they cause something with M’s intrinsic features to exist. If the contentconstituting connections are necessary for Ophelia’s success in this case, they are still not necessary qua content-constituting connections, and that is what correspondentists would need for the connections to carry explanatory weight. The very same argument applies if we turn to an even weaker notion than nomological necessity, counterfactual dependence. One might claim that M’s content-constituting connection explains Ophelia’s systematic success because, if Ophelia had not had a token with that content-constituting connection, she would not have succeeded (even though, perhaps, she could have succeeded). Let us suppose, for the sake of argument, that Ophelia would not have succeeded without a token with M’s content-constituting connection. Still we must ask why that is so. There are two candidate explanations. One is that, if Ophelia did not have a token with M’s content-constituting connection, she would not have a token with M’s content. The other is that, if she did not have a token with M’s content-constituting connection, she would not have a token with M’s intrinsic causal powers. So long as Ophelia has a token with M’s intrinsic causal powers, though, she will enjoy systematic navigational success, regardless of that token’s content properties. So, if Ophelia would not succeed without a token with M’s content-constituting connection, it would have to be because such a token would lack M’s intrinsic causal powers, not because it would lack M’s content. But then her success does not counterfactually depend on the content-constituting connection qua content-constituting connection, and so the connection does not bear the explanatory weight causal correspondentists would need it to.

123

Synthese (2011) 181:451–470

465

There is also reason to deny that Ophelia’s success counterfactually depends on M’s content-constituting connection. Suppose Ophelia did not have a token with M’s content-constituting connection. What would the world be like? How one answers depends on how one interprets counterfactual conditionals. We might treat ‘If Ophelia had not had a token with M’s content-constituting connection, she would not have succeeded’ as a “backtracking” conditional (Lewis 1979). In that case, we would have to consider what in Ophelia’s past would have been different to lead to her not having a token with M’s content-constituting connection, and we would have to ask whether such a different past would undermine Ophelia’s later success. One possibility is that Elsinore’s geography was the same but Ophelia wound up with various mistaken beliefs about where things in Elsinore were. Other possibilities also exist, though. For example, if Fr. Jenssen rather than Fr. Nielssen had consecrated the cemetery, then Ophelia would not have had a token with M’s content constituting connection, and yet she still might have enjoyed systematic success. Also, so long as we allow backtracking, we should take into account the possibility that Ophelia winds up with a token whose content-constituting connection is different because Elsinore is different, and Ophelia still succeeds systematically in getting where she wants to go. It is not clear what to say about whether Ophelia would have succeeded without a token that has M’s content-constituting connection, if we allow backtracking. If we do not allow backtracking, though, we simply try to hold as much constant as we can, while deleting M’s content-constituting connection from the world. We use, as Bennett (2003) has put it, “metaphysical hole punchers” to eliminate the connection while leaving everything else the same. In particular, we leave the arrangement of Elsinore the same, and we leave M’s intrinsic features the same. M no longer has its content, but it still has its intrinsic causal powers, and those are enough (given the arrangement of Elsinore) to get Ophelia where she wants to go. If M did not have its content-constituting connection, Ophelia would still succeed. So, her success does not depend counterfactually on M’s content-constituting connection. Ophelia’s success does not require M’s content-constituting connection, logically, nomologically, or merely counterfactually. And the connection neither suffices for her success nor makes her success more probable. If the connection explains her success, then, we are left with only the possibility that the content-constituting connection is an INUS condition for Ophelia’s success—an insufficient but necessary part of an unnecessary, (minimal) sufficient condition for her success. (A sufficient condition for something is “minimal” when none of its proper parts are sufficient for it. Without the restriction to minimal sufficient conditions, everything turns out to be an INUS condition of everything logically independent of it.11 ) M’s content-constituting causal connection is not an explanatory INUS condition for Ophelia’s success. Suppose it were. Then it would be a necessary part of some condition S, such that S is sufficient for Ophelia’s success, and no proper part of S is sufficient for Ophelia’s success. But, as the Swamp Ophelia case shows, we can hold 11 Let P and Q be logically independent of one another. Then P&Q is an unnecessary sufficient condition for P, and Q is an insufficient necessary condition for P&Q. So, Q is, in this trivial way, an INUS condition for P.

123

466

Synthese (2011) 181:451–470

S’s intrinsic features and causal powers constant while varying M’s content-constituting causal connection (or removing it entirely). So, either the content-constituting connection isn’t necessary for S or S isn’t a minimal sufficient condition for Ophelia’s success. Either way, the content-constituting connection is not an explanatory INUS condition for Ophelia’s success. Kitcher claims that psychological and effective pertinence coincide because of M’s “correspondence” to Elsinore. If he means correspondence in the sense that is available to deflationists, then this point does not count against their view. On the other hand, if by ‘correspondence’ Kitcher means M’s content-constituting causal connection, then what he says looks false. Those connections appear to bear no explanatory relation to her success. The other two points Kitcher raises are that M is causally relevant to the actions Ophelia takes and that the actions succeed because things in Elsinore tend to be where M says they are. But deflationists are not in the business of denying the causal efficacy of mental representations, nor are they in the business of denying that representations with different contents are apt to cause different actions. Indeed, the canonical deflationary explanation of Ophelia’s tendency to succeed is that she succeeds because things are (or tend to be) where she thinks they are. So, deflationists have nothing to fear from these other points. None of this is to say that the deflationist really can resolve the mystery and explain why psychological and effective pertinence coincide. For now, my only point is that causal correspondentists are in at least as much trouble as deflationists. The claims Kitcher makes in favor of correspondentism do not really support that view over deflationism. In Sect. 6, I consider what conclusions we ought to draw from all this. But first, there is a more general argument against correspondence explanations of success that I would like to make.

5 A more general argument Let it be granted that deflationists owe an account of why psychological and effective pertinence should ever coincide. The more general argument I offer in this section aims to show that causal correspondentists have the same, undischarged debt. The argument depends on a substantive but widely accepted metaphysical assumption. According to that assumption, genuine properties are individuated by the causal powers they bestow on their bearers. Objects cannot differ in their causal powers without differing in their properties, and they cannot differ in their properties without differing in their causal powers. Causal powers are dispositions. As such, they are intrinsic to the objects that have them, and they need not all be manifested all the time. A brick has the power to break a vase of a certain kind, but that does not mean that it ever actually does break the vase. Its power to break such a vase (in the right circumstances) is a standing, intrinsic feature of the brick, which can exist even if there are no vases and even if the right circumstances do not ever obtain. It is not that the brick, upon striking a vase with a certain force, suddenly acquires the power to break the vase just in time to break it.

123

Synthese (2011) 181:451–470

467

Rather, the brick has the power to break the vase all along, and that power manifests itself when the brick hits the vase. Truth, by this standard, is not a genuine property.12 Imagine two versions of Ophelia, True-Ophelia and False-Ophelia, who are intrinsic duplicates of one another. TrueOphelia lives in a world where her token of M is accurate. False-Ophelia’s token is inaccurate. The accuracy of the tokens makes no difference to their causal powers. It matters only to how the powers are manifested. A hammer in a world where all vases are made of steel and a microphysical duplicate in a world where all vases are made of crystal do not differ in their causal powers. They differ only in how their causal powers are manifested and in their opportunities to manifest them. Now, we might still want to explain Ophelia’s success by pointing to the truth of her beliefs (or the accuracy of M). In so doing, we would be pointing simultaneously to some of the circumstances in which her causal powers are manifested (e.g., the cemetery’s being by the brook) and to some of her causal powers (e.g., those involved in thinking the cemetery is by the brook). Such a move is, surprisingly enough, off limits on the causal correspondence view of truth. The reason is that the causal correspondence view makes a token’s causal history, rather than its intrinsic causal powers, constitutive of its content. (Remember, “success semantics” and “teleosemantics” are out of the running as bases for a causal theory of truth; they explain truth by appeal to success, rather than explaining success by appeal to truth.) But a token’s causal history is extrinsic to it, and so having a given causal history is not a genuine property. Causal history does not bestow causal powers on a token, it just tells how the token came to be. So, on the causal correspondence view, to explain Ophelia’s success by appeal to M’s accuracy is to explain it in terms of M’s causal antecedents and the geography of Elsinore. Such an explanation leaves out what is explanatorily crucial: the causal powers of M that are involved in the production of effectively pertinent actions. The causal correspondence view has no more to say about psychological pertinence than the deflationary view has. It has nothing to say about the causal powers of mental representation tokens, and so it has nothing to say about why what is psychologically pertinent to a given end (relative to a given agent and belief) would also be effectively pertinent (relative to that agent and belief). The causal correspondence view does not resolve Kitcher’s explanatory mystery. 6 Everybody’s mystery I have been arguing that explaining systematic success is at least as much of a problem for causal correspondence theories of truth as it is for deflationary theories. This is because causal correspondence theories are not better suited to solving Kitcher’s 12 Some philosophers think this test is too strict; it rules that all manner of logical and mathematical properties are not “genuine.” What matters for my purposes, though, is less the claim that all properties are individuated by the powers they bestow than the claim that there is a class of properties (the “natural” properties, perhaps) that are individuated in that way, and that truth isn’t one because truth does not bestow causal powers on representations. As I argue below, causal-historical properties do not bestow causal powers either, and that makes them bad candidates to qualify as “causal-explanatory properties.”

123

468

Synthese (2011) 181:451–470

“explanatory mystery” than deflationary theories. I now take up the task of explaining why it is deflationism that has the advantage. Kitcher’s explanatory mystery is everybody’s mystery, but everybody’s mystery is still everybody’s mystery. If neither the causal correspondence view nor deflationism is well positioned to solve the mystery, then one might think that counts against both views, and that some other view of truth—pragmatism, coherentism, or functionalism,13 perhaps—is called for. In my view, that would be a mistake. It would be a mistake because Kitcher’s explanatory mystery has nothing to do with truth. It is the mystery of why it should often be the case that one and the same action satisfies the following two descriptions: a. It is what a given agent with a given belief, whose content is that p, would do in pursuit of some end, q. b. It is what would bring it about that q if a given agent performed it and p. This is a mystery to be addressed by theories in the philosophy of mind, not by theories of truth. At least two views in the philosophy of mind are readily available to explain why psychological and effective pertinence so often coincide. One of those views concerns the nature of belief. According to it, what makes an attitude with the content that p a belief is that it is disposed to cause actions that would succeed if p. It is not that those dispositions make it into a belief with that content, for various attitudes with the same content (such as desires and hopes) do not have those dispositions. Rather, those are the sorts of dispositions that constitute the beliefhood of an attitude with a given content. The view is not implausible. Suppose that Bill has an attitude X with the propositional content that there is beer in the refrigerator, yet it is not the case that the actions that attitude tends to cause are actions that would tend to succeed if there were beer in the refrigerator. This seems to be very powerful prima facie evidence that attitude X is not the attitude of belief. Maybe it is hope or desire, but it isn’t belief. And if we suppose that attitude X is belief, then we owe a story about what other factors are in the picture that override Bill’s disposition to take actions that would succeed if there were beer in the refrigerator. If this view of the nature of belief is correct, the “explanatory mystery” is pretty easy to solve. Psychological and effective pertinence coincide because, by its nature, belief is an attitude that causes effectively pertinent actions. An attitude that does not do that is not belief. And, because deflationists are not committed to denying that belief is a genuine property, this is an explanation to which they and causal correspondentists can both help themselves. Some philosophers will object to this view of the nature of belief. They might be more persuaded by a second sort of view, which does not suppose that belief is essentially an attitude that causes effectively pertinent actions. The second view is generally adaptationist and naturalistic in spirit. Our psychological mechanisms are the products of natural selection. Those that take beliefs and desires as inputs and produce actions as outputs can do better or 13 For functionalism, see Lynch (2004).

123

Synthese (2011) 181:451–470

469

worse at producing effectively pertinent outputs. The worse they do, the worse job a critter with such mechanisms will do at satisfying its desires. Critters that do badly enough will display what Quine calls the “pathetic but praiseworthy tendency to die before reproducing their kind.” Natural selection will favor mechanisms that are better at making psychological and effective pertinence coincide. The coincidence does not arise because of the nature of truth, but because of the details of our cognitive mechanisms and their phylogeny. This explanation is also available to deflationists.14 There are probably other ways we might resolve the explanatory mystery in the philosophy of mind.15 Let us call all such explanations “PM explanations.” (‘PM’ stands for “Philosophy of Mind.”) Deflationists could use PM explanations that do not presuppose inflationary theories of truth to answer Kitcher’s challenge, but so too could causal correspondentists. Nevertheless, I think deflationism has the advantage for two reasons. First, if the dispute is to be between deflationism plus a PM explanation and a causal correspondence theory plus a PM explanation, the causal correspondence theorist owes a justification for preferring the logically stronger theory. What reason is there for thinking that truth is a genuine property? By employing a PM explanation, the correspondence theorist cannot answer that we must treat truth as a genuine property to explain practical success. So the additional logical strength of correspondentism is, for the moment, anyway, unmotivated. Second, if the dispute is between deflationism plus a PM explanation and a causal correspondence theory without a PM explanation, such as what Kitcher recommends, then the deflationist’s explanation of systematic success is just better. The correspondence explanation appeals to various extraneous factors—particularly the causal history of mental representation tokens—that fail to produce deeper or more powerful explanations than deflationism provides before adding a PM explanation. Adding the PM explanation to a deflationary view of truth then allows one to answer the correspondence challenge directly, giving precisely the deeper and more powerful explanation the challenge calls for, without having to compromise the view the truth is not a genuine property. 7 Conclusion Kitcher’s argument for the superiority of causal correspondence theories to deflationary theories of truth depends on the claim that the former, but not the latter, can explain 14 Of course, it is available only if the explanation does not ultimately depend on the view that systems where the pertinences coincide were selected for in virtue of causal correspondence to the world. But such a view is unnecessary. Critters that do what is effectively pertinent have a fitness advantage over critters that do not; they are selected for doing what is effectively pertinent. But critters will do what is psychologically pertinent, by definition. That means selection for doing what is effectively pertinent is automatically selection for the coincidence of psychological and effective pertinence, independent of any putative content-constituting causal connections. 15 One might even take Leeds’ (1995) proposal as a PM explanation. If Leeds’ point is that psychological and effective pertinence coincide because (a) the contents of our mental states are interpretation-dependent and (b) the interpretation we standardly use assigns contents in such a way that the two sorts of pertinence coincide by design, then once more the solution to Kitcher’s mystery is not to be found in the nature of truth but in the nature of our mental states and our practices of mentalistic explanation.

123

470

Synthese (2011) 181:451–470

the coincidence of psychological and effective pertinence. Consideration of Kitcher’s “explanatory mystery,” though, shows that it is no less a mystery for causal correspondentists than for deflationists. Neither theory of truth offers a clear solution to the mystery, but we can expect a solution to come from a theory of mind rather than a theory of truth. If the adoption of the logically stronger, correspondence view of truth is to be justified, it will have to be on other grounds. Acknowledgements This paper arose from work begun at John Heil’s 2006 Mind & Metaphysics NEH Summer Seminar. I owe thanks to the National Endowment for the Humanities, John Heil, the participants in that seminar, and several anonymous referees. Special thanks are due to Torin Alter, Adam Kovach, Michael Lynch, Ted Parent, Richard Richards, Stuart Rachels, and Michelle Wrenn.

References Bennett, K. (2003). Why the exclusion problem seems intractable, how, just maybe, to tract it. Nous, 37, 471–497. Clark, A. (2001). Mindware: An introduction the philosophy of cognitive science. Oxford: Oxford University Press. Clark, A., & Grush, R. (1999). Towards a cognitive robotics. Adaptive Behavior, 7(1), 5–16. Damnjanovic, N. (2005). Deflationism and the success argument. The Philosophical Quarterly, 55(218), 54–67. Davidson, D. (1990). The structure and content of truth. Journal of Philosophy, 87(6), 279–328. Field, H. (1972). Tarski’s theory of truth. The Journal of Philosophy, 69(13), 347–375. Grush, R. (2004). The emulation theory of representation: Motor control, imagery, and perception. Behavior and Brain Sciences, 27, 377–396. Horwich, P. (1998). Truth (2nd ed.). Oxford: Oxford University Press. Kitcher, P. (2004). On the explanatory role of correspondence truth. In F. F. Schmitt (Ed.), Theories of truth (pp. 197–215). Malden, MA: Blackwell. Leeds, S. (1995). Truth, correspondence, and success. Philosophical Studies, 79, 1–36. Lewis, D. (1979). Counterfactual dependence and time’s arrow. Nous, 13, 455–476. Lynch, M. (2004). Truth and multiple realizability. Australasian Journal of Philosophy, 82(3), 384–408. Mackie, J. (1974). The cement of the universe. Oxford: Clarendon Press. Stich, S. (1990). The fragmentation of reason: Preface to a pragmatic theory of cognitive evaluation. Cambridge, MA: MIT Press.

123

Synthese (2011) 181:471–488 DOI 10.1007/s11229-010-9742-2

Two claims about epistemic propriety E. J. Coffman

Received: 13 July 2009 / Accepted: 8 March 2010 / Published online: 27 March 2010 © Springer Science+Business Media B.V. 2010

Abstract This paper has two main parts. In the first part, I argue that prominent moves in two related current debates in epistemology—viz., the debates over classical invariantism and the knowledge first movement—depend on one or the other of two claims about epistemic propriety: (1) Impropriety due to lack of a particular epistemic feature suffices for epistemic impropriety; and (2) Having justification to believe P suffices for having warrant to assert P. In the second part, I present and defend novel arguments against both claims. Keywords

Epistemic justification · Warrant · Knowledge · Belief · Assertion

Consider two claims about epistemic propriety: Defect Implies Defective (DID): If an epistemically evaluable item (e.g., a belief) is in some way improper because it lacks a particular epistemic feature, then the item is epistemically improper. In a slogan: “Impropriety due to epistemic defect implies epistemically defective.” Justification Secures Warrant (JSW): If you’re positioned to hold an epistemically proper (‘justified’) belief that P, then you’re positioned to make an epistemically proper (‘warranted’) assertion that P.1 In a slogan: “Justification to believe secures warrant to assert.”

1 My usage of ‘epistemically proper’ and related terms will follow that of (e.g.) Pryor (2005). On this

usage, an item is epistemically proper only if it has a more objective kind of positive epistemic status than is entailed by a subject’s simply being intellectually blameless relative to the item. In using ‘justified belief’ and ‘warranted assertion’ to abbreviate (respectively) ‘epistemically proper belief’ and ‘epistemically proper assertion’, I take myself to be following common usage in the relevant literature. E. J. Coffman (B) The University of Tennessee, Knoxville, TN, USA e-mail: [email protected]

123

472

Synthese (2011) 181:471–488

In Sect. 1, I’ll argue that prominent parts of two related current debates in epistemology depend crucially on one or the other of DID and JSW. The debates I have in mind are those over classical invariantism (the thesis that a given knowledge-ascribing sentence’s truth-value neither varies across contexts nor depends on “non-truth-relevant” factors—e.g., how important it is to the subject that the proposition be true) and the knowledge first movement (which “takes the simple distinction between knowledge and ignorance as a starting point from which to explain other things, not as something itself to be explained” [Williamson 2000, p. v]). In Sect. 2, I’ll present and defend novel arguments against both DID and JSW. There, I’ll be trying to establish the following two conclusions: • •

Even if an item is in some way improper because it lacks a particular epistemic feature, the item might nevertheless be epistemically proper. Even if you’re not positioned to make an epistemically proper assertion that P, you might nevertheless be positioned to hold an epistemically proper belief that P.

Given the findings of Sect. 1, my arguments for these claims threaten to undercut prominent parts of current debates over classical invariantism and the knowledge first movement. I’m hopeful, however, that the arguments developed in Sect. 2 will hold interest independent of the ramifications they arguably have for those debates. 1 This section’s goal is to show that prominent parts of the debates over classical invariantism and the knowledge first movement depend on one or the other of DID and JSW. These debates are unified in that arguments central to each employ the so called knowledge account of assertion (KAA)—the thesis that you’re positioned to make an epistemically proper assertion that P iff you know that P, “roughly summarized in the slogan ‘Only knowledge warrants assertion”’ (Williamson 2000, p. 243).2 In the debate over classical invariantism, DeRose (2002, pp. 187–188; cf. 2009, pp. 106–107) develops and defends this influential argument from the KAA against classical invariantism: The knowledge account of assertion provides a powerful argument… against the classical invariantist. […] Recall that the classical invariantist denies that varying epistemic standards provide the truth-conditions for knowledge attributions, including first-person claims to ‘know’… But if, as KAA would have it, the standards for when one is in a position to warrantedly assert that P are the same as those that constitute a truth-condition for ‘I know that P’, then if the former vary with context, so do the latter. In short: KAA… together with the context-sensitivity of assertability yields the conclusion that the truth-conditions 2 The knowledge account of assertion can, and here will, be detached from two related claims that William-

son (2000, pp. 238–241) makes on its behalf (cf. Brown 2008, pp. 96–99; DeRose 2009, pp. 92–93)—viz., that the account focuses on a standard for assertion that (a) is the only one governing every possible assertion (in Williamson’s terminology: the account captures “the single constitutive rule of assertion”) and, as a result, (b) serves to distinguish assertion from every other speech act.

123

Synthese (2011) 181:471–488

473

of ‘I know that P’ are in fact context-sensitive in the way the classical invariantist denies. We’ll return to this argument—what DeRose calls the argument from variable assertability conditions—below, for I’ll be contending that it depends crucially on DID. What needs noting here is the essential role the KAA plays in this portion of the debate over classical invariantism. Shifting focus to the debate over the knowledge first movement, consider Williamson’s (2000, p. 251) insight that one can “argue for [the thesis that one’s evidence is just what one knows] from the knowledge account of assertion, given the… premise that one’s evidence consists of just those propositions which the rules of assertion permit one to assert outright.” Combined with the view (endorsed by Williamson 2000, pp. 207–208 and many others) that epistemic justification supervenes on evidence, Williamson’s suggested argument from the KAA to the equation of evidence with knowledge yields an account of a third important epistemic concept— viz., epistemic justification—in terms of knowledge.3 In this way, the KAA can be employed in a promising attempt to advance the knowledge first movement. The current debates over classical invariantism and the knowledge first movement are thus unified in that arguments central to each employ the KAA. I’ll soon begin my argument that prominent moves in these related debates depend crucially (though sometimes subtly) on one or the other of DID and JSW. At this juncture, though, it’s worth noting that some who discuss the KAA—which will be a main character in the sequel—don’t state it quite like I did above: (S1) You’re positioned to make an epistemically proper assertion that P iff you know that P. Consider three statements of the KAA suggested in recent work by Brown (2008, pp. 89–90): (S2) You are “in a good enough epistemic position to assert that P” iff you know that P. (S3) You’re positioned to make an assertion of P that’s not (on balance) improper for epistemic reasons iff you know that P. (S4) You’re positioned to make an assertion of P with which “there is nothing epistemically wrong” iff you know that P. For reasons of charity, Brown prefers S2–S4 (she treats them as though they’re equivalent) as statements of the KAA to ones like the following: (S5) You’re positioned to make a proper assertion that P iff you know that P. The KAA is an obvious nonstarter when understood as S5 (“[f]or instance, even if one knows that one’s boss is bald, it may not be polite or prudent to say so”), but not (Brown thinks) when understood as (one of) S2–S4. Charity thus demands, Brown suggests, that we understand the KAA as (one of) S2–S4. 3 Writes Williamson (2000, p. 9): “[J]ustification is primarily a status which knowledge can confer on

beliefs that look good in its light without themselves amounting to knowledge. Knowledge itself enjoys the status of justification only as a limiting case, just as, trivially, every shade of green counts as similar to a shade of green.”

123

474

Synthese (2011) 181:471–488

My reasons for preferring S1 to Brown’s S2–S4 are similar to those Brown gives for preferring S2–S4 to S5: S1 is a more charitable expression of the KAA than are S2–S4. Starting with S2, it’s unclear what’s meant by ‘in a good enough epistemic position to assert P’. On one natural reading, you’re in a good enough epistemic position to assert P iff you understand P well enough to assertively utter a sentence expressing P. But if that’s how we read the indicated locution, the KAA amounts (on S2) to the absurdity that you know P iff you understand P well enough to assertively utter a sentence expressing P. As for S3, Williamson himself (2000, p. 256) describes a counterexample to its “left-to-right” direction that (he says) leaves the KAA (whatever exactly it is) intact: Sometimes one knows that one does not know P, but the urgency of the situation requires one to assert P anyway. I shout ‘That is your train’, knowing that I do not know that it is, because it probably is and you have only moments to catch it. Such cases do not show that the knowledge rule is not the rule of assertion. They merely show that it can be overridden by other norms not specific to assertion. The other norms do not give me warrant to assert P, for to have such warrant is to satisfy the rule of assertion. According to Williamson, while he doesn’t know the content of his assertion about the train, his assertion is nevertheless (on balance) proper (it’s what “the urgency of the situation requires”)—and so, not made (on balance) improper by epistemic considerations. Thus, understanding the KAA as S3 would in effect be to unnecessarily ascribe serious confusion to Williamson and other proponents of the KAA who make points similar to that made in the above passage. Turning finally to S4, it too is unclear: what’s meant by ‘assertion with which there is nothing epistemically wrong’? On one natural interpretation, that would be an assertion having no epistemic defects whatsoever. But if we understand the locution in this way, the “right-to-left” direction of S4 clearly seems false. Suppose you know P, but don’t know that you know P. S4 implies you’re positioned to make an assertion of P having no epistemic defects whatsoever. Plausibly, though, any assertion whose subject doesn’t know whether he knows its content has at least one (perhaps very minor) epistemic defect. So, if understood as S4, the KAA clearly seems false: given that you’re ignorant of your knowledge that P, you’re not positioned to make an epistemically flawless assertion that P, despite the fact that you do know P. The upshot of the last few paragraphs is that the KAA is a pretty obvious nonstarter when understood as any of S2–S4. The KAA at least seems worth considering, however, when understood as S1. Accordingly, S1 is the statement of the KAA we’ll adopt in what follows. Of course, S1 can be no clearer than is the concept of epistemically proper assertion. Before moving on, then, I should pause briefly to clarify that concept. To begin, given that assertion is an essential source of information (it’s one of the main ways we affect each other’s beliefs), it’s quite plausible to think that assertions can promote or frustrate achievement of epistemic goals (goods, values)—i.e., those goals (goods, values) that constitute the epistemic perspective or viewpoint. This is especially so on a plausible “pluralist” construal of the epistemic perspective, according to which the epistemic perspective is constituted by such goods as (e.g.) each person in your community’s having a large and diverse belief set with a high

123

Synthese (2011) 181:471–488

475

percentage of beliefs that are strong candidates for knowledge.4 These points should help mitigate initial skepticism about the very idea that assertions can be evaluated from the epistemic perspective. We can further tighten our grasp on the distinction between epistemically proper and epistemically improper assertion by considering certain well-chosen pairs of sample assertions. Consider, e.g., the following pair of assertions DeRose contrasts in his (1996, p. 568): In some lottery situations, the probability that your ticket is a loser can get very close to 1. Suppose, for instance, that yours is one of 20 million tickets, only one of which is a winner. Still, it seems that… [y]ou are in no position to flat-out assert that your ticket is a loser. ‘It’s probably a loser’, ‘It’s all but certain that it’s a loser’, or even, ‘It’s quite certain that it’s a loser’ seem quite alright to say, but, it seems, you are in no position to declare simply, ‘It’s a loser’. […] Things are quite different when you report the results of last night’s basketball game. Suppose your only source is your morning newspaper, which did not carry a story about the game, but simply listed the score, ‘Knicks 83, at Bulls 95’, under ‘Yesterday’s Results’. Now, it doesn’t happen very frequently, but, as we all should suspect, newspapers do misreport scores from time to time. […] Still, when asked, ‘Did the Bulls win yesterday?’, ‘Probably’ and ‘In all likelihood’ seem quite unnecessary. ‘Yes, they did’, seems just fine. I think the “lottery” and “score” assertions DeRose contrasts here serve to illustrate the difference between (respectively) epistemically improper and epistemically proper assertion. Finally, in Sect. 2 below, I’ll present two plausible sufficient conditions for an assertion’s being epistemically defective or improper, and describe sample assertions that meet those conditions. The discussion there should further clarify the notion of warranted assertion in play here. Having made an effort to shed some light on the concept of warranted assertion, I’ll now devote the remainder of this section to arguing for the following two conclusions: •

•

DID underwrites (a) DeRose’s (2002, 2009) argument from variable assertability conditions against classical invariantism, as well as (b) the main counterexamples to the view that knowledge suffices for warrant to assert (the “right-to-left” direction of the KAA). JSW underwrites (c) Sutton’s (2005, 2007) assertion argument for the thesis that epistemically justified belief suffices for knowledge, as well as (d) a recent argument from Wright (1996) against the view that knowledge is required for warrant to assert (the “left-to-right” direction of the KAA).

In short, prominent parts of the debates over classical invariantism and the knowledge first movement depend on one or the other of DID and JSW. (a) As we saw above, DeRose (2002, 2009) presents the following influential KAA-based argument against classical invariantism. Your having warrant to assert P can vary across conversational contexts in which you bear precisely the same 4 For articulation of—and supporting arguments for—such a “liberal” conception of the epistemic perspec-

tive, see (e.g.) Kawall (2002) and Kvanvig (2005).

123

476

Synthese (2011) 181:471–488

epistemic relation to P. That is, for some proposition(s) P, there are pairs of “epistemically identical” conversational contexts such that you have warrant to assert P in one but not the other of the contexts. But by the KAA, the conditions in virtue of which you have warrant to assert P are logically equivalent to those in virtue of which you know P. So, since the former can vary across “epistemically identical” conversational contexts, so can the latter. Pace the classical invariantist, then, the truth-value of a knowledge-ascribing sentence can vary across “epistemically identical” conversational contexts. Here is DeRose’s (2002, pp. 187–188; cf. 2009, p. 107) support for the premise that warrant to assert can vary across such conversational contexts: It is difficult to deny that the matter of how well positioned one must be with respect to a matter to be able to assert it varies with context: What one can flatout assert in some “easy” contexts can be put forward in only a hedged manner (“I think…,” “I believe…,” “Probably…,” etc.) when more stringent standards hold sway. […] And it’s clear that this is true not only of knowledge attributions, but of assertions generally—clear enough that I trust that illustrative examples are not needed here. As DeRose makes clear (2002, pp. 198–199, n. 19; cf. 2009, p. 91, n. 12), the modal terms ‘able’ and ‘can’ here express the notion of conversational permissibility. DeRose’s thought is that there can be a pair of “epistemically identical” conversational contexts such that your asserting P would convey a false proposition about your epistemic position relative to P—and so, would be conversationally improper—in one but not the other of the contexts. It follows from this, DeRose thinks, that warrant to assert can vary across “epistemically identical” contexts. DeRose’s attempt to buttress the relevant premise of his argument from variable assertability conditions depends on DID. Here’s what clearly seems true in light of DeRose’s subargument: there can be a pair of conversational contexts such that your asserting P would misrepresent your epistemic relation to P in one context (the “hard” context) but not the other (the “easy” context). In the “hard” (but not the “easy”) context, your asserting P would be conversationally improper for epistemic reasons: for in the “hard” (but not the “easy”) context, you don’t bear to P the epistemic relation your asserting P would there represent you as bearing to P. DeRose infers from this that warrant to assert can vary across “epistemically identical” conversational contexts. But you can’t justifiedly infer the context-sensitivity of warranted assertability from the above considerations unless you’re justified in thinking that an assertion’s being conversationally improper for epistemic reasons suffices for the assertion’s being epistemically improper (unwarranted). And it’s difficult to see how you could be justified in believing the indicated sufficiency claim absent justification to believe the more general DID—which, in a slogan, says that impropriety due to epistemic considerations suffices for epistemic impropriety. In this way, DeRose’s argument from variable assertability conditions against classical invariantism seems dependent on DID. (b) We’ve seen that much recent work on assertion centers on the question how having warrant to assert relates to knowing. Many philosophers have jointly built an

123

Synthese (2011) 181:471–488

477

impressive prima facie case that you have warrant to assert P only if you know P.5 Call this thesis—the “left-to-right” direction of the KAA—Knowledge Is Necessary (KIN). Others aim to undermine the case for KIN by arguing that the data it appeals to are explained at least as well by certain weaker requirements on warranted assertability.6 Most parties to the debate over KIN seem to agree, however, that knowledge suffices for warranted assertability: provided that you know P, you have warrant to assert P. Call this—the “right-to-left” direction of the KAA—Knowledge Is Sufficient (KIS).7 The apparent widespread agreement on KIS is noteworthy, as the view bears importantly on (among other things) the debate over classical invariantism. We’ve just seen how DeRose combines KIS with KIN to argue against the classical invariantist. Further, Brown (2008, pp. 95–96) has recently shown that KIS itself arguably entails the denial of classical invariantism (when combined with a couple other prima facie plausible claims).8 But while there’s widespread agreement on KIS, the view has recently come under attack. Levin (2008), Brown (forthcoming), and Lackey (forthcoming) all present what they regard as counterexamples to KIS, cases in which (they claim) a person knows P yet lacks warrant to assert P. After relaying a representative case, I’ll argue that any such attempted counterexample to KIS depends on DID. Consider the following representative attempted counterexample from Lackey (forthcoming, pp. 3–4, 6 of ms): Doctor: Matilda is an oncologist at a teaching hospital who has been diagnosing and treating various kinds of cancers for the past fifteen years. One of her patients, Derek, was recently referred to her office because he has been experiencing intense abdominal pain for a couple of weeks. After requesting an ultrasound and MRI, the results of the tests arrived on Matilda’s day off; consequently, all of the relevant data were reviewed by Nancy, a competent medical student in oncology training at her hospital. Being able to confer for only a very brief period of time prior to Derek’s appointment today, Nancy communicated to Matilda simply that her diagnosis is pancreatic cancer, without offering any of the details of the test results or the reasons underlying her conclusion. Shortly thereafter, Matilda had her appointment with Derek, where she truly asserts to him purely on the basis of Nancy’s reliable testimony, “I am very sorry to tell you this, but you have pancreatic cancer.” […] [W]hile Nancy’s reliable testimony may be sufficient for Matilda’s knowing that Derek has pancreatic cancer, and while its

5 See, e.g., Unger (1975), DeRose (1991, 1996, 2002), Williamson (2000), Adler (2002), Reynolds (2002), and Hawthorne (2004). 6 See, e.g., Weiner (2005), Douven (2006), Kvanvig (2003, 2009), Lackey (2007), and Coffman(ms). 7 Prominent advocates of KIS include Williamson (2000), DeRose (2002), Reynolds (2002), Douven

(2006), and Kvanvig (2009). Williamson in particular is better known as a proponent of KIN; but see his (2000, pp. 241, 252), where he says (respectively) that (a) “If an assertion satisfies the [C rule, which soon becomes KIN], whatever derivative norms it violates, it is correct in a salient sense [=warranted]” and (b) “To have the (epistemic) authority [=warrant] to assert p is to know p.” 8 Notably, Brown ultimately rejects her formulation of the indicated KIS-based argument against classical invariantism—what she calls the sufficiency argument. For a reformulation of the sufficiency argument and a defense of it from Brown’s main objection, see Coffman (Forthcoming a).

123

478

Synthese (2011) 181:471–488

isolated nature may not pose an epistemic obstacle to this being the case, the isolated secondhand nature of Matilda’s knowledge makes it improper for her to flat out assert this diagnosis to Derek. One reason for this is that Matilda is an expert—she is an oncologist and Derek’s physician, and such roles carry with them certain epistemic duties. I contend that putative counterexamples to KIS like Doctor depend on DID: you’re justified in thinking them counterexamples to KIS only if you’re justified in believing DID.9 Contraposed, my contention is this: if you’re not justified in believing DID, then you’re not justified in thinking that cases like Doctor are counterexamples to KIS. As many readers will have noticed, Lackey’s own commentary on Doctor suggests that what’s clearly true of Matilda is this: because she bears a somewhat suboptimal epistemic relation to (or: harbors certain “epistemic deficiencies” regarding) the content of her assertion to Derek, that assertion is (in the first place) professionally improper. Main factors making Matilda’s assertion professionally improper include “the isolated secondhand nature of” Matilda’s belief in her assertion’s content, as well as the fact that her testifee (Derek) hasn’t yet recognized the source of Matilda’s belief (her student Nancy) as an expert on the relevant subject matter. When Matilda tells Derek that he has pancreatic cancer, she isn’t able to convey to him any of the “reasons underlying [the diagnosis]”. All Matilda can now say in support of her distressing news is “That’s what my competent student Nancy told me”. Given how unprepared Matilda is to discuss Derek’s condition with him, starting that discussion as she does seems quite unprofessional. Given Matilda’s apparent failure to fulfill important professional responsibilities, her behavior will also strike many as both morally and prudentially problematic. What’s clearly true of Matilda, then, is this: her assertion to Derek is professionally (and, as a result, likely morally and prudentially) inappropriate because she bears a somewhat suboptimal epistemic relation to (or: harbors certain epistemic deficiencies regarding) the content of her assertion. But you can’t justifiedly infer from this that Matilda lacks warrant to assert the indicated proposition unless you’re justified in believing DID—in particular, unless you’re justified in believing that an assertion’s being somehow inappropriate because it lacks a particular epistemic feature suffices for its being epistemically inappropriate.10 We can conclude that if you’re not 9 For other attempted counterexamples to KIS relevantly similar to Doctor, see Lackey (forthcoming,

pp. 4–5, 13 of ms), Levin (2008, pp. 373–375), and Brown (forthcoming, p. 5 of ms). My argument that viewing Doctor as a counterexample to KIS depends on DID can be applied, mutatis mutandis, to all the attempted counterexamples to KIS described by Lackey, Levin, and Brown. I must leave such application as homework for interested readers. 10 It’s worth noting here how very close Lackey comes to explicitly endorsing DID. Consider the following two quotations in which she articulates her paper’s target (she calls it ‘KNA-S*’):

[P]roponents of the KNA-S* emphasize that knowing that p is sufficient for possessing the requisite epistemic credentials to properly assert that p. (Forthcoming, p. 25 of ms) These general considerations provide further reason to reject the thesis that knowledge is sufficient for epistemically proper assertion. Let us now turn to some responses that may be offered on behalf of the KNA-S*. (Forthcoming, p. 17 of ms)

123

Synthese (2011) 181:471–488

479

justified in believing DID, then you’re not justified in regarding cases like Doctor as counterexamples to KIS—i.e., as cases involving a subject who knows a proposition yet isn’t positioned to make an epistemically proper assertion of it. This and other similar putative counterexamples to KIS thus depend on DID. (c) Sutton (2005, 2007) argues that epistemically justified belief suffices for knowledge: you hold a justified belief in P only if you know P. Call this thesis justification entails knowledge (JEK). Obviously, Sutton’s arguments for JEK threaten numerous common views—e.g., that there can be justified false beliefs; that “Gettierized” beliefs are justified but not knowledge; that you can justifiedly believe you’ll lose the lottery without knowing it; that we can use epistemic justification to provide a noncircular analysis of knowledge; and so on. In short, his arguments promise to significantly advance the so called knowledge first movement. In what follows, I’ll show how one of Sutton’s main arguments for JEK—what he calls the assertion argument—depends on JSW (the claim that justification to believe suffices for warrant to assert).11 Here is Sutton’s assertion argument for JEK (2005, pp. 375–376; cf. 2007, pp. 44– 48): [S]uppose that Andy has a justified… belief that p that does not amount to knowledge that p… [Andy] asserts that p to Bob who has, we can suppose, the very best reasons for thinking—indeed, he knows—that Andy is expressing what is for him a justified belief. Bob, then, has acquired a… belief that p that is justified… And yet, the knowledge rule12 tells us, Andy should not have asserted that p. This is exceptionally puzzling… If the beliefs transmitted [from one thinker to another via assertion] meet the primary standards governing good belief for both speaker and hearer…, it would be mysterious if the assertions transmitting the beliefs failed to meet the standards governing good assertion. On the contrary, the assertions in question have to meet the standards governing good assertion impeccably since they transmit impeccable beliefs. It is not, however, the knowledge rule that is at fault… It is our initial supposition that was at fault. There are no justified… beliefs falling short of knowledge… Focus on this key explicit premise of Sutton’s assertion argument: “[Assertions like Andy’s] have to meet the standards governing good assertion impeccably since they transmit impeccable beliefs.” Employing this premise commits the assertion argument’s proponent to the following more general claim: • If you are positioned to express an epistemically proper belief in P, then you’re positioned to make an epistemically proper assertion of P. Footnote 10 continued These and other passages strongly suggest that Lackey equates the property of meeting all the epistemic requirements on properly asserting P with the property of being positioned to make an epistemically proper assertion of P. Endorsing such an equation clearly commits one to DID, which (when contraposed) says that an “epistemically evaluable” item like an assertion is epistemically proper only if it meets all the epistemic requirements on other (not purely epistemic) kinds of propriety (“general” or “vanilla” propriety included, if there is such a thing; for worries that there’s not, see Sect. III of Feldman (2000)). 11 For critical assessment of Sutton’s other arguments for JEK, see Coffman (Forthcoming b). 12 That is, the claim that “warranted assertion requires knowledge” (Sutton 2005, p. 374), what we’re here

calling Knowledge Is Necessary (KIN).

123

480

Synthese (2011) 181:471–488

And this more general claim seems to carry commitment to JSW: it’s difficult to see how you could coherently endorse the former while being disposed to reject—or even withhold on—the latter. So, it seems you’re not justified in believing the above key explicit premise unless you’re also justified in endorsing JSW. Sutton’s assertion argument for JEK thus seems dependent on JSW. (d) Consider, finally, the following anti-KIN argument sketched by Wright (1996, p. 935): [W]arranted assertion… is simply the exterior counterpart of warranted belief and there is, prima facie, no plausibility whatever in the suggestion that possession of a sufficient reason to believe a proposition demands nothing less than knowledge of it. Notice that Wright’s argument against KIN “reverses” Sutton’s assertion argument (cf. Sutton 2005, p. 394). Whereas Sutton argues from JSW and KIN to JEK, Wright argues from JSW and ∼JEK to ∼KIN. The important thing to see, of course, is that Wright’s argument against KIN—and so, the KAA—depends on JSW no less than does Sutton’s argument for JEK. We’ve now seen that prominent parts of the debates over classical invariantism and the knowledge first movement depend crucially on one or the other of DID and JSW. Besides being interesting in its own right, the question whether DID and JSW are correct bears on several important current issues in epistemology. Recognition of such facts helps motivate the following critical assessment of DID and JSW. 2 (a) My first argument against DID is an argument from analogy. Its key premise is that analogues of DID for other kinds of propriety are clearly false. I’ll focus on analogues of DID for moral and prudential propriety: Moral Analogue (MA): If a morally evaluable item is in some way improper because it lacks a particular moral feature, then the item is morally improper. Prudential Analogue (PA): If a prudentially evaluable item is in some way improper because it lacks a particular prudential feature, then the item is prudentially improper. I begin with an argument against MA. Suppose I do at noon an act that’s morally permissible—though not obligatory—for me to do around noon: I send you an e-mail. Suppose also that you had offered me a modest monetary reward for doing at noon an act I’m obligated to do around noon—e.g., feed my children. Assuming I could just as easily have fed my children at noon, sending you an e-mail then was somewhat imprudent: by sending you a note instead of feeding my children, I sacrificed that modest reward you had offered.13 By hypothesis, though, sending you that e-mail was morally permissible (I’m not obligated to feed my children at noon, just around noon). 13 If the reward had been sufficiently large, perhaps it would have been morally impermissible for me to

sacrifice it in this way. Pretend the reward was not that large.

123

Synthese (2011) 181:471–488

481

So, sending you the e-mail was morally permissible yet imprudent because it lacked a certain moral feature—viz., being an act I’m obligated to do around noon. We can conclude that a thing’s being somehow improper due to lack of a certain moral feature doesn’t suffice for the thing’s being morally improper. MA is false. Now for an argument against PA. Presumably, there can be prudent acts that aren’t preceded by an explicit calculation of expected utilities. Suppose that one evening I prudently work on a paper of my own instead of grading my students’ exams. This act wasn’t preceded by an explicit calculation of expected utilities. Now, while I didn’t promise my students I’d grade their exams that evening, I did make them the following conditional promise: I promised to work on my paper only if I found this course of action to be the one favored by an explicit calculation of expected utilities. Fill in the details so that I’ve done something morally wrong in breaking this conditional promise to my students. Then (by hypothesis) working on my paper was prudent yet morally wrong because it lacked a certain “prudential” property—viz., being preceded by an explicit calculation of expected utilities.14 We can conclude that a thing’s being somehow improper due to lack of a certain prudential property doesn’t suffice for the thing’s being prudentially improper. PA is false. So, analogues of DID for moral and prudential propriety are false. Given that epistemic propriety is in many ways similar to these other kinds of propriety, the falsity of MA and PA constitutes some reason to think DID is false too. Of course, there are also important differences between epistemic propriety, on the one hand, and moral and prudential propriety, on the other. So I shouldn’t rest my case against DID just yet. Rather, I’ll amplify that case by describing what strikes me as a straightforward counterexample to DID. Consider the following passage from Chisholm (1977, pp. 12–13): [I]t is at least conceivable that a man may have the duty to accept a true proposition which he does not know to be true. For example, a man may have the duty to believe that the members of his family are honest or faithful without in fact knowing that they are. Or a sick man, who has various unfulfilled obligations, may have the duty to accept certain propositions [about his prospects for recovery] if, by accepting them, he can make himself well and useful once again. A person in circumstances like those Chisholm describes may be epistemically justified in believing propositions it would be morally or prudentially improper for him to believe. For example, we can imagine a scenario in which it would be morally improper for you to make a certain negative moral judgment about a close friend or relative of yours: you harbor some doubt as to whether the judgment is true, and as a result morally ought to let this person benefit from your doubt (you morally should “give this person the benefit of the doubt”). Nevertheless, we can suppose that whatever doubt you harbor isn’t enough to keep you from being epistemically justified in making the relevant negative moral judgment: consistent with your harboring some doubt about 14 I assume that being preceded by an explicit calculation of expected utilities is a “prudential” property in the same way that (e.g.) being an instance of knowledge is an “epistemic” property, being an act I’m obligated to perform is a “moral” property, being an instance of C-fiber stimulation is a “physical” property, and being an instance of pain is a “mental” property.

123

482

Synthese (2011) 181:471–488

the judgment, you may well have evidence strong enough to (epistemically) justify you in making that judgment (at least when combined with, e.g., your reliability on such matters, the proper function of your relevant cognitive faculties, and so on). Now suppose that, under such conditions, you go ahead and make an epistemically justified negative moral judgment about the relevant person (e.g., that he’s dishonest and manipulative). Because you harbor some doubt as to whether the judgment is true, you’re not (yet) morally justified in making that judgment. So, your negative moral judgment is epistemically justified yet morally unjustified, and for epistemic reasons: you harbor some doubt that you morally should be letting this person benefit from. What this kind of case shows is that there could be an item that’s epistemically proper yet improper in some other (“nonepistemic”) way for epistemic reasons or considerations. Upshot: while impropriety due to epistemic reasons and being epistemically improper often travel together, they sometimes part ways. DID is therefore false. Of course, a strong argument for DID would at least partially offset the above arguments against it. But I’m not aware of any promising arguments for DID.15 It’s worth noting that my argument from analogy against DID reveals that we can’t plausibly argue to DID from parallel principles involving moral and prudential propriety, for such principles are clearly false. On the basis of the arguments presented above, then, I conclude that DID is false. This conclusion impugns any project dependent for its success on the plausibility of DID—e.g., DeRose’s argument from variable assertability conditions against classical invariantism, and the main attempted counterexamples to KIS. (b) We turn now to JSW. I’ll soon present two “example-driven” arguments against JSW. First, I want to make a couple preliminary points in order to soften up readers for the forthcoming anti-JSW arguments. Justification to believe and warrant to assert are doubtless similar in many ways. So, if the forthcoming arguments succeed and JSW is indeed false, then it’s possible that you bear some but not all of a group of very similar relations to a given proposition or state of affairs. But this point doesn’t by itself lend a whit of support to JSW, since it’s clearly possible that you bear some but not all of a group of very similar relations to a given proposition or state of affairs. Examples abound; here’s one that doesn’t involve epistemic relations. Unlike some of those who legitimately attend departmental meetings (e.g., student representatives), I have the authority to move that the meeting be adjourned, and also to vote that the meeting be adjourned. Unfortunately, I don’t (now) have the authority to adjourn departmental meetings—i.e., to directly bring it about that a departmental meeting is adjourned. So, I have the departmental authority to bear some but not all of a group of similar relations (moving that, voting that, directly bringing it about that) to [The departmental meeting is adjourned].16 If the forthcoming anti-JSW arguments succeed, then you could find yourself in the following predicament which is relevantly similar to my departmental predicament: though you’re well positioned enough

15 Notably, none of the DID-dependent work discussed in Sect. 1 above offers so much as a hint of an argument for DID. 16 Here and elsewhere, ‘[P]’ abbreviates ‘the proposition that P’.

123

Synthese (2011) 181:471–488

483

relative to P to justifiedly believe it, you’re not well positioned enough relative to P to warrantedly assert it. Second preliminary point. A bit of elementary reflection on the nature of belief and assertion should at least make us wonder whether JSW is true. Things happen when you assert P that don’t happen when you merely believe P. First (and most obviously), when you make an assertion, you “propel a proposition out into a conversation with assertoric force” (Slote 1979, p. 178). Further, in asserting P, you represent yourself as being in certain cognitive states relative to P. For example, you represent yourself as knowing—and so, as believing—P.17 Finally, one plausible element of the so called dialectical model of assertion18 —on which an assertion is “a move in the game of giving and asking for reasons”—is that “by asserting a proposition, [you] commit [yourself] to defending the proposition when faced with challenges and counterarguments” (Rescorla 2009, p. 3). The point that needs emphasizing here is that none of these things that happen when you assert P happen when you simply believe P. By merely believing a proposition, you don’t propel that proposition out into conversational space; represent yourself as knowing—or even believing—the proposition; or commit yourself to defending the proposition should someone challenge it. So, in light of the point that things happen when you assert that don’t happen when you merely believe, it’s worth wondering whether being well positioned enough relative to P to justifiedly believe P ensures that you’re well positioned enough relative to P to warrantedly assert it. I’ll now present two arguments that the former doesn’t ensure the latter. My first argument against JSW employs the notion of a lottery proposition—i.e., a proposition which “while highly likely, is [one] that we would be intuitively disinclined to take ourselves to know” (Hawthorne 2004, p. 5; cf. Vogel 1990, p. 17). Examples include [I won’t win the upcoming lottery I hold a ticket in]; [The letter I just sent overseas will arrive safely at its destination]; [The next flight I take won’t crash]; and [I won’t be in a car accident on my way home from work tonight]. It seems possible that a person, S, have justification to believe some or other lottery proposition, L.19 But we can now fill in the details of S’s case so that S lacks warrant to assert L. Here’s one way to tell the story. Start by supposing that S is “intuitively disinclined to take [herself] to know” L, and thus has justification to think she doesn’t know L. On this occasion, S is right: S doesn’t know L. (I leave it open whether we ever know lottery propositions; for the record, I think we often do.) Indeed, S doesn’t even believe L—she’s withholding on L—and has justification to believe this fact about L’s cognitive status for her. Now suppose further that S is justified in regarding herself as a reliable source of information about her attitudes to propositions she has considered. Provided S has justification to think herself so reliable, if S were to assert L, [S believes L] would be conveyed to her by a source (S) she’d be justified in thinking reliable on the relevant subject 17 Cf. Black (1952), Moore (1963), Unger (1975), Williamson (2000), and Hawthorne (2004). 18 For helpful recent explanation and discussion of this approach to assertion, see Rescorla (2009). Advo-

cates of this approach include Brandom (1983, 1994), Watson (2004), and Rescorla (2009). 19 See Hawthorne (2004) for an argument that denying we can know lottery propositions leads (via modest

epistemic closure principles) to implausible skeptical positions.

123

484

Synthese (2011) 181:471–488

matter (S’s attitudes to propositions S has considered).20 We can suppose that, under such circumstances, S would come to hold a justified belief, B, that a reliable source of information about S’s attitudes to propositions considered by S has just conveyed [S believes L]. Plausibly, S’s justified belief B would (for S) constitute evidence that S believes L. So, by asserting L, S would thereby provide herself with evidence that S believes L. And this would, we can suppose, defeat (at least temporarily) S’s justification to believe the fact that she doesn’t believe L. Hence, if S were to assert L, S would thereby defeat the justification she currently has to believe a certain fact about L’s cognitive status for her—viz., that she doesn’t believe L. At this point, the following plausible thesis can kick in: •

If you’re positioned to make an epistemically proper assertion of L, then you’re positioned to assert L while retaining whatever justification you currently have to believe facts about L’s cognitive status for you.

Since S isn’t positioned to make such an assertion of L, the above plausible thesis implies that S lacks warrant to assert L. But recall that S is (by hypothesis) justified in believing L. So JSW is false. I should pause here to (a) defend the above line of argument—the lottery proposition argument against JSW—from a potentially serious worry, as well as (b) contrast the argument with a prominent objection from Williamson (2000, pp. 244–263) against accounts of warranted assertability which, like JSW, imply that some or other epistemic status short of knowledge suffices for warrant to assert. The potentially serious worry is that the lottery proposition argument commits its proponent to one or the other (perhaps both) of the following related absurdities: A1: Merely by asserting the most blatant falsehood, I momentarily provide myself with evidence in favor of it. A2: Someone could acquire previously unpossessed justification to believe that P by merely asserting that P.21 If the lottery proposition argument commits its proponent to one or the other of these absurdities, it does so by virtue of its dependence on the following modal claim: MC: Possibly, S gains a justified belief that S believes L by asserting L at a time when S is justified in regarding herself as a reliable source of information about her attitudes to propositions she has considered. But MC doesn’t imply either of A1 or A2. To begin to see that MC doesn’t imply A1, notice that MC is compatible with the following principle: Principle: For some P, even if S (now) has justification to think herself a reliable source of (certain kinds of) information, if S were to assert P, the fact that S asserted P might render S unjustified in thinking herself a reliable source of information. 20 I assume that if S were to assert L, she’d thereby represent herself as believing L—and so, convey (without asserting) that S believes L. 21 The quoted sentences are from anonymous referees’ reports that helped guide me toward the current formulation of the lottery proposition argument. I thank these referees for raising such worries about previous versions of the lottery proposition argument.

123

Synthese (2011) 181:471–488

485

Principle entails that A1 is false. Given Principle, even if you’re now justified in thinking yourself a reliable source of information (and so, positioned to provide yourself with certain kinds of evidence by asserting certain propositions), there are some propositions—including, e.g., “the most blatant falsehoods”—such that your asserting them at least might result in your lacking justification to think yourself reliable. But if A1 is true, there are no such propositions: A1 implies that, no matter what you might assert, you’d still be justified in thinking yourself a reliable source of information. So Principle entails that A1 is false. But MC, remember, is compatible with Principle. So MC doesn’t entail A1 after all. Let’s turn to A2. Does MC imply that S could, merely by asserting L, gain previously unpossessed justification to believe L itself? Not that I can see. Granted, MC arguably would hold such an implication if MC were (explicitly or implicitly) tied to something like the following claim: • Possibly, S gains justification to believe L itself by asserting L at a time when S is justified in regarding herself as reliable on L’s subject matter. As should be clear by now, though, MC isn’t tied to anything nearly as implausible as the above claim. Rather, MC depends at most on the following different and much more plausible thought: • Possibly, S gains justification to think S believes L by asserting L at a time when S is justified in regarding herself as a reliable source of information about her attitudes to propositions she has considered. One can quite comfortably endorse the above thought while denying claims like A2. On close inspection, then, it seems that MC—and more broadly, the lottery proposition argument—doesn’t commit its proponent to any absurdities in the neighborhood of A1 or A2. Now for the promised contrast between the lottery proposition argument against JSW and a somewhat similar line of argument due to Williamson (2000, pp. 244–263). Begin by noting the following two features of the lottery proposition argument: • It’s neutral on the question whether we can ever have warrant to assert a lottery proposition—and so, its proponent can countenance examples suggesting we sometimes do have warrant to assert such propositions.22 • It includes a subargument—utilizing a plausible sufficient condition for an assertion’s being epistemically improper—for its premise that the indicated lottery assertion is in fact epistemically improper. These features give my argument from lottery propositions against JSW a distinct advantage over the lately mentioned objection from Williamson, which the following passage helpfully summarizes: I may believe on good evidence that your lottery ticket did not win; I am not warranted in asserting that it did not win. I may believe on good evidence that I shall not be knocked down by a bus tomorrow; I am not warranted in asserting that I shall not be knocked down by a bus tomorrow. Neither belief nor belief on good evidence warrants assertion. (2000, p. 260) 22 See, e.g., the examples Lackey sketches in her (2007, p. 618).

123

486

Synthese (2011) 181:471–488

As these and other passages make clear, Williamson’s objection to JSW-like claims about warranted assertability depends on an intuition to the effect that we can’t have warrant to assert lottery propositions. By contrast, my lottery proposition argument employs the much weaker premise that we can lack warrant to assert lottery propositions we’re justified in believing, and includes a subargument for this weaker premise (utilizing a plausible sufficient condition for an assertion’s being epistemically improper). In light of these differences, my lottery proposition argument against JSW seems to have a distinct advantage over Williamson’s similar line of argument.23 My second argument against JSW—the argument from Moorean propositions— starts with the claim that a person, S, could have justification to believe a Moorean proposition of the form [P and S doesn’t believe P]. Here’s one proposal for how this could happen (cf. Lackey 2007, pp. 613–616; Douven 2006, p. 461): while S has justification to believe P, S also has justification to believe he’s presently “in denial” with respect to P. Here’s another proposal: while S has justification to believe P, S is also justified in endorsing requirements on believing P that are in fact a good deal too restrictive. Now, provided S has justification to believe a Moorean proposition of the relevant form,24 there will be ways to fill in the details of his case so that he lacks warrant to assert the proposition. Suppose (as before) that S is justified in believing he’s a reliable source of information about his attitudes toward propositions he has considered. Accordingly, if S were to assert the first conjunct of the Moorean proposition he now has justification to believe, the proposition that S believes the indicated proposition would be conveyed to him by a source he’s justified in regarding as reliable on the relevant subject matter.25 So, by a line of reasoning parallel to that employed in the argument from lottery propositions, if S were to assert the first conjunct, S would thereby provide himself with evidence against the second conjunct.26 And that evidence would defeat the justification S currently has to believe the Moorean proposition in question. So, if S were to assert the Moorean proposition he now has justification to believe, he’d thereby defeat his justification to believe that proposition. We can now invoke the following plausible thesis: 23 Thanks to an anonymous referee for urging me to contrast my argument from lottery propositions with Williamson’s related line of argument. 24 Some theorists argue that Moorean propositions can’t have certain positive epistemic features for us. For a helpful introductory discussion, see Green and Williams (2007). A more complete development of the prima facie case now under construction would interact with such arguments. I hope to do so elsewhere. 25 As before, I assume that if S were to assert the proposition in question, he’d thereby represent himself as believing the proposition—and so, convey (without asserting) that S believes the proposition. 26 The parallel argument runs as follows. We can suppose that, as a result of asserting the first conjunct

(P) of the relevant Moorean proposition, S forms a justified belief (B) that a source reliable on S’s attitudes toward propositions considered by S has just conveyed [S believes P]. Plausibly, S’s justified belief B would (for S) constitute evidence that S believes P. So, if S were to assert the first conjunct of the relevant Moorean proposition, he’d thereby provide himself with evidence against the second conjunct (= [S doesn’t believe P]). Notably, while the argument from Moorean propositions is vulnerable to the same kinds of worries considered above in connection with the argument from lottery propositions, the same kind of reply given on behalf of the lottery proposition argument is also available to the proponent of the Moorean propositions argument: nothing in the latter commits its adherent to claims in the neighborhood of A1 or A2.

123

Synthese (2011) 181:471–488

•

487

Any assertion that defeats its agent’s justification to believe its own content is an epistemically defective assertion.

Since S’s asserting (the first conjunct of) the Moorean proposition he now has justification to believe would render him unjustified in believing that proposition, the above plausible thesis entails that S lacks warrant to assert a proposition he nevertheless has justification to believe. JSW is false.27 We’ve now seen that a strong prima facie case can be made against each of DID and JSW. The arguments developed above threaten those claims along with any work dependent on them—e.g., (a) DeRose’s variable assertability conditions argument against classical invariantism; (b) the main putative counterexamples to KIS; (c) Sutton’s assertion argument for JEK; and (d) Wright’s argument against KIN. In all probability, the influence of DID and JSW extends beyond current debates over classical invariantism and the knowledge first movement. I thus suspect that the above arguments against DID and JSW threaten to undermine quite a bit of recent work in epistemology.28 References Adler, J. E. (2002). Belief’s own ethics. Cambridge, MA: MIT Press. Black, M. (1952). Saying and disbelieving. Analysis, 13, 25–32. Brandom, R. (1983). Asserting. Nous, 17, 637–650. Brandom, R. (1994). Making it explicit. Cambridge, MA: Harvard University Press. Brown, J. (2008). The knowledge norm for assertion. Philosophical Issues, 18, 89–103. Brown, J. (Forthcoming). Knowledge and assertion. Philosophy and Phenomenological Research. Chisholm, R. (1977). Theory of Knowledge. Upper Saddle River, NJ. Prentice Hall. Coffman, E.J. (ms). Lenient accounts of warranted assertability. Coffman, E.J. (Forthcoming a). Does knowledge secure warrant to assert? Philosophical studies. Coffman, E.J. (Forthcoming b). Is justified belief knowledge? Philosophical books. DeRose, K. (1991). Epistemic possibilities. Philosophical Review, 100, 581–605. DeRose, K. (1996). Knowledge, assertion and lotteries. Australasian Journal of Philosophy, 74, 568–579. DeRose, K. (2002). Assertion, knowledge, and context. Philosophical Review, 111, 167–203. DeRose, K. (2009). The case for contextualism. Oxford: Oxford University Press. Douven, I. (2006). Assertion, knowledge, and rational credibility. Philosophical Review, 115, 449–485. Feldman, R. (2000). The ethics of belief. Philosophy and Phenomenological Research, 60, 667–695. Green, M., & Williams, J. (2007). Introduction. In M. Green & J. Williams (Eds.), Moore’s paradox. Oxford: Oxford University Press. Hawthorne, J. (2004). Knowledge and lotteries. Oxford: Oxford University Press. 27 In effect, this argument that a person can lack warrant to assert a Moorean proposition blends elements from different attempts to explain the putative fact that any Moorean assertion is somehow inappropriate or defective (including the view that when you assert P, you represent yourself as being in certain cognitive states relative to P). For helpful introductory discussion of such attempts, see Green and Williams (2007). It’s also worth noting that my argument from Moorean propositions against JSW constitutes an objection to Douven’s (2006) overall position on warranted assertion. For Douven maintains that (i) “rational credibility” (roughly, what’s here meant by ‘justification to believe’) suffices for warrant to assert (pp. 449–450), but also concedes that (ii) Moorean propositions can be rationally credible for us (p. 474). Essentially, the above argument shows that (ii) implies the denial of (i). 28 I presented material from this paper at the University of Oklahoma and the University of Tennessee. Thanks to those in attendance for stimulating discussion and helpful feedback. Special thanks to Nathan Ballantyne, Jessica Brown, Jon Kvanvig, Jennifer Lackey, Aidan McGlynn, Mark Sainsbury, John Turri, and some anonymous referees.

123

488

Synthese (2011) 181:471–488

Kawall, J. (2002). Other-regarding epistemic virtues. Ratio, 15, 257–275. Kvanvig, J. (2003). The value of knowledge and the pursuit of understanding. Cambridge: Cambridge University Press. Kvanvig, J. (2005). Truth is not the primary epistemic goal. In M. Steup & E. Sosa (Eds.), Contemporary debates in epistemology. Malden, MA: Blackwell. Kvanvig, J. (2009). Assertion, knowledge, and lotteries. In D. Pritchard & P. Greenough (Eds.), Williamson on knowledge. Oxford: Oxford University Press. Lackey, J. (2007). Norms of assertion. Nous, 41, 594–626. Lackey, J. (Forthcoming). Assertion and isolated secondhand knowledge. In J. Brown & H. Cappelen (Eds.), Assertion: New philosophical essays. Oxford: Oxford University Press. Levin, J. (2008). Assertion, practical reason, and pragmatic theories of knowledge. Philosophy and Phenomenological Research, 76, 359–384. Moore, G. E. (1963). Commonplace book, 1919–1953. London: Allen & Unwin. Pryor, J. (2005). There is immediate justification. In M. Steup & E. Sosa (Eds.), Contemporary debates in epistemology. Malden, MA: Blackwell. Rescorla, M. (2009). Assertion and its constitutive norms. Philosophy and Phenomenological Research, 79, 98–130. Reynolds, S. L. (2002). Testimony, knowledge, and epistemic goals. Philosophical Studies, 110, 139–161. Slote, M. (1979). Assertion and belief. In J. Dancy (Ed.), Papers on language and logic. Keele: Keele University Library. Sutton, J. (2005). Stick to what you know. Nous, 39, 359–396. Sutton, J. (2007). Without justification. Cambridge, MA: MIT Press. Unger, P. (1975). Ignorance: A case for scepticism. Oxford: Oxford University Press. Vogel, J. (1990). Are there counterexamples to the closure principle?. In M. Roth & G. Ross (Eds.), Doubting: Contemporary perspectives on skepticism. Dordrecht: Kluwer. Watson, G. (2004). Asserting and promising. Philosophical Studies, 117, 57–77. Weiner, M. (2005). Must we know what we say?. Philosophical Review, 114, 227–251. Williamson, T. (2000). Knowledge and its limits. Oxford: Oxford University Press. Wright, C. (1996). Response to commentators. Philosophy and Phenomenological Research, 56, 911–941.

123

Synthese (2011) 181:489–514 DOI 10.1007/s11229-010-9743-1

Ambiguous figures and representationalism Athanasios Raftopoulos

Received: 3 December 2007 / Accepted: 16 March 2010 / Published online: 30 March 2010 © Springer Science+Business Media B.V. 2010

Abstract Macpherson (Nous 40(1):82–117, 2006) argues that the square/regular diamond figure threatens representationalism, construed as the theory which holds that the phenomenal character is explained by the nonconceptual content of experience. Her argument is the claim that representationalism is committed to the thesis that differences in the experience of ambiguous figures, the gestalt switch, should be explained by differences in the NCC of perception of these figures. However, with respect to the square/regular diamond and some other ambiguous figure representationalism fails to offer a unified account of how representational content makes them ambiguous. In this paper, I aim, first, to offer a representationalist account of ambiguous figures and, second, to examine and rebut Macpherson’s arguments. My main point is that in each ambiguous figure Macpherson discusses there are differences in representational content that can explain differences in phenomenal character or content. The representational differences are due to the ways the Cartesian frame of reference in which perceptual content is always cast cuts the figure, underlying different properties of the figure with respect to the axes of the Cartesian frame of reference. Keywords Ambiguous figures · Nonconceptual content · Phenomenal character of experience · Representationalism 1 Introduction Ambiguous figures present a challenge to the philosophy of perception. Nonconceptual Representationalism, a theory which holds that the phenomenal character of

A. Raftopoulos (B) Department of Psychology, University of Cyprus, Nicosia, Cyprus e-mail: [email protected]

123

490

Synthese (2011) 181:489–514

experience is either identical to, or supervenes on, some part of the nonconceptual content of experience (NCC), explains the ways ambiguous figures are perceived by reference to this NCC. The same figure is seen in two different ways if on each occasion a different NCC is being represented in the perceiver’s perceptual states (Peacocke 1992; Tye 2002, 2005). Macpherson (2006) argues that Mach’s figure (see Fig. 1) threatens representationalism. The brunt of her argument is that: If one identifies phenomenal character with nonconceptual content then one must suppose that there is some change of nonconceptual content when undergoing a gestalt switch, and so the representationalist must be able to account for the difference in nonconceptual content between an experience associated with seeing Mach’s figure as a square and the experience associated with seeing Mach’s figure as a diamond. (Macpherson 2006, p. 101) Macpherson discusses other ambiguous figures too and examines three attempts purporting to provide a representationalist account of them. She concludes that none of these attempts can account for all ambiguous figures. Therefore, the representationalists should either abandon their position, or they should specify the relevant differences in content among the various ambiguous figures so that switches in the phenomenal content of experience could be attributed to differences in representational contents.1 Let me state, first, that since Macpherson (2006, p. 83) concedes that perception has NCC, in what follows I take the existence of NCC for granted. The aim of this paper is twofold. To sketch a representationalist account of ambiguous figures in which the gestalt switch is caused by differences in NCC, and to show that Macpherson’s challenge fails to undermine representationalism. In Sect. 2, I present in some detail Macpherson’s argument. In Sect. 3, I introduce the distinction between phenomenal seeing and doxastic seeing and relate it to the phenomenal content of experience, which according to nonconceptual representationalism is representational NCC [Tye’s (2002, 2005) PANIC theory of phenomenal content]. In the fourth section, I offer a representationalist account of some among the ambiguous figures. Finally, I examine Macpherson’s arguments and show that they do not undercut representationalism. My main point is that Macpherson’s challenge to representationalism can be met in two ways, depending on the account of what happens when subjects view ambiguous figures (which one of the two is correct is a matter for psychological research to establish). On one account there are two NCC contents involved in all cases of ambiguous figures. On another, there is one NCC but two different conceptualizations of that content. In either case, representationalism faces no problems. If successful, this paper will contribute to the philosophical literature in two ways. First, it offers a detailed representationalist account of ambiguous figures based on psychological and neuropsychological evidence. Such an account is currently missing 1 Macpherson’s thesis is that phenomenal NCC cannot exhaustively explain the phenomenal character. One of my goals is to explain why switches in phenomenal character can be explained by changes in phenomenal content. Thus, by using the term ‘phenomenal content’ to state what is up for explanation I do not try to force Macpherson to a starting point which tacitly presupposes that she is wrong. I am just stating what I purport to do.

123

Synthese (2011) 181:489–514

491

a

Duck –rabbit figure

diamond-square figure (Mach’s figure)

b

regular diamond

square

Fig. 1 Ambiguous figures

although, of course, several philosophers have attempted to explain bits and pieces of what happens when one views an ambiguous figure (Peacocke 1992; Tye 2002, 2005). Second, it mounts an adequate defense of representationalism against the challenge from ambiguous figure by extending and modifying traditional defenses like that of Peacocke’s (1992).

2 Macpherson’s argument against representationalism According to Macpherson (2006, p. 97) representationalists are committed to the view that the two experiences with different phenomenal characters one can have looking at an ambiguous figure must have different NCC’s since the latter are supposed to explain differences in the phenomenal character of the experiences. Macpherson concedes that with some ambiguous figures, as the rabbit/duck figure, such differences in NCC can be found. Specifically, the following account is available to representationalism. Specifying the differences in content between the two experiences that one can have while viewing the pictures, and showing how experiences could come to have those contents…will be reasonably straightforward. This is because there are independent occurrences of ducks and rabbits…that correspond to the two contents in question. When one sees a picture of a duck or a rabbit the content of those experiences will not straightforwardly be ‘rabbit’ or ‘duck’ but

123

492

Synthese (2011) 181:489–514

perhaps something like, ‘duck-looking’, or ‘rabbit-form or ‘picture of a duck.’ (Macpherson 2006, p. 97) Thus, the ‘duck-form’ and ‘rabbit-form’ can be thought of as different contents because there can be something duck-like present without something rabbit-like present. However, this explanation will not do in the case of the square/diamond ambiguous figure because there are no properties that are parts of NCC that are represented when one views a diamond but are lacking when one views a square. Thus, there cannot be two distinct states of affairs, one of which corresponds to diamonds, and the other of which corresponds to squares, because each time a square is instantiated so does a regular diamond (Macpherson 2006, p. 101). One way out would be to argue that in seeing a diamond there is some property that is represented and which is not represented in seeing a square. If there are some worldly states that exemplify this property and some others that do not, then one cold use this to account for the representational difference between the square and the diamond, even though the square/diamond figure necessarily has both properties. In this case, the same figure could be seen in two different ways, depending on whether this property is represented or not. Let me state, first, that there is some confusion lurking in Macpherson’s account of ambiguous figures that somewhat vitiates her account. Often, in discussing the square/diamond figure, the ambiguity pertains to the fact that the two figures in Fig. 1b are seen as a diamond and a square, respectively. When, for example, Macpherson (2006, pp. 88–89, 105–106) discusses Palmer (1992) and Ferrante et al. (1997) she refers to two different token-figures being seen as two different shapes. Ferrante et al. specifically address the problem why a square when it is rotated 45◦ looks like a diamond and not a square; they do not discuss the case of why the same token-figure can be seen first as a diamond and then as a square, which is the Gestalt switch that Macpherson purports to examine. Other times Macpherson’s ambiguity (2006, pp. 84, 91, 93, 95, 101) refers to the fact that a single figure, the Mach figure, on some occasions is reported as a diamond and on other occasions it is reported as a tilted square, which is described as a gestalt switch in experience. The gestalt switch happens when looking at the same stimulus one has an experience with a certain phenomenal character followed by another experience with a different phenomenal character. The same confusion occurs when Macpherson (2006, pp. 106–107) discusses the ambiguous parallelogram. She draws two different figure-tokens (see Fig. 2a), an upright parallelogram that she calls a “non-regular diamond” and a parallelogram that is rotated 45◦ with respect to the first figure. When on the next page she discusses an extension to Ferrante et al. (1997) theory involving the frame of reference that is imposed upon the figure, one would expect her to draw the two different figures framed in the same frame of reference and discuss whether there are different properties represented (see Fig. 2b). Instead, she draws the same figure to which the frame of reference is applied into two different ways (see Fig. 2c) and discusses the claim that owing to the different ways the frames cut the figure different properties are being represented. It follows that the ambiguity Macpherson has in mind is that there are two different ways the same token-figure could be perceived, and not that the two figures in Fig. 2a are seen into two different ways. These confusions, as we shall see, vitiate the way Macpherson handles the empirical evidence.

123

Synthese (2011) 181:489–514

493

a

b

c

Fig. 2 a On the left: an upright parallelogram (a non-regular diamond). On the right: a parallelogram rotated 45◦ . b The upright parallelogram (on the left) and the tilted parallelogram (right) cast in a normal Cartesian framework. c The upright parallelogram (on the left) and the tilted parallelogram (right) cast in a normal and in a tilted Cartesian framework, respectively

Before addressing Macpherson’s attack on various representationalist defenses, one should note that in order for her attack against representationalism to succeed, she must show that differences in experiencing the diamond/square figure are differences in phenomenal awareness and not differences in conceptual seeing. If they were the latter, then one would not have to explain them on the basis of NCC and her argument against representationalism would become irrelevant. Against this possibility, Macpherson (2006, pp. 91–93) argues that a conceptual account of ambiguous figures will not do. She scrutinizes and rejects the idea that “seeing a square as a

123

494

Synthese (2011) 181:489–514

square and then as a diamond did not involve experiences with different phenomenal characters, but instead only involved differences in the categorising, cognising or conceptualisation of the object.” Against that idea she adduces empirical evidence and argues that changes in judgements are not sufficient to generate changes in phenomenal awareness. Macpherson uses this to conclude that the gestalt changes experienced upon viewing the diamond/square figure are not caused by changes in judgment or in general by cognitive/conceptual interferences. Although I agree with Macpherson’s conclusion, we shall see that there is some empirical evidence suggesting that cognitive influences may determine the perception of ambiguous figures, in which case and given the existence of NCC, the differences involved are differences in categorizing and cognizing of the object. Having argued that the gestalt switch cannot be caused by different conceptual interpretations of the image, Macpherson proceeds to block the representationalist strategy of explaining the differences in phenomenal awareness in terms of differences in NCC by finding suitable nonconceptual properties that are represented when a diamond is perceived but are not represented when a square is perceived. Macpherson (2006, pp. 102–110) discusses various possible representationalist defenses along this line starting with Peacocke (1992, p. 77) who assigns different NCC to the two percepts of the Mach figure on account of the fact that when one views the Mach figure and sees a square one has the NCC that there is a symmetry about the bisectors of the shape’s sides, while when one sees the figure as a diamond, one’s experience has the NCC that there is a symmetry about the bisectors of the shape’s angles. Macpherson raises two objections to Peacocke’s account. First, this account works only when the two representational contents that supposedly produce the difference in the phenomenal character of the experiences are inconsistent, that is, when they cannot both fill the same scenario in Peacocke’s theory of NCC. Only then could one argue that the one content but not the other is the NCC of experience and thus determine its character. However, in the square/diamond figure this is not the case because the NCC that represents the symmetry about the axes is consistent with the NCC that represents the axes about the angles and, therefore, there is no reason to assume that one cannot have an experience with both contents. Second, if the axes of symmetry determine the percept when one views the Mach figure, then how is it possible to see a square while focusing one’s attention on its angle bisector symmetry? This possibility entails that the contents pertaining to both types of symmetry are present in one’s experience and, therefore, differences in the type of symmetry by itself cannot explain the different phenomenal characters. Macpherson (2006, pp. 105–106) then moves to discuss another attempt to explain the difference in the phenomenal characters of the experiences of Mach’s figure. This is Ferrante et al. (1997) theory that one can accurately judge whether an angle is a right angle or not when the angle is normal, that is, when it is formed by the two cardinal orientations (horizontal and vertical axes). One cannot do that when the angle is not normal. This is attributed to the fact that the visual system is more sensitive to normal right angles than to non-normal right angles. When one experiences a square, which has normal right angles, the content of one’s experience represents a right angle, while when one experiences a diamond, which does not have normal right angle, the NCC of that experience does not represent a right angle. There are, therefore, two different

123

Synthese (2011) 181:489–514

495

representational contents that are associated with the two different ways the figure is seen. It is at this juncture that the confusion I noted above vitiates Macpherson’s account. Ferrante’s theory explains why upon viewing a square and a regular diamond one sees two different shapes, despite their having the same shape, on account of the fact that when one sees a square one represents a right angle, while when one sees a diamond one does not. The theory cannot explain by itself why a regular diamond can be seen sometimes as a diamond and other times as a square. (In Sect. 4, I present an account that purports to explain this phenomenon.) Macpherson admits that although Ferrante’s theory provides a successful representationalist defense of Mach’s figure, it cannot explain the different phenomenal characters associated with a parallelogram and the same figure when it is rotated 45◦ and yields a non-regular diamond. Given that these figures have oblique angles, an appeal to the privileged perception of normal right angles will not do. (Notice here that Macpherson talks about two different token-figures been seeing into two different ways.) One could attempt to extend Ferrante’s account of the Mach figure to cover the parallelogram case by taking into consideration not right normal angles but the orientation of a framework imposed within or upon one’s experience (Macpherson 2006, p. 107). If a viewer frames the contents of her experience in a set of perpendicular axes labeled top/down and left/right, this framework is imposed on the figures of her experience irrespective of the orientation of the figure with respect to the viewer and irrespective of whether the figure has right angles or not. Then a representationalist could claim that the way the frame cuts the figure yields a different NCC in the two alternative experiences of ambiguous figures. For example, in the square/diamond ambiguous figure, depending on how one imposes a Cartesian framework, one represents the angles first in a normal position and then one represents them as in a not normal position and this causes one to see a square and a diamond, respectively. Similarly, in the parallelogram case the ways the differently oriented frames of reference cut the figure result into two different contents being represented, which readily explains the two different phenomenal contents associated with the non-regular diamond and the parallelogram (see Fig. 2c); the experience corresponding to the left hand figure would represent an angle pointing in the direction of ‘up’. In the experience corresponding to the right hand figure in Fig. 2c, this would not be represented. One should notice that Macpherson shifts the discussion from the ways two different token-figures (see Fig. 2a) are perceived, which is how she introduced the issue on page 106, to the ways the same figure (the non-regular diamond) is perceived. This, again, creates a problem in Macpherson’s attempt to recount the representationalist defense. Although, as I will say in Sect. 4, why a parallelogram is seeing differently from a non-regular diamond can be explained by the different sensitivities of V1 neurons to cardinal orientations (horizontal and vertical lines) than to non-cardinal orientations, this by itself cannot explain why the non-regular diamond is sometimes seen as a diamond and other times as a parallelogram. Be that as it may, this last move would allow representationalists to argue that in all ambiguous figures there are different NCC represented and that these differences in NCC explain the differences in the phenomenal characters of the experiences. Thus, one would have a unified representationalist account of the experience of all ambiguous figures.

123

496

Synthese (2011) 181:489–514

To dismiss this possibility, Macpherson (2006, pp. 107–108) adduces evidence from studies of subjects that view an A and a tilted A and interprets the results of these studies, namely that the subjects report seeing an A in both cases, to suggest that the subjects have experiences with the same phenomenal contents. Based on this, she argues that since subjects experience the same contents on both occasions despite the different ways in which the frame of references in which they are both cast cut the two figures, the account of the gestalt switches based on differences in the way frames of reference cut the figures does not work. This is so because if the way the frames cut the figures determined how they were phenomenally experienced, then the subjects should have experienced a gestalt switch in the case of the A/tilted A as well, which they did not. Since an explanation of why when the frames are imposed on the ambiguous figures cause different phenomenal contents and why when they are imposed on non-ambiguous figures such as the A do not cause different phenomenal contents is lacking, one cannot appeal to the way the frames of reference cut the figures to explain difference in phenomenal characters. This undermines the most empirically adequate representationalist attempt to account for the way one experiences ambiguous figures; if representationalists cannot appeal to the frames of reference to explain differences in representational NCC, then they cannot account for the parallelogram case and, therefore, that figure presents an unsolved challenge for representationalism. Since it is the final obstacle that blocks a successful representationalist account, the A/tilted A argument is crucial in Macpherson’s strategy against representationalism.

3 Phenomenal content, epistemic seeing and non epistemic seeing According to representationalism, the phenomenal character of experience is identical with, or is determined by, the NCC of that experience, that is, by what the experience represents the world as being. Restating this to avoid the lurking content/vehicle confusion, one may say that the phenomenal character of experience is identical with, or is determined by, the representational NCC (that presents worldly states as being in a certain way) of the experiential states (that represent worldly states as being in a certain way). Now, there are two kinds of NCC: representational NCC that is the content of the personal-level experience that one has, for example, when one sees a token of a specific shade of red, and NCC that is the content of sub-personal perceptual states (say, spatial frequencies, edges, or contours) that occur early in perceptual processing before awareness occurs. The second kind of NCC, which may or may not be representational, is the subpersonal computational content (Bermudez 1995). Phenomenal content is representational NCC at the personal level. Perceptual states have a phenomenal character in virtue of having some kind of (re)presentational content. The phenomenal content of experience includes (re)presentations of properties of things, events, or places, that these things, or events, or places have in virtue of appearing to us, or being disposed to appear to us, in certain ways. As we saw, Macpherson agrees that perception has NCC. I have argued elsewhere that for visual experience to have states with NCC there must be a stage of visual processing that is immune to conceptual influences, that is, a stage that is cognitively impenetrable or conceptually encapsulated. The states that are formed at this stage

123

Synthese (2011) 181:489–514

497

have NCC. Following Pylyshyn (2003), let us call this stage ‘early vision’. To distinguish the nonconceptual stage of visual processing from the conceptual stage of visual processing whose states have conceptual content, let us call the latter, ‘late vision’. The distinction between a nonconceptual seeing and a conceptual seeing has a long history. Jackson (1977) proposed that expressions of the form “X looks (or appears) F to S” where ‘F’ expresses a phenomenal property should be distinguished from expressions of the form “X looks (or appears to) as if it is F to S” or “X looks (or appears to) to be F to S”. The former are phenomenal, in that they do not involve concepts, whereas the latter are epistemic in that they involve concepts. Dretske (1997) distinguishes a ‘phenomenal sense of see’ from a ‘doxastic sense of see’, the first corresponding to a non-epistemic or nonconceptual sense of seeing and the second corresponding to an epistemic or conceptual sense of seeing. In the latter case, the content delivered may be the content of judgments and beliefs.2 I call the nonconceptual type of seeing phenomenal seeing or ‘seeingph ’, and the conceptual epistemic type of seeing doxastic seeing or ‘seeingdox ’. In rebutting representationalism, Macpherson discusses an account of ambiguous figures that is based on the way the frame of reference in which perceptual content is cast cuts a figure and, thus, allows different aspects of the figure to be represented. In general, accounts of NCC that try to elucidate its nature underline the importance of the frame of reference in which the content of perception is cast (Campbell 2002; Matthen 2005; Peacocke 1992; Raftopoulos 2009; Raftopoulos and Muller 2006; Tye 2000). Since from the two main visual streams that exist in the brain, the dorsal and the ventral system (Goodale and Milner 1992; Jacob and Jeannerod 2003), consciousness characterizes only the latter, the states that have phenomenal content belong to the ventral pathway. The ventral system uses a relational metric scene-based reference frame; the positions and characteristics of objects are represented in relation to the other objects in the scene (Goodale and Milner 1992; Jacob and Jeannerod 2003). Locations and distances are relative and are not represented in absolute terms, that is, as locations and distances from the observer. The frame of reference in which the relevant positions and distances in space of a set of objects are described is the Cartesian coordinate system that is defined with respect to the position of the body of the perceiver. Peacocke (1992, pp. 61–62) makes this point in elaborating his scenarios. He searches for a level of NCC on which to anchor phenomenal concepts in a noncircular way. This kind of NCC is provided by the spatial types “the type being that under which fall precisely those ways of filling the space around the subject that are consistent with the correctness of the content.” To specify the spatial types one needs to fix an origin and the axes of a spatial frame. These cannot be defined with reference to the world, since a spatial type may be instantiated at different places. They should be defined with respect to something that is present irrespective of the location at which the type is instantiated, which is the subject’s body. The origin may be the center of

2 More specifically, to say that S sees O to be P in the doxastic sense of “seeing” is to say that S believes, or tends to believe, or (ceteris paribus) would believe, on the basis of perception that O is P. This means that P is conceptually articulated content.

123

498

Synthese (2011) 181:489–514

gravity of the body and the axes of reference the directions of right–left, up–down, back–front. NCC is always cast in such a frame of reference.

4 Representationalism and ambiguous figures Macpherson discusses ambiguous figures such as the duck/rabbit and the square/ diamond configurations (see Fig. 1a). She argues that on viewing such figures gestalt switches occur in which the former configuration is perceived as either a duck or a rabbit and the latter is perceived as either a square or a regular diamond. Representationalism’s task is to explain how the two different phenomenal contents could be explained by differences in NCC. Here, I propose a representationalist account of ambiguous figure that is compatible with the relevant scientific evidence.

4.1 The duck/rabbit ambiguous figure During perception figures are individuated from the ground and from other objects in a scene (figure-ground segmentation). Properties of the object such as size, shape, color, orientation, motion are also bottom-up retrieved from the visual scene, and what is seeingph consists in a configuration with the aforementioned properties. (To be more precise, the perceptual states of whose content the perceiver is aware have as content configurations of such properties born by an object). However, on certain occasions more than one figure-ground segmentations in an image are possible, in which case the image is ambiguous. Or, perceptual ambiguity may arise because a figure may be decomposed in more than one ways, or may have more than one internal organization. Ambiguous figures are those that can be perceptually organized in more than one ways. The duck/rabbit ambiguous figure belongs in this category. Though none of the properties of the figure allows by itself the distinction between a rabbit-like and a duck-like figure, it is the way the figure is perceptually organized that determines whether a duck or a rabbit is perceived. Studies with ambiguous figures suggest that seeing a duck or a rabbit when presented with the ambiguous configuration depends on where spatial attention, either bottom-up driven (exogenous attention) or top-down driven (endogenous attention) focuses on the image. Attention enhances or attenuates the availability of certain perceptual categories. The term ‘perceptual set’ is used to denote the factors that guide perceptual acts and consist in the perceptual goals held by the observer during the task. There is extensive research regarding the role of spatial attention in determining what one sees when viewing ambiguous figures, which suggest that both the locus of spatial attention and the perspective from which the figure is viewed are important in determining how one perceives ambiguous figures. Spatial attention can be either bottom-up or image driven, as when the subjects are asked to fixate a certain position on the figure, or top-down conceptually driven, as when subjects are asked to view a figure from different perspectives (to attempt to see a Necker cube, from instance, from the top view or the bottom view). Studies show that fixation position can bias perception of the Necker cube to some extent (Kawabata 1986; Kawabata et al. 1978;

123

Synthese (2011) 181:489–514

499

Magnuson 1970; Peterson and Gibson 1994; Toppino 2003).3 This view falls within the framework of the focal-feature hypothesis in which different points on a figure are assumed to favor one or the other perceptual organization of the figure. In this view, bottom-up factors determine how the ambiguous figure is perceived. Even though such points clearly can cause figure reversals, they are not necessary for them since when subjects are asked to fixate on a single point, this does not stop reversals (Long and Toppino 2004). Other studies suggest a strong effect of top-down selective/active spatial attention on the perception of the Necker cube that is more powerful and independent of the effect of fixation position (Meng and Tong 2004), or a strong interaction between top-down and bottom-up spatial factors in the perception of bistable apparent motion (Suzuki and Peterson 2000). These studies bring to the fore the role of spatial attention in influencing the way the ambiguous figure is experienced by determining its perceptual organization and suggest that the perception of ambiguous figures can be adequately explained by a combination of both bottom-up and top-down processes. These processes are differentially engaged by different viewing conditions; on some occasions bottom-up stimulus-driven attention can disambiguate the image and on other occasions top-down voluntary cognitive-driven attention is needed to modulate bottom-up perceptual processing. In exogenous attention it is the features of the image that “pop-up” and capture attention by making more salient the crucial points, whereas in endogenous attention the focus is determined by the conceptual framework of the perceiver. Suppose one view the duck/rabbit configuration of Fig. 1 that moves from left to right. Then one’s attention would more likely4 focus on the right side of the figure, where the direction of motion is, and the figure would be decomposed and organized in such a way that a rabbit-like figure (that is, a figure that could be described or identified as a rabbit should the perceiver possess the salient concept), would be seenph or perceived. If the figure were moving from right to left then one would perceive a duck-like figure (that is, a figure that could be identified as a duck), since one’s attention would more likely have focused on the left side of the figure and decompose the figure in another way; the image perceived is determined by exogenous attention since the locus of attentional focus is determined by the moving image. If the perceiver possessed the concepts “rabbit” or “duck” then she would seedox a rabbit or a duck, respectively. In the case of endogenous attention, the activation of the perceptual set may explain the process by which perceptual set operates with bi-stable stimuli (stimuli that support two perceptual interpretations), in tasks in which a single stimulus is present and there is not a target object that must be selected amidst other distractors. Perceptual set facilitates one interpretation over another. Object recognition contributions contribute to determining perceptual set (Peterson and Gibson 1994); it is known that familiar objects have an advantage over nonfamiliar objects in object figure-ground segregation (Peterson 1994; Vecera and O’Reilly 2000) and in object attention (Vecera and Farah 1997). Using Desimone (1999) account of attention as biased competition 3 Although these studies examine the Necker cube, they can be extended to cover other reversible figures as well (Meng and Tong 2004, p. 540). 4 The qualification “more likely” is needed because motion may not be the only factor that determines

where attention will focus, although it is a very important factor.

123

500

Synthese (2011) 181:489–514

one can say that familiar objects stored in long term memory, when they become task relevant, activate cells in visual working memory that represent the familiar objects’ features, providing top-down feedback that enhances the activation of neurons in the visual cortex that respond to these objects giving them an edge in their competition against neuronal assemblies that respond to nonfamiliar objects. Other cognitive factors that may contribute to the perceptual set include volitional effects, learning effects, knowing of or familiarity with reversibility, and expectancy effects (Long and Toppino 2004). The finding that when multiple figures are simultaneously presented to different regions of the retina they are perceived to reverse (that is, they switch from one phenomenal content to another) independently suggests that a single global top-down processes is not necessary for figural reversal (Long and Toppino 2004). This entails that attention and perceptual learning are not necessary for reversals in perceiving ambiguous figures and, thus, weakens the plausibility of attentional-decisional theories. These studies underlie the necessary role of strictly localized conditions in perceiving ambiguous figures (Long and Toppino 2004). Such a localized condition could be that a single stimulus at a single retinal location cannot be seen simultaneously as two different objects, a suggestion that figures in Attneave (1971) account of the property of exclusivity, namely, that there is only one perceptual organization that can be experienced at any given time. (Note that this property cannot be adequately explained by attention-decisional theories.) This caveat notwithstanding, work with bi-stable stimuli (Attneave 1971; Driver and Baylis 1996; Hochberg and Peterson 1987; Peterson and Hochberg 1983) sheds light on the mechanisms that underlie the way perceptual set biases object segmentation. Their findings suggest that the cognitive states of the observer do not affect by themselves the organization of the stimulus. Some crucial points of fixation influence the organization of the stimulus. In other words, the way a bi-stable stimulus can be visually interpreted depends on where the observer fixes her attention, because there are crucial points fixation on which determines the perceptual interpretation. This means that the mechanism underlying the effect of perceptual set in ambiguous figures involves the voluntary control of spatial attention; the perceptual set induces observers to allocate their attention to specific regions in the stimulus (Peterson and Gibson 1994). This causes the figure to be experienced the way favored by perceptual set. If the cognitive effects affect perceptual organization of ambiguous figures through spatial attention’s role in focusing the eyes to specific regions in the figure, the cognitive effects influence the way the figure is perceived only in an indirect way. This is so because once spatial attention has focused on some region, information is retrieved from the visual scene by means of processes that are stimulus-driven and not cognitivedriven. Despite the role of cognitive effects in determining the perceptual organization of an ambiguous figure, there is always a passive, stimulus-driven, automatic process underlying the perception of ambiguous figures, which Hochberg and Peterson (1987) call “figural instability” (the cognitive factors’ effects are referred to as “figural malleability”). Now, if (a) cognitive factors affect the way ambiguous figures are experienced through the modulating effects of voluntary top-down attention; (b) voluntarily attentional control allows or “forces” in a top-down manner the subjects to focus on critical

123

Synthese (2011) 181:489–514

501

points or regions in the figure and (c) it is this focus that determine bottom-up the perceptual organization of the ambiguous figure, the cognitive effects, the effects of top-down voluntary spatial attention, and the bottom-up effects posited by the focal-feature hypothesis are largely coextensive. As Long and Toppino (2004, p. 756) remark, it is difficult to separate the cases in which volitional control affects perception of ambiguous figures from those of the focal-feature hypothesis in which different points on a figure seem to favor one of the perceptual organizations because the observer may facilitate reversals by changing the points on the figure to which focuses attention though volitional control. However, Toppino (2003) adduces evidence that voluntary control can be independent of processing crucial focal features. Thus, cognitive factors may directly control figural reversal bypassing spatial attention. I return to this possibility when I examine the factors that may determine the experience of Mach’s figure. We have examined the role of endogenous attention in disambiguating ambiguous figures. Endogenous attention is related to working memory. Working memory stores information and performs executive control governing retrieval, encoding, and commands for the expression of attention (Baddeley 1995). These two functions underlie the distinction in the attentional control processes between expectancy of an upcoming event and preparation for that event (LaBerge 2000). The expectation of an event is not necessarily accompanied by an attentional preparation for it. The top-down attentional control of perception amounts to the attentional expectation for an upcoming event, not necessarily to the preparation for that event. That is, cognitive factors determine the expectation for an event, but this is not sufficient for attentional preparation for that event; information regarding the upcoming display of an object may be kept in working memory while attention may be directed elsewhere. Attentional preparation follows expectation, if the event is task-relative. Suppose that a stationary duck/rabbit configuration is viewed by a perceiver X who possesses both concepts but who has for some reason or other activated the concept “rabbit” before the ambiguous figure appears. In other words, cognitive factors have created a context in which X is biased towards rabbits or equivalently X is in perceptual readiness for rabbits. This has resulted in an expectation in X’s spatial attention. Let us assume second, that X performs a task that is potentially related with rabbits. This leads X to be prepared for rabbits. This preparation combined with expectation means that spatial attention focuses onto points in space where previous experience indicates that, with an acceptable probability, the information contained at those points suffices to determine the presence of a rabbit. Such characteristic information may be the relative position of the ears with respect to the face, which depends on the orientation of the image. In the case of perceptual readiness for ducks, the characteristic information pertains to the relative position of the beak with respect to the face. When X is presented with a rabbit/duck ambiguous shape, then X, focusing her attention onto the location at which she expects the characteristic ears will see a rabbit. If X were expecting a duck, she would have focused elsewhere and would have seen a duck. Thus, when the ambiguous figure appears, the perceiver, because of the perceptual set, focuses on some part of the image and retrieves from it bottom-up, and therefore seesph , either a duck-like or a rabbit-like figure. This image is fed to higher cortical

123

502

Synthese (2011) 181:489–514

areas where working memory activation affects visual processing. At this stage, cognitively driven object-based attention intervenes and determines, depending on the kind of perceptual readiness, whether, upon seeingph a duck or a rabbit, one seesdox a duck or a rabbit, respectively. The above analysis shows in what sense the duck/rabbit configuration is perceptually ambiguous. The image that is formed during the first stages of perceptual processing is neither a duck nor a rabbit, but rather, a duck/rabbit configuration. There is no separation between a duck-like and a rabbit-like image during the first stages of visual processing, since how the image is eventually seen does not have any effects on early processing. Early on in perceptual processing, spatial attention, whether it be endogenously or exogenously driven, decomposes that figure in ways that the ensuing phenomenal content of perception is that of a duck-like or a rabbit-like figure. This takes place at areas V 4 and MT, in which shape is encoded, and which seem to be associated with the point of reversal from one perceptual interpretation of the ambiguous figure to the other. This is supported by Leopold and Logothetis (1996) research, which suggests that the perception of bi-stable figures is not determined during early visual processing, for example by suppression of monocular cells, but in higher visual areas, such as V 4 and IT, in which shape is encoded and in which attentional effects are first registered. The phenomenal content may eventually be conceptualized and the perceiver sees either as a duck or as a rabbit, respectively; the perceiver passes from representational states in which she seesph a duck-like or a rabbit-like figure to representational states of seeingdox a duck or a rabbit.

4.2 The diamond/square and the non-regular diamond/parallelogram figures I turn now to the square/diamond figure and the non-regular/parallelogram figure. Let me start by addressing the cases in which two figures are seen and the subjects, despite the fact that these figures have the same shape, see different figures. Thus, the lefthand figure in Fig. 1b is seenph as a diamond, whereas the right-hand figure is seenph as a square, and the left-hand figure in Fig. 2a is seenph as a non-regular diamond, whereas the right-hand figure is seenph as a parallelogram. If the problem consists in the different ways the two figures look, this is a problem of failure of shape constancy and not of a gestalt switch that has to be explained. Moreover, there is no ambiguity here. Each of these figures is usually reported as a diamond or as a square, and as a non-regular diamond and as parallelogram, respectively. The problem at hand is not that there are two different reports, one for each figure, but that the one figure is a tilted version of the other and yet it is reported as a different figure. This, however, is not an ambiguity regarding the perception of an image but failure of shape-constancy. In the diamond/square case, this failure of shape constancy may be caused by the fact that, as Ferrante et al. (1997) argue, people can accurately judge whether an angle is right or not when the angle is normal but they cannot do it as accurately when the angle is not normal. This phenomenon is attributed to the fact that the visual system is more sensitive to normal right angles than to non-normal right angles. Therefore, a figure with normal angles is represented differently from a figure with not normal angles. Thus, if this is Macpherson’s case, then representationalists have an easy task;

123

Synthese (2011) 181:489–514

503

it suffices to appeal to this different sensitivity that creates differences in the content represented in perception. With respect to the non-regular diamond and the parallelogram figure it is not the sensitivity to normal right angles that causes different representational contents to be had, but the sensitivity of the visual system to cardinal orientations (horizontal and vertical lines) as opposed to oblique lines. Appelle (1972) adduces evidence to the effect that for a plethora of perceptual tasks (such as, the resolution of targets, estimation of stimulus position, learning and discrimination of objects, etc.,) performance is superior for stimuli aligned in horizontal or vertical orientations (this is called the “oblique effect”). Appelle also argues that there are neurophysiological mechanisms for orientation analysis in the higher visual pathways that subserve the orientation preferences reported behaviorally. The oblique effect is supported by Foster and Ward (1991) studies, except that these authors locate the neurophysiological mechanisms of this effect in two orthogonal filters in early vision. The view that there is an early cortical locus for the oblique effect receives support from experiments with single-neuron electrophysiology (De Valois et al. 1982) and evoked-potential studies (Maffei and Campbell 1970). Furmanski and Engel (2000) fMRI study suggests that the oblique effect results from asymmetries between populations of V1 neurons, specifically from differences in the neural activity or the relative number of cardinal neurons. The oblique effect is probably absent from regions outside of V1. Finally, Boynton (2005) reports studies that use algorithms to overcome the difficulties associated with the fact that the spatial resolution of fMRI responses does not allow studying the activations of individual neurons. These studies recognized patterns of activation of neurons that form columns about 500 µm across large areas of V1 and confirmed the oblique effect. Kamitani and Tong (2005) were able to predict from the images in V1 the orientation the subject was thinking about. Haynes and Rees (2005) predicted the orientation of a stimulus the subjects could not see. These findings show the close correlation between the activation across columns in V1 neurons and the orientation of the stimulus. Representationalists can appeal to the differentiated sensitivity to cardinal orientations to account for different representational contents of the two figures and Macpherson’s argument that there are two phenomenal characters but only one NCC. I turn now to examining the diamond/square and the diamond/parallelogram ambiguous figures. The case that could poses a real threat to representationalism is that in which a single figure is seeingph into two different ways. Suppose that subjects view the square/diamond ambiguous figure in Fig. 1a. The image that is seeingph is a figure with a certain shape and a certain orientation (i.e., with non-normal right angles).5 The NCC that is involved in seeingph and is formed along the ventral visual pathway is framed in a relational coordinate system that is best captured by Cartesian coordinates, in which the directions top-down and right/left are determined with respect to the position of the body of the perceiver.6

5 Normal right angles are the angles whose lines are close to the horizontal and vertical axes. 6 Which part of the body is the center of reference of the coordinate system depends on the stage of per-

ceptual processing. It can be the middle point between the eyes, or the center of the head, when it comes to vision, or the felt direction of gravity when it other senses.

123

504

Synthese (2011) 181:489–514

a

b

Fig. 3 Mach’s figure cast in two differently oriented Cartesian frames

In the square/diamond figure, the application of the coordinate system bisects the angles of the figure. So, what is phenomenally perceived is something like the configuration in Fig. 3a. This phenomenal content is conceptualized as a diamond. There is evidence that subjects that are unfamiliar with the figure (and other ambiguous figures, such as the Necker cube) and who are provided with no instruction by the experimenters fail to experience or experience very little ambiguity (Rock et al. 1994). Since most all subjects report seeing a diamond, there is some controversy as to whether this figure is really an ambiguous figure. Recently, it is more common to call this figure “reversible”, that is, a figure the perception of which may change under certain circumstances (by rotating the figure, for instance). As we saw, spatial attention, either bottom-up image driven or top-down cognitively driven, can influence the way the figure is perceived. One could shift attention to the axes that bisect the sides of the square/diamond figure and then one could report seeing a square (see Fig. 3b). In this case, one applies a rotated Cartesian frame of reference to the figure. Such an attentional shift may be due to either exogenous or endogenous reason. A cue, for example, may appear at the middle of one side and cause attention to disengage from one of the corners and shift to that middle point. This is a case of exogenous attention. In endogenous or top-down voluntary spatial attention, attention mediates the effects of perceptual learning, expectancy, volitional factors, knowledge of reversibility or previous experience with reversible figures, and contextual factors that are all covered under the term “perceptual set”. The subject may be in perceptual readiness for squares and, as a result, decomposes the configuration in her visual field in a way that results in the perception of a square-like figure. With both kinds of attention, the ensuing phenomenal content is that of a square-like figure. Notice, first, that the rotated frame of reference cuts the figure differently from way the non-rotated frame of reference cuts the figure (see Fig. 3a and b, respectively) causing a different perceptual organization of the figure and, thus, emphasizing different properties of the figure (in the one case the figure is cut into two fours similar figures, whereas in the other is cut into four triangles).

123

Synthese (2011) 181:489–514

505

It is worth pointing out that the upright frame of reference and the rotated frame of reference act as the two types of axes of symmetry that Peacocke used to explain the different representational contents that can be associated with Mach’s figure. This shows that Peacocke’s account was on the right track, except that instead of construing them as axes of symmetry, which of course, they are, he should have construed them under their more general role as the axes of the frame of reference in which representational contents are cast. This would have allowed them to extend their use in the case of non-symmetrical figures, avoiding thus some of the objections raised by Macpherson. It is not certain how attentional effects cause what is usually perceived as a diamond to be perceived as a square, that is, how the cause a figure reversal. Research examining the effect of the presentation of the unambiguous version (the square) of the typical ambiguous figure prior to the presentation of the typical ambiguous figure suggests that for short presentations there is a positive-bias effect favoring the same perceptual interpretation of the subsequently presented ambiguous figure. In other words, if a square is presented for short periods of time (less than 5 s) followed by the presentation of Mach’s figure, the ambiguous figure is experienced as a square (Long and Toppino 2004). In this case, the prior presentation of the square creates a perceptual set favoring squares and it may be that when one comes across a figure that might be a square even though it does not look like one (the diamond-like figure that is being presented after the square), one rotates it to see whether it is a square that is rotated. Thus, probably the subjects perform a mental rotation that changes the orientation of the figure. This is an example of the role of voluntary action in imposing a certain perceptual organization that favors one or the other interpretation of the ambiguous figure. There is abundant evidence (Pylyshyn 1979; Shepard and Cooper 1982) that subjects under various conditions perform mental rotations in order to identify and classify figures. It is clear, then, that Ferrante et al. (1997) theory cannot explain why Mach’s figure is sometimes seen as a diamond and other times as a square, for Mach’s figure has non-normal right angles. Only when an account has been given why the figure is rotated, Ferrante’s et al., account can be brought to bear, because only if it is rotated 45◦ the Mach’s figure acquires normal angles. The phenomenal content of the experience of a square is different from the phenomenal content of the experience of a diamond in that the two figures have different orientations and the Cartesian framework cuts them differently. Since these attributes of the image are retrieved bottom-up from the scene, they are parts of NCC. Thus, the original figure and the rotated figure have different NCC. They also have different phenomenal contents, which is all that representationalism demands. A similar account can be given for the non-regular diamond/parallelogram ambiguous figure (see Fig. 2b). Emphasis could be given either to the angles or the sides of the figure, resulting in different orientations of the frame in which the figure is cast. As in the square/diamond figure so here the axes of reference cut the figures differently highlighting different properties that differentiate the corresponding nonconceptual representations. For example, as Macpherson (2006, p. 107) remarks “the experience corresponding to the first figure would represent an angle pointing in the direction of ‘up’. In the experience corresponding to the second figure, this would not be represented. Instead, two sides of the figure would be represented as being parallel to

123

506

Synthese (2011) 181:489–514

the up/down axes”. Or, in the former case four triangles are formed, whereas in the latter four quadrilaterals are formed. As with Mach’s figure so here, appealing to the oblique effect cannot explain by itself why a non-regular diamond can be seen as a parallelogram. Only by factoring in top-down or bottom-up spatial attention and the way it affects the orientation of the frame of reference in which the figure is cast can one explain why the figure is ambiguous. One might object that the above account of ambiguous figures is beset by a blatant inconsistency. For if phenomenal content is NCC and if, as I have argued, NCC is formed in a preattentive stage of visual processing, then how could top-down cognitively driven attentional shifts lead to shifts in phenomenal content? Once attention is in play, the content is conceptual not nonconceptual and, therefore, could not be phenomenal. Notice, first, Macpherson’s disregard of the role of attention in seeing ambiguous figures, even though the evidence I review here clearly shows its effects. This is particularly important in view of the fact that cognitively driven or top-down spatial attention clearly influences the perception of ambiguous figures. However, if it does, then how is this consistent with Macpherson’s view that cognitive or conceptual factors do not produce gestalt switches? Be that as it may, the same problem threatens my account as well. The answer to that is that although spatial attention affects perceptual processing very early, it does so in an indirect way that does not entail conceptual modulation of the perceptual processing itself, in that the content of the perceptual states is determined exclusively by the stimulus (it is stimulus driven) and is not cognitively driven by top-down flow of information (Raftopoulos 2009). However, we have seen that Toppino (2003) has made a case that voluntary attentional control of ambiguous figures may be independent of selectively processing particular focal features. This entails that cognitive factors control directly perceptual processing of ambiguous figures, that is, without being mediated by spatial attention. If this is true, then two things may happen. First, there is simply no seeingph in the case of ambiguous figures since cognitive factors determine what one perceives by fixing the perceptual organization of the figure. That is, if the perceptual processes involved in perceiving ambiguous figures are cognitively penetrable, then perception and cognition are continuous and there is not a stage of visual processing that is cognitively impenetrable. This entails that there is no NCC and that the phenomenal content of experience as construed by nonconceptual representationalists does not exist. Since both Macpherson and nonconceptualists talk about the NCC phenomenal content of the experience of ambiguous figures, if such content does not exist, they all loose. Second, there is an alternative explanation as to what happens when one sees a diamond like figure and reports seeing a square, which still leaves room for NCC. Suppose again that the subject is in perceptual readiness for (and searches the scene for) squares. The phenomenal content that corresponds to a diamond is first retrieved bottom-up from the scene on account of the nature of perceptual processing and the way the figure is oriented in space. When object-centered attention modulates the processing and the conceptual framework of the perceiver applies to the phenomenal content of perception, the object is identified as a square, as a result of the perceptual readiness for the class “square”. The subject seesdox a square even though she is seeingph a diamond; there is only one phenomenal content but different conceptualizations. In this case, there is no gestalt switch in seeing a diamond or a square, since the switch is not

123

Synthese (2011) 181:489–514

507

a change in phenomenal content but a change in doxastic content; it is a change in conceptual content not in NCC. If there is no change in phenomenal character but only different cases of doxastic seeing, then infants and animals (lacking concepts and thus unable to seedox ) should seeph the same figure. As I said in the introduction, the problem of which one of the two alternative accounts that I offered here is a more adequate account of the perception of ambiguous figures is a matter for empirical research to solve. If it could be shown that infants or animals do not undergo the gestalt switch upon viewing the square/diamond figure, which would mean that the figure is not ambiguous for them, then that would tip the scale in favor of the second account. Notice that in either case, that is, whether in perceiving the reversible figure one has either different NCC and different phenomenal contents or the same phenomenal content but different conceptualizations of that content, representationalism is out of the woods. Recall that for Macpherson’s arguments to succeed, the ambiguous figures must have different phenomenal contents and the same NCC. This is not the case in any of the above scenarios. This analysis implies a potential difference between the duck/rabbit and the square/diamond ambiguous figures. If it is possible in the latter figure that there be only one NCC content retrieved from the image (a diamond-like figure) and the ambiguity results from the conceptualization of that NCC as a square, then the square/diamond figure is not perceptually ambiguous, or at least is not ambiguous in the same sense in which the duck/rabbit figure is ambiguous. The duck/rabbit is ambiguous because it can be decomposed perceptually and being seeingph into two different ways. The square/diamond figure, on the other hand, can be seenph only in one way (as a diamond-like figure) but can be conceptualized differently (either as a diamond or a square).

5 Do ambiguous figures challenge representationalism? I will assess now Macpherson’s argument against representationalism. Macpherson’s point is that there is not a unified representationalist account of all ambiguous figures. Ferrante’s views may account for Mach figure, but they cannot explain the parallelogram case. The oblique effect (which Macpherson does not discuss) may account for the parallelogram case, and the extension of Ferrante’s strategy involving frames of reference and the way they cut the figure underlying different properties may account for the parallelogram case as well, but the latter fails to explain why the A and the tilted A are seenph the same way (as Macpherson thinks they do) despite the fact that they are oriented differently and the frame of reference cuts them differently. Since representationalists cannot explain why the frame of references does not influence the way the A and tilted A are perceived, it seems that frames of reference do not play the role that representationalists wish to assign to them. Hence, they cannot be used to explain the parallelogram case, which means that despite some initial successes, representationalism does not have a unified description of the ways ambiguous figures are perceived. In the previous section, I offered a detailed account of ambiguous figures and argued that in all these cases when the figures are seenph as having different phenom-

123

508

Synthese (2011) 181:489–514

enal content, different NCC are also being represented owing to differences in figure orientation and in the way the frame of reference, which all viewers impose on their perceptual contents, allows them to organize perceptually the image. However, for my argument to succeed, I must counteract two of Macpherson’s objections. The first concerns Macpherson’s (pp. 102–110) criticism of Peacocke’s argument that could also be used against the account based on frames of reference proposed here. Macpherson argues that for this account to work the two different representational contents that are formed in virtue of the application of the different frames of reference should be inconsistent, that is, they must be such that they cannot both fill the same scenario. Only then could one argue that the one content but not the other could be the NCC of experience. However, in the square/diamond figure the NCC that represents the symmetry about the axes is not inconsistent with the NCC that represents the symmetry about the angles and, therefore, there is no reason to assume that one cannot have an experience with both contents. Moreover, if the frames of reference or axes of symmetry determine the percept when one views the Mach figure, then how is it possible to see a square while focusing one’s attention on its angle bisector symmetry? The fact that this is possible means that the contents pertaining to both types of symmetry are present in experience and, thus, differences in the type of symmetry cannot explain the different phenomenal characters associated with the two percepts. Despite its initial attraction, Macpherson’s objection does not stand to scrutiny. It is true that imposing one frame of reference to the figure is not geometrically incompatible with imposing the other frame and, therefore, both representational properties that result from the different perceptual organizations of the figure could co-occur. However, one can focus one’s attention only to the sides of the figure or to its angles but not to both and, therefore, one can cast the figure only in one frame of reference. Given the nature of selective attention, the attended feature or location is enhanced and the alternative unwanted feature or location is suppressed. With regard to the second point, Macpherson chooses the wrong figure, because the square that she uses as an example is not an ambiguous figure (remember that Mach’s ambiguous figure is the diamond); in psychology, this is called the “unambiguous version of the figure”. Therefore, it is expected that focusing on the symmetry about the angles of the square does not change the way the figure is perceived. However, focusing on the symmetry around the sides of a diamond may change the way the figure is perceived. The second objection concerns the A/tilted A counterargument that was offered precisely to block the kind of theory that I have expounded. This is arguably the most important of Macpherson’s points in that it undermines the strongest and more empirically adequate representationalist attempt to account for the experience of ambiguous figures. If representationalists cannot appeal to the frames of reference and the way they cut the figure to explain differences in representational content, then they cannot account for the parallelogram case and, therefore, that ambiguous figure presents a real unsolved challenge for representationalism. Macpherson bases her argument on evidence that shows that subjects that view rotated A’s recognize them as A’s. This evidence, as Macpherson interprets it, shows that in this case the subjects have the same phenomenal experience as they have when they see a straight A. Macpherson (2006, p. 108) writes that an option available to representationalism would be to explain “why

123

Synthese (2011) 181:489–514 Fig. 4 The A and the tilted A cast in a normal Cartesian frame of reference

509

a

b

A the experiences [that is, the experience of the frame of reference cutting the figure] that represent different properties do not give rise to experiences with different phenomenal characters in non-ambiguous figures, but do so in the other ambiguous figure cases”. So, unless one could provide an account as to why the frames are applied to the ambiguous figure (the diamond/square) but not to the non-ambiguous figure (A /tilted A), the evidence suggests that the subjects do not apply such frames. Since representationalism has not provided such an account, its appeal to different frames of reference to account for the perception of ambiguity fails. My reply to that argument is simple. When subjects view an A and a tilted A (see Fig. 4), they have different phenomenal contents. The fact that they report seeing an A in both cases means only that they seedox an A; what they see is contaminated by semantic knowledge. Thus, Macpherson confuses seeingdox with seeingph . Let me explain why. Macpherson (2006, p. 91) admits that subjects judge the tilted figure to be a tilted A, but then she adds that this does not mean that “one undergoes the distinctive gestalt switching phenomenon, which one experiences looking at the left hand figure (the square/diamond figure)—nothing one does can make one undergo a Gestalt switch with respect to the A.” Comparing the case of the A/tilted A with that of the diamond/square, she argues that in the latter case one undergoes a gestalt switch, whereas in the former one does not. But why is that so? One could argue that the subjects’ report that they see a tilted A in the one case and an A in the other case may be taken as evidence that the subjects perceive two different phenomenal contents but lacking a concept for the tilted A as a different letter they describe their experience as that of an A. Consider the analogous case in which one tries to explain the difference between two shades of blue of which one just had a phenomenal experience by taking recourse to descriptions like, “the one blue was darker than the other blue”. The fact that both descriptions involve the concept “blue” does not imply that the subject seesph the same shades. Similarly, the characterization “tilted A” as opposed to “A” may signify the two different phenomenal contents despite the subjects’ report. Why does not Macpherson consider that possibility? I think the answer is that Macpherson takes the subjects’ reporting an A that is tilted when they see a tilted A as evidence that they seeph an A despite its being tilted and that is why she mentions shape constancy with regard to that experiment. In other words, the fact that the letter in both cases is reported as an A, albeit as a tilted one in one of the cases, shows that the same phenomenal content is involved. To neutralize the objection that since the subjects report seeing an A and a tilted A this may mean that they seeph two different figures, Macpherson offers the

123

510

Synthese (2011) 181:489–514

rejoinder that the fact that the tilted A is described as a “tilted A” as opposed to an “A” is the result of a judgment and not of a gestalt switch. Against this, representationalism’s thesis that phenomenal content is NCC suggests that the phenomenal awareness of an A is different from the phenomenal awareness of a tilted A, simply because the properties that are retrieved bottom-up when an A is present are different from the properties that are similarly retrieved when a tilted A is present. The orientations of the figures are different and the frame of reference in which NCC is cast cuts the figures differently underlying different properties and making, thus, their perception different (see Fig. 4). What happens is that the subjects, in order to report what they see, must appeal to their conceptual repertory regarding letters. Thus, the task shows what they seedox and not what they seeph . The tilted A is recognized and described by subjects as an A because there is a letter-concept that describes A’s but no letter-concept that describes tilted A’s. Imagine a civilization that has both an A and a rotated A as letters in its alphabet and calls the former “alpha” and the latter “kappa”. The subjects from that civilization in the same experiment would see first an alpha and then a kappa. Or, to reverse the case, suppose a civilization that does not have the term “square” in its conceptual vocabulary, and suppose that subjects from that civilization perform the experiment in which a diamond-like figure is rotated. Lacking the concept square, they would describe their experience as that of a tilted diamond and they would identify the figure as a diamond. Are they are aware of the same phenomenal content? Provided that they have the same perceptual system as earth-subjects, the answer is negative.

6 Conclusion I have claimed that the way ambiguous or bi-stable figures are perceived is determined by the locus of focus of attention. The way a bi-stable stimulus can be perceptually interpreted depends on where the observer fixes her attention, because there are in the figure crucial points fixation on which determines the perceptual interpretation, that is the ways the objects are seeingdox . Spatial attention intervenes early during perceptual processing to determine where a search for a figure or a search within an image will take place. However, attention does not modulate perceptual processing and thus the cognitive content that drives it does not enter into the content of perceptual states. The figure is first organized in a certain way and one of the two possible images is processed and phenomenally perceived; the perceiver is phenomenally aware either of a rabbit or a duck, or of a diamond or a square-like figure. Then, the phenomenal content is embedded in a conceptual framework and the figure is identified and categorized as the member of a class of objects, and the viewer acquires access to that content, in the sense that she can employ concepts to bear on it. When the viewer is presented with two figures, say a square and a diamond, or when a square is presented first and then is tilted, then the viewer has two different phenomenal contents, a diamond-like and a square-like figure, but these two phenomenal contents are easily accounted for by differences in NCC, which are established by considering the information that is retrieved bottom-up from the scene. The two figures have a different orientation in space and the Cartesian coordinate system in which

123

Synthese (2011) 181:489–514

511

NCC is represented cuts them differently bringing to the fore different properties that the same figure has in relation to that frame of reference. When an ambiguous figure is presented, there are two possibilities. First, one may seeph the phenomenal content that corresponds to a diamond-like figure and report seeingdox a diamond, or, because of the role of perceptual set, one performs a mental rotation of the diamond and then one seesph a square-like figure. However, in this case, one also perceives a different NCC. Second, it may be that one seesph a diamond-like figure but identifies the figure as a square because of the role of object-based attention that imposes one’s conceptual framework on visual processing and makes one seedox a square. In this case, there is a unique phenomenal content and there is no problem for representationalism either. The distinction between a conceptually encapsulated visual processing (seeingph ) and a conceptually penetrated visual processing (seeingdox ) is crucial for in my account of ambiguous figure. Phenomenal content is formed in the first stage. It is interesting that Macpherson (2006, p. 110) comes close to the account of ambiguous figures presented in this paper by arguing that the fact that one could have an experience of a square and, therefore, represent symmetry about the bisectors of the sides, while at the same time paying attention to, and thereby representing information about, the symmetry about the bisectors of the angles suggests that one should distinguish between layers of content in experience. “The crucial factor that determines how an ambiguous figure looks would not have its influence simply in virtue of whether or not it is represented in experience. Rather, there might be a difference between experiences of the same ambiguous figure according to which content is being attended to or focused on and which content is merely implicit in one’s experience (that is, which content one’s experience has that does not depend for attention on its existence)”. Macpherson distinguishes between a bottom-up content of experience and a content of experience that is formed by top-down cognitive influences that through attention modulate perceptual processing. Since NCC is content that is retrieved bottom-up from a visual scene and conceptual content is content that is cognitively modulated, Macpherson distinguishes in effect between NCC and conceptual content of experience. She agrees that spatial attention plays a role in how an ambiguous figure is experienced and that an equally important factor in determining how an ambiguous figure looks is the implicit content of an experience, that is, content that does not depend on attention. This is, indeed, the basis of my account of ambiguous figures. The reason that Macpherson and I differ can be uncovered in her statement that “the crucial factor that determines how an ambiguous figure looks would not have its influence simply in virtue of whether or not it is represented in experience”. What Macpherson means is that since the way an ambiguous figure looks depends in part on the implicit content of experience there is a factor that influences the way the figure looks that is not represented in experience. This is the main reason behind her conclusion that representationalism cannot account for ambiguous figures; representationalism deals with content that is represented. However, the perception of ambiguous figures partly depends on content that is implicit and not represented. The argument is sound only if implicit content (content that does not depend on attention for its existence) is not represented. This is exactly where Macpherson and I disagree. As I have argued here and elsewhere (Raftopoulos 2009; Raftopoulos and Muller 2006),

123

512

Synthese (2011) 181:489–514

the content that is retrieved bottom-up from a scene before the onset of attention is representational NCC. Furthermore, the representational content of one’s perceptual states is independent of what one reports seeing. Explicit reports related to descriptions of objects in a scene provide a poor measure of the detail of the scene representation in that explicit reports are not sensitive to the presence of all information that is perceived and represented. To draw conclusions about the nature of representations based exclusively on subjective reports assumes a one-to-one relation between awareness and attention that simply does not exist, as studies on implicit perception (Dehaene et al. 1998a,b; Evans et al. 2000; Koivisto and Revonsuo 2004; Merikle et al. 2001) and on change blindness (Fernandez-Duque et al. 2003; Hollingworth and Henderson 2004) suggest. Even explicit reports of the percept may not be an exhaustive measure of all explicit contributions to perception. Paraphrasing Block (2007, p. 354), one could say that the subjects’ reports provide us with access to what they access in their experiential content and not access to the experiential content of their states. Observers’ failing to report some object, property, or change does not mean that they lack explicit information about them (Mitroff et al. 2002). Subjects may be aware of the object, property, or change even though they do not report it because their awareness is below some threshold criterion (Simons and Silverman 2004).

References Appelle, S. (1972). Perception and discrimination as a function of stimulus orientation: The “oblique effect” in man and animals. Psychological Bulletin, 78, 266–278. Attneave, F. (1971). Multistability in perception. Scientific American, 225, 63–71. Baddeley, A. (1995). Working memory. In M. S. Gazzaniga (Ed.), The cognitive neurosciences. Cambridge, MA: The MIT Press. Bermudez, J. L. (1995). Nonconceptual content: From perceptual experience to subpersonal computational states. Mind and Language, 10(4), 333–369. Block, N. (2007). Two neural correlates of consciousness. In Consciousness, function, and representation, collected papers (Vol. 1). Cambridge, MA: The MIT Press. Boynton, G. M. (2005). Imaging orientation selectivity: Decoding conscious perception in V1. Nature Neuroscience, 8(5), 541–542. Campbell, J. (2002). Reference and consciousness. Oxford: Clarendon Press. De Valois, R. L., Yund, E. W., & Hepler, N. (1982). The orientation and direction selectivity of cells in macaque visual cortex. Vision Research, 22, 531–544. Dehaene, S., Kerszberg, M., & Changeux, J. P. (1998a). A neuronal model of global workspace in effortful cognitive tasks. Proceedings of the National Academy of Science of the USA, 95, 14529–14534. Dehaene, S., Naccashe, L., Le’ Clec’H, G., Koechlin, E., Mueller, M., Dehaene-Lambertz, G., van de Moortele, P.-F., & Le Bihan, D. (1998b). Imaging unconscious semantic priming. Nature, 395, 597–600. Desimone, R. (1999). Visual attention mediated by biased competition in extrastriate visual cortex. In G. W. Humphreys, J. Duncan, & A. Treisman (Eds.), Attention, space and action: Studies in cognitive neuroscience. Oxford: Oxford University Press. Dretske, F. (1997). Naturalizing the mind. Cambridge, MA: MIT University Press. Driver, J., & Baylis, G. S. (1996). Eye-assignment and figure-ground segregation in short-term visual matching. Cognitive Psychology, 31, 248–306. Evans, M. A., Shedden, J. M., Hevenor, S. J., & Hahn, M. C. (2000). The effect of variability of unattended information on global and local processing: Evidence from lateralization at early stages of processing. Neurophysiologia, 38, 225–239.

123

Synthese (2011) 181:489–514

513

Fernandez-Duque, D., Grossi, G., Thornton, I. M., & Neville, H. J. (2003). Representation of change: Separate electrophysiological markers of attention, awareness, and implicit processing. Journal of Cognitive Neuroscience, 15(4), 491–507. Ferrante, D., Gerbino, D., & Rock, I. (1997). The right angle. In I. Rock (Ed.), Indirect perception. Cambridge, MA: The MIT Press. Foster, D. H., & Ward, P. A. (1991). Asymmetries in oriented-line detection indicate two orthogonal filters in early vision. Proceedings of the Royal Society B, 243, 75–81. Furmanski, C. S., & Engel, S. A. (2000). An oblique effect in human primary visual cortex. Nature Neuroscience, 3, 535–536. Goodale, M. A., & Milner, D. A. (1992). Separate visual pathways for perception and action. Trends in Neuroscience, 15, 20–25. Haynes, J.-D., & Rees, G. (2005). Predicting the stream of consciousness from activity in human visual cortex. Current Biology, 15(14), 1301–1307. Hochberg, J., & Peterson, M. A. (1987). Piecemeal organization and cognitive components in object perception. Journal of Experimental Psychology: General, 116, 370–380. Hollingworth, A., & Henderson, J. M. (2004). Sustained change blindness to incremental scene rotation: A dissociation between explicit change detection and visual memory. Perception and Psychophysics, 66(5), 800–807. Jackson, F. (1977). Perception. Cambridge: Cambridge University Press. Jacob, P., & Jeannerod, M. (2003). Ways of seeing: The scope and limits of visual cognition. New York: Oxford University Press. Kamitani, Y., & Tong, F. (2005). Decoding the visual and subjective contents of the human brain. Nature Neuroscience, 8, 679–685. Kawabata, N. (1986). Attention and depth perception. Perception, 15, 563–572. Kawabata, N., Yamagami, K., & Noaki, M. (1978). Visual fixation points and depth perception. Vision Research, 18, 853–854. Koivisto, M., & Revonsuo, A. (2004). Preconscious analysis of global structure: Evidence from masked priming. Visual Cognition, 11(1), 105–127. LaBerge, D. (2000). Networks of attention. In M. S. Gazzaniga (Ed.), The new cognitive neurosciences (2nd ed.). Cambridge, MA: The MIT Press. Leopold, D. A., & Logothetis, N. K. (1996). Activity changes in early visual cortex reflect monkeys percepts during binocular rivalry. Nature, 379, 549–554. Long, G. M., & Toppino, Th. C. (2004). Enduring interest in perceptual ambiguity: Alternating views of reversible figures. Psychological Bulletin, 130(5), 748–768. Macpherson, F. (2006). Ambiguous figures and the content of experience. Nous, 40(1), 82–117. Maffei, L., & Campbell, F. W. (1970). Neurophysiological localization of the vertical and horizontal visual coordinates in man. Science, 167, 386–387. Magnuson, S. (1970). Reversibility of perspective in normal and stabilized viewing. Scandinavian Journal of Psychology, 11, 153–156. Matthen, M. (2005). Seeing, doing, and knowing: A philosophical theory of sense-perception. Oxford: Clarendon Press. Meng, M., & Tong, F. (2004). Can attention selectively bias bistable perception? Differences between binocular rivalry and ambiguous figures. Journal of Vision, 4, 539–551. Merikle, P. M., Smilek, D., & Eastwood, J. D. (2001). Perception without awareness: Perspectives from cognitive psychology. Cognition, 79, 115–134. Mitroff, S., Simons, D. J., & Franconeri, S. L. (2002). The siren song of implicit change detection. Journal of Experimental Psychology: Human Perception and Performance, 28(4), 798–815. Palmer, S. E. (1992). Modern theories of gestalt perception. In G. W. Humphreys (Ed.), Understanding vision. Cambridge, MA: Blackwell. Peacocke, C. (1992). A study of concepts. Cambridge, MA: The MIT Press. Peterson, M. A. (1994). Object recognition processes can and do operate before figure-ground organization. Current Directions in Psychological Science, 3, 105–111. Peterson, M. A., & Hochberg, J. (1983). Opposed set-measurements procedure: A quantitative analysis of the role of local cues and intention in form perception. Journal of Experimental Psychology: Human Perception and Performance, 9, 183–193. Peterson, M. A., & Gibson, B. S. (1994). Object recognition contributions to figure-ground organization: Operations and outlines and subjective contours. Psychological Science, 5, 253–259.

123

514

Synthese (2011) 181:489–514

Pylyshyn, Z. (1979). The rate of mental rotation of images: A test of a holistic analogue hypothesis. Memory and Cognition, 7, 19–28. Pylyshyn, Z. (2003). Seeing and visualizing. Cambridge, MA: The MIT Press. Raftopoulos, A. (2009). Perception and cognition: How do psychology and the neural sciences inform philosophy. Cambridge, MA: The MIT Press. Raftopoulos, A., & Muller, V. (2006). The phenomenal content of experience. Mind and Language, 27(2), 187–219. Rock, I., Halls, S., & Davis, J. (1994). Why do ambiguous figures reverse?. Acta Psychologica, 87, 33–59. Shepard, R. N. & Cooper, L. A. (Eds.). (1982). Mental images and their transformations. Cambridge, MA: The MIT Press. Simons, D. J., & Silverman, M. (2004). Neural and behavioral measures of change detection. In L. M. Chalupa & J. S. Werner (Eds.), The visual neurosciences. Cambridge, MA: The MIT Press. Suzuki, S., & Peterson, M. A. (2000). Multiplicative effects of intention on the perception of bistable apparent motion. Psychological Science, 11, 202–209. Toppino, T. (2003). Reversible-figure perception. Perception and Psychophysics, 65, 1285–1295. Tye, M. (2000). Consciousness, color, and content. Cambridge, MA: MIT Press. Tye, M. (2002). Visual qualia and visual content revisited. In D. Chalmers (Ed.), Philosophy of mind. Oxford: Oxford University Press. Tye, M. (2005). Nonconceptual content, richness and fineness of grain. In T. Gendler & J. Hawthorne (Eds.), Perceptual experience. Oxford: Oxford University Press. Vecera, P., & Farah, J. (1997). Is visual image segmentation a bottom-up or an interactive process?. Perception and Psychophysics, 59, 1280–1296. Vecera, P., & O’Reilly, R. C. (2000). Graded effects in hierarchical figure-ground organization. Journal of Experimental Psychology, 26, 1221–1231.

123

Synthese (2011) 181:515–529 DOI 10.1007/s11229-010-9744-0

Knowledge without credit, exhibit 4: extended cognition Krist Vaesen

Received: 19 June 2009 / Accepted: 25 March 2010 / Published online: 8 April 2010 © The Author(s) 2010. This article is published with open access at Springerlink.com

Abstract The Credit Theory of Knowledge (CTK)—as expressed by such figures as John Greco, Wayne Riggs, and Ernest Sosa—holds that knowing that p implies deserving epistemic credit for truly believing that p. Opponents have presented three sorts of counterexamples to CTK: S might know that p without deserving credit in cases of (1) innate knowledge (Lackey, Kvanvig); (2) testimonial knowledge (Lackey); or (3) perceptual knowledge (Pritchard). The arguments of Lackey, Kvanvig and Pritchard, however, are effective only in so far as one is willing to accept a set of controversial background assumptions (for instance, that innate knowledge exists or that doxastic voluntarism is wrong). In this paper I mount a fourth argument against CTK, that doesn’t rest on any such controversial premise, and therefore should convince a much wider audience. In particular, I show that in cases of extended cognition (very broadly conceived), the most salient feature explaining S’s believing the truth regarding p may well be external to S, that is, it might be a feature of S’s (non-human, artifactual) environment. If so, the cognitive achievement of knowing that p is not (or only marginally) creditable to S, and hence, CTK is false. Keywords

Knowledge · Credit · Extended cognition

1 Introduction Consider a lousy archer, called Franz. Franz places a lousy shot, but, due to the fortunate interference of a strong western wind, hits the boar nonetheless. The boar is brought down through luck, rather than through Franz’ skill, so although the shot is attributable to him, the success is not.

K. Vaesen (B) Eindhoven University of Technology, Eindhoven, The Netherlands e-mail: [email protected]

123

516

Synthese (2011) 181:515–529

This is a fairly typical way of framing what is special about knowledge. Knowledge is like hitting the boar through skill rather than through luck; it is believing the truth because of the correct application of one’s own cognitive abilities. So we would not attribute knowledge to someone in case she holds a true belief that p, but her truly believing that p is explained by, say, the environment accidentally making her belief true. Stated otherwise: to know that p is to deserve credit for believing the truth regarding p—voilà, the Credit Theory of Knowledge (CTK) in its most basic form. One can find it in, among other places, Greco (2007),1 Riggs (2007)2 and Sosa (2007).3 The strategy of those who oppose CTK is straightforward: find an instance of knowledge in which the credit condition isn’t met. Three such counterexamplars have been suggested. Both Lackey (2007) and Kvanvig (2009) argue that innate knowledge (at least: if it exists) would effectively undermine CTK—and indeed, prima facie it seems pretty awkward to credit someone for the innate beliefs she happens (or better: would happen) to have. Second, there is testimonial knowledge (see Lackey 2007, 2009). Lackey argues that one can acquire knowledge by, so to speak, passively absorbing what one is told; if so, credit should go, if anywhere, to the testifier rather than to her audience. And third, Pritchard (2005a) has expressed his doubts regarding credit theories in simple cases of perceptual knowledge: since basic perception is not within one’s immediate control, speaking of creditworthiness seems problematic at least. Each of these counterarguments has its own problems, though. First, the argument of innate knowledge obviously is compelling in so far as (one believes that) such knowledge exists—a controversial conjecture, to say the least. Moreover, even if one is willing to concede the existence of innate knowledge, the CTK-theorist can easily downplay its significance; Riggs (2009), for example, says to have no problem “limiting” his brand of CTK to the most common cases of knowledge, namely those involving empirical (as opposed to innate) knowledge. The same author also questions the effectiveness of Lackey’s case of testimonial knowledge. In Lackey’s scenario, a person called Morris, who has just arrived at Chicago train station, forms a true belief about the whereabouts of the Sears Tower, after having asked the first adult passerby. Lackey argues that while Morris knows the location of the Sears Tower, he doesn’t deserve epistemic credit for his believing the truth; hence, knowledge and credit do not always go hand-in-hand. Riggs, now, simply doubts that Morris has knowledge regarding the Sears Tower (ibid., p. 11): ‘Why on earth would we say that Morris knows where the tower is when he has picked a stranger at random, and unhesitatingly (and, one assumes, unreflectively) accepted what that person said? On the face of it, this is terrible epistemic practice.’ The dilemma Lackey faces is this. On one hand, she may raise the standards for testimonial knowledge. She could assume that Morris doesn’t simply open up his brain, but does some extra cognitive work; Morris, for instance, might select not the first passerby, but walk to a tourist office and ask for information there. If so, the 1 Greco (p. 57): “[…] knowledge attributions can be understood as credit attributions: when we say that someone knows something, we credit them for getting it right.” 2 Riggs (p. 329): “[…] knowledge is, complications aside, credit-worthy true believing.” 3 Sosa (p. 92): “[b]elief amounts to knowledge when apt: that is to say, when its correctness is attributable

to a competence exercised in appropriate conditions.”

123

Synthese (2011) 181:515–529

517

CTK-theorist presumably accepts that Morris through testimony comes to know about the Sears Tower; the problem now is, however, that Morris’ discriminatory behavior (i.e. carefully selecting a reliable testimonial source) makes him deserve credit again. Consequently, the effectiveness of Lackey’s case against CTK strongly depends on how high one sets the standards for knowledge. My intuitions, for what it’s worth, tend more toward Morris indeed acquiring knowledge; but for a strong, generic argument against CTK, a scenario indisputably involving knowledge would be helpful. Third, according to Pritchard’s perceptual knowledge argument, our lack of control over our perceptual (and many other) beliefs is at odds with us deserving credit for them. We seem to get such ideas “willy nilly”, they are hardly our cognitive achievement. In light of this, the argument goes, CTK-theorists either should admit that perceptual knowledge is illusory, or give up their credit condition. Unfortunately, as Riggs (2007) notes, quite a number of philosophers have argued a case for exactly the opposite; they contend that people do have sufficient control over their beliefs to be rightly attributed credit.4 These defenders of what is called “doxastic voluntarism” will not be impressed by Pritchard’s charges. So unless the dispute between voluntarists and anti-voluntarists is settled first, the prospects for a refutation of CTK that is palatable for all (or most) look dim. In sum, what Kvanvig’s, Lackey’s, and Pritchard’s counterarguments have in common is that they can perform as intended, only when some extra, contested assumptions are made—respectively, about the existence of innate knowledge, about the standards for knowledge, and about our lack of doxastic control. And that is, obviously, their weakness. In this paper I develop a fourth line of argument against CTK, the efficacy of which, I think, isn’t conditional on any such contestable premise(s). It concerns knowledge acquired through extended forms of cognition. Those familiar with recent discussions in the philosophy of mind might at this point dismiss my argument already, for to many the hypothesis of extended cognition is contestable par excellence. But to reassure the reader: the brand of extended cognition my argument relies on is, as will become apparent, quite weak, hence easily digestible. Basically, the only thing one needs to accept is that humans may use cognitive aids to produce cognitive outputs, that we may acquire knowledge by putting to work simple things like glasses, thermometers and computers. Indeed, my argument works whenever one is willing to subscribe to this quite trivial claim. The baseline of my argument then will be this: S can come to know that p, even when the most salient feature explaining S’s true belief is a feature of S’s (cognitive) artifactual environment. But whereas in Gettier-style cases the environment’s contribution is a fortunate coincidence, it is intended (non-lucky) in the scenarios I have in mind. As a rough-and-ready preview to my argument, let me draw a parallel to archery again. Suppose Franz, our lousy archer, hits a target, not through skill, but thanks to a bow-and-arrow set, neatly installed on a tripod, perfectly aimed at the target,

4 Riggs refers to Hieronymi (2006), Raz (1999), Adler (2002), Owens (2000), and Audi (2001).

123

518

Synthese (2011) 181:515–529

and equipped with a simple “shoot”-button. Knowledge might be like this; hitting the target, while all relevant cognitive work was delegated to external aids. 2 What epistemic credit is (supposed to be) Let’s sort out some terminological issues first. What exactly does the CTK stand for? What does it mean to say that knowledge is credit-worthy true belief? In answering these questions, I will restrict myself to two authors, John Greco and Wayne Riggs, since they both are quite explicit on this score. My case will be particularly strong if I succeed in refuting both CTKGr eco and CTK Riggs . Take CTKGr eco first. According to Greco (2003), S knows that p, iff S deserves intellectual credit for believing the truth regarding p. Intellectual credit, in turn, requires that: (G.1) believing the truth that p has intellectual value; (G.2) believing the truth regarding p can be ascribed to S; (G.3) S’s reliable cognitive character is an important necessary part of the total set of causal factors that give rise to S’s believing the truth that p. Condition (G.3) is crucial, since it is here that cognitive skill is introduced so as to separate the wheat (knowledge) from the lucky chaff (accidentally true belief). Lackey (2007) points out, though, that Greco doesn’t endorse so much (G.3) as (G.3*): (G.3*) S’s reliable cognitive character is the most salient part of the total set of causal factors that give rise to S’s believing the truth that p. The reason for that is simple. In Gettier-type cases, such as Chisholm’s dog-lookinglike-a-sheep scenario,5 a necessary part of the total set of causal factors giving rise to S’s truly believing that p still is part of S’s cognitive character, namely: S relying on her perceptual faculties to form the belief that p. To deny S knowledge in Gettier-style scenarios, Greco himself writes that S’s cognitive character ‘is [in such cases] not the most salient part [of the total set of causal factors that give rise to S’s believing the truth that p] (Greco 2003, p. 130, italics mine)’. Hence the plausibility of (G.3*). Lackey raises another interesting issue. How should we understand cognitive character in (G.3*)? Does it simply refer to the cognitive faculties S actually applied to form the belief that p or is it something more stable, a trait of S which is typically revealed when she is cognizing? In the latter case, it is particularly hard for a lousy thinker to come to know something extraordinary, that is, something unexpected given her stable (and lousy) cognitive character. Therefore, Lackey suggests the following amendment: (G.3**) S’s reliable cognitive faculties are the most salient part of the total set of causal factors that give rise to S’s believing the truth that p.

5 The scenario is as follows. S forms the belief that there is a sheep in the field, based on her seeing a

sheep-looking dog. S belief happens to be true, though. There actually is a sheep in the field, invisible to S, grazing behind a rock.

123

Synthese (2011) 181:515–529

519

What (G.3**) is supposed to do, thus, is to get into focus the actual cognitive faculties applied by S; it is these that should be decisive in (withholding) attributions of credit, not S’s track record of cognitive achievements. So if a lousy thinker would “reliably” stumble over a brilliant discovery, she would deserve credit (and thus the stamp of knowledge), even though the discovery wouldn’t be representative for her typical cognitive bent. In the remainder, whenever referring to CTKGr eco , I will assume (G.3**) rather than (G.3) or (G.3*), since (G.3**) seems the most convincing interpretation available. Lowering the stakes—that is adopting (G.3) or (G.3*)—would, I suppose, make the rest of my argument less compelling. Then there is CTK Riggs —which, according to the author (Riggs 2009), differs in important respects from CTKGr eco . CTK Riggs is basically an “anti-luck” theory of knowledge. For Riggs, saying that something is due to luck means it is not attributable to the beneficiary (or victim) of the luck in question. Credit, as a cognate of attributability, then ‘is simply a shorthand for saying that some event, state of affairs, or consequence thereof is attributable to an agent, as an agent (ibid., p. 3).’ Applied to matters of knowledge: S knows that p only if S’s holding the true belief that p is attributable to her as a cognitive agent. But to what kinds of luck is attributability supposed to be antithetical? Riggs discerns two types. First, there is the veritic luck of Gettier-style cases (such as Chisholm’s dog-looking-like-a-sheep scenario, see footnote 5). In such cases, S’s coming to hold a true belief that p is not ‘the product of S’s actual [cognitive] abilities (Riggs 2007, p. 335)’, and hence, S does not really know that p. Second, attributability as anti-luck implies that S coming to hold a true belief that p wasn’t accidental, in the sense of being (sufficiently) caused by S’s intention to have true beliefs. Riggs (1998) gives the following example. Suppose I have patent evidence that my business partner is guilty of fraud. Yet, because I have deep inhibitions against becoming involved in conflicts, I go to great mental lengths to explain away the evidence and to believe my partner is innocent. In fact, my partner really is innocent; he is just being framed by someone else. So while my belief about my partner is true, it doesn’t qualify as knowledge, since it is true accidentally. My actions aren’t guided by my intention (or better: desire) to have true beliefs any longer, but by another desire (viz. the desire to avoid confrontation). In such cases, my cognitive success is, as Riggs puts it, inadvertent. Putting this together, Riggs thinks that (ibid., p. 335): S knows that p iff S holds the true belief that p and: (R.1) S’s coming to hold a true belief in this instance is the product of S’s actual abilities; and (R.2) S’s coming to hold a true belief in this instance is not inadvertent. Given this formulation, the difference between CTKGr eco and CTK Riggs appears pretty small. After all, (R.1) is a statement concerning causal efficacy: we want ‘the outcome [i.e. S coming to hold the true belief that p] [to be] causally due to the agent’s abilities (ibid., p. 335)’. Or more in line with Greco’s terminology: the most salient causal feature explaining the true belief must be S’s cognitive faculties. The

123

520

Synthese (2011) 181:515–529

fact that Riggs talks about actual abilities makes (R.1) indistinguishable perhaps not from (G.3) or (G.3*), but quite plausibly from (G.3**).6 Regarding (R.2): indeed, such a claim is absent in CTKGr eco , but two points in response. First of all, as Riggs admits, it is quite hard to articulate what “inadvertently coming to form a true belief that p” actually amounts to. Second, even if one could spell out condition (R.2) in sufficient detail, I can allow myself to focus on condition (R.1) alone; given the biconditional (“iff”), CTK Riggs is false just in case one of its conditions doesn’t hold. This is precisely what I will show in subsequent sections. To recap, CTKGr eco and CTK Riggs share one intuition, that can be formulated either as (G.3**) or as (R.1). As such, it suffices to rebut that shared intuition in order to rebut both CTKGr eco and CTK Riggs . Since Greco was the first to propose a theory of knowledge in terms of credit-worthy belief and for purposes of simplicity, I will adopt Greco’s terminology and work only with (G.3**). One should bear in mind, however, that my criticism applies just as much to (R.1), and thus to CTK Riggs .

3 Credit and extended cognition: first pass In this section and the next, I introduce two examples of what one may call extended cognition (e-cog), a thesis proposed fairly recently in the philosophy of mind. According to proponents of e-cog (Andy Clark, Mark Rowlands, John Haugeland, to name but a few), cognitive processes may take place outside the boundaries of the human skin—extending into the subject’s environment—implying that the human skin doesn’t delimit the thinking subject. One prototypical example (given by Clark and Chalmers 1998) is Otto, an Alzheimer patient who uses a notebook as a substitute for his failing memory. Otto relies on his notebook in the same way as a healthy person relies on her internal memory; given this functional equivalence between internal (biological) memory and external (artifactual) memory, Clark and Chalmers claim that there is no principled reason to consider the former as a cognitive entity, but not the latter: If, as we confront some task, a part of the world functions as a process which, were it to go on in the head, we would have no hesitation in accepting as part of the cognitive process, then that part of the world is […] part of the cognitive process. (Clark and Chalmers 1998, p. 8) Much has been said about the plausibility of e-cog. One obvious problem is conceptual: despite functional equivalence, external processes do not seem to meet traditional criteria of the cognitive. Adams and Aizawa (2001), for instance, argue that real cognitive processes involve non-derived content (which e-cog processes don’t), and that the causal mechanisms underlying external and internal processes are too different to form a cognitive kind.7 6 Put otherwise, I don’t think that the difference between Greco and Riggs is a salience requirement (as Riggs suggests, p. 1); in my opinion, causal salience and efficacy can be construed as equivalent. The difference rather is (i) in typical (Greco) versus actual (Riggs) applications of cognitive abilities as a litmus test for credit, and (ii) in consideration (R.2). 7 For another much cited critique of e-cog, see Rupert (2004).

123

Synthese (2011) 181:515–529

521

These are thorny issues indeed,8 but whatever we decide the mark of the cognitive to be, e-cog contains a fairly uncontroversial (but in epistemology manifestly underplayed)9 part: the fact that human cognition is strongly dependent on external resources (whether or not we call them cognitive). Some features of the world actively scaffold us in our cognitive endeavors and as such are causally relevant to the kinds of beliefs we happen to have. And as long as CTK-theorists recognize this, my argument will appear effective. The ultralight version of e-cog I will exploit, thus, is supposed to be attractive to a wide audience and sidesteps the conceptual morass surrounding the notion “cognitive”. A very basic example of external scaffolding is in fact invoked quite commonly in epistemology (but not recognized as such): S using a thermometer to form beliefs about her environment. In matters of temperature, S indeed is quite dependent on a device like a thermometer.10 So an external feature of S’s external environment (i.e. the thermometer) certainly is causally relevant to S’s true beliefs regarding, say, room temperature. And the instrument is not one of those environmental features (prominent in Gettier-type cases) that contribute only accidentally to true belief; no, the device is designed and used to play the particular role it plays. Let me point out how this might bear on CTK.11 Consider the following scenario, due to Pritchard (2005b). Ferdinand likes to be up to date with respect to room temperature, and therefore regularly consults a thermometer on the wall. Ferdinand doesn’t know the thermometer is defective and that it fluctuates randomly within a given range. Nonetheless, Ferdinand’s readings are reliable, since unbeknownst to him, a benevolent demon sees to it that, whenever Ferdinand consults the thermometer, room temperature is adjusted so that it actually corresponds to what is displayed on the device. Although reliably formed, Ferdinand’s beliefs do not count as knowledge. Why not? According to CTK: Ferdinand doesn’t know, because his cognitive faculties are not the most salient feature explaining his correct temperature readings. He doesn’t deserve credit for his true beliefs; if credit (as causal salience) should go to anyone, it would be to the benevolent demon. But what would happen if we removed the benevolent demon and let Ferdinand use a perfectly functioning thermometer? He would figure out the temperature correctly, of course. But would his cognitive faculties really be the most salient feature explaining his success? As he is helped in the demon case (viz. by the demon), he now seems helped just by another agent: the manufacturer of the device.12 If not for her, Ferdinand wouldn’t have temperature beliefs at all. If the manufacturer doesn’t at least earn some epistemic credit, a striking (and I guess unfair) inequality arises: if 8 This is not to say that they can’t be handled satisfactorily. Quite convincing e-cog (counter)arguments are e.g., Clark (2007), Menary (2006) and Rowlands (2009). 9 See Pritchard (forthcoming, p. 2), for a similar remark. 10 This in fact holds for all sorts of instrumentation; many augment (e.g. telescopes) or substitute (e.g.

calculators) some of the cognitive faculties we are naturally endowed with. 11 Recall that CTK here (and in the remainder) assumes (G.3**). 12 Not to mention the efforts of those scientists involved in the invention of temperature. For an overview of

this intricate achievement, one may read Hasok Chang’s Inventing temperature: Measurement and scientific progress (2004).

123

522

Synthese (2011) 181:515–529

something goes wrong, we most likely blame the manufacturer, but if all goes well, we leave her out of the credit equation. The simple point I am trying to make is in fact structurally similar to Lackey’s argument regarding innate knowledge (Lackey 2007). As said, Lackey contends that the most salient explanatory feature for an innate true belief wouldn’t be some cognitive trait of the person entertaining it, but the origin of the belief, such as natural selection or some other evolutionary mechanism. What this suggests is that outcomes such as true belief may be generated by causal factors extending across the boundaries of the individual cognizer. In Lackey’s case, though, these external factors are (would be) outside human control, undeserving attributions of credit, whereas in the thermometer scenario, the causal contributions are within the realm of human intent—and therefore a proper target for normative appraisal. Another strength of the thermometer example is, of course, that it doesn’t invoke a dubitable entity like innate knowledge. Nonetheless, the scenario is vulnerable to at least two objections (more will follow in Sect. 5). The first one is that causal relevance isn’t the same as causal salience. So the CTK proponent might concede that several external factors contribute to Ferdinand’s success, but stress that salience is in the truth-conducive cognitive work performed by Ferdinand, not that performed by the manufacturer (let alone by the thermometer itself). It is Ferdinand who has decided to use the thermometer in the first place; it is Ferdinand who has developed a sense of trust regarding the reliability of the instrument, either through inductive generalization (trial-and-error, perhaps) or through testimonial evidence (provided by the manufacturer, who Ferdinand has chosen to trust).13 Ferdinand (no one else) is in control of the thermometer, so it makes good sense to consider him as causally privileged. Second, Greco and Riggs might point out that causal salience is not supposed to explain S merely believing that p, but, as stated in (G.3**), S believing the truth about p. To demonstrate the causal salience of an external feature, therefore, I shouldn’t refer to a counterfactual in which no belief is produced—such as Ferdinand not forming temperature beliefs in the absence of a thermometer—but to one in which, given similar cognitive processes, a false (rather than a true) belief is formed. I think both objections are pertinent. So instead of trying to counter them, I will rather translate them into requirements for the design of a more effective scenario. As a refutation of CTK, such a scenario should thus at least: (i) downplay the cognizer’s control in his coming to know that p—instead of S pulling information out of the device, the device should as it were push truths toward S; and (ii) involve external features explaining true belief, not just (the presence of) belief. I give it another try below. 4 Credit and extended cognition: second pass The e-cog scenario developed in this section takes its cue from a book titled The human factor (2003) by cognitive engineer Kim Vicente, and concerns airport security, in particular the X-ray scanners used for baggage inspection. Such scanners make 13 This point is made for instance in Sosa (2006, p. 118).

123

Synthese (2011) 181:515–529

523

the life of inspectors easier in a way (no need to open up and check the inside of every piece of luggage), but also more tedious and demanding. After all, illegal objects are exceedingly rare, which implies that operators typically have to stare for long uninterrupted periods of time at a monitor showing unsuspicious (and boring) goods (like toothbrushes, deodorants, computers, books). Under such conditions, people’s vigilance decreases dramatically; after half an hour, performance drops by almost 50% (ibid., p. 138). The solution (implemented worldwide after 9/11) relates to the fact that people remain more attentive if “false signals” are introduced periodically. In case of baggage scanners, these false signals are images of illegal objects superimposed on the actual images of the suitcases projected on screen. The false positives wake up the inspectors, and stimulate them to remain alert. Suppose, now, that SYSTEM1 is an ordinary pre-9/11 baggage scanner, whereas SYSTEM2 is a post-9/11 upgrade, including a “false signal” engine. Whenever SYSTEM2 projects a false image and the operator notices (she informs the system by, say, clicking on the image), the following message pops up: “False alarm: you were being tested!” If no message appears, the operator knows the threat is real. Consider, then, the following scenario: SISSICASE: Sissi has been a baggage inspector all her life. She used to work with an old-fashioned SYSTEM1 , but since 9/11, the airport she is working for introduced a SYSTEM2 . Her supervisor Joseph, a cognitive engineer who was actually involved in the design of the device, has informed her how it works (how its operation is almost identical to the operation of the old system). Currently Sissi is inspecting a piece of luggage which contains a bomb. She notices and forms a true belief regarding the contents of the suitcase. As such, the bomb is intercepted and a catastrophe prevented from happening. What is the most salient feature explaining Sissi’s true belief? Note here that the relevant counterfactual is Sissi using the old-fashioned SYSTEM1 , resulting, we may reasonably suppose, in her failing to notice the bomb, and thus in her forming a false belief (rather than her just failing to form a belief). As such, requirement (ii), formulated at the end of the previous section, is met. Since Sissi’s cognitive faculties are assumed constant over SISSICASE and the counterfactual, the difference between false and true belief is explained by a difference in the set-up of the machinery in both cases; hence, the most salient causal feature to the effect of true belief is external to Sissi. The control condition (requirement (i)) is handled adequately, too. Salience is not in some discriminatory faculty of Sissi, deciding to replace SYSTEM1 with a more truth-conducive SYSTEM2 , for that decision was Joseph’s, her supervisor. Nor can it be in her treating Joseph as a reliable testimonial source, because Joseph plays an identical role in SISSICASE and its counterpart. We have no reason to suppose that Sissi applied different faculties of trust in either case; as an employee she trusts her supervisor, irrespective of whether he is explaining the workings of SYSTEM1 or of SYSTEM2 . My diagnosis, then, is this: Sissi knows about the bomb, but doesn’t really deserve credit (in the sense of (G.3**)), simply because it is a feature external to her that makes

123

524

Synthese (2011) 181:515–529

the difference between her believing the truth (“there is a bomb”) and her believing a falsehood (“no bomb in here”).

5 Objections and replies In this section I will go through some objections one may raise. Objection :: Partial credit One line of argument might go as follows. SISSICASE assumes that credit for true belief cannot be shared. CTK isn’t wedded to such an idea, though: a knower doesn’t need to get all credit for the knowledge she acquires. So perhaps Joseph (Sissi’s supervisor) or even the machine deserve some credit for their contribution to Sissi’s true belief, still the accomplishment is hers, and that suffices for the CTK-kite to fly.14 Reply First of all, this response goes quite some way in acknowledging Sissi’s strong dependence on external features (Joseph, the machinery)—a feather in e-cog’s cap, I would say. The problem is, however, that such a distribution of credit makes CTK vulnerable to Gettier-style charges again. For what the partial credit objection suggests is to downplay Sissi’s contribution from being the most salient feature to being just one feature among many; in other words, it is proposing (G.3) instead of (G.3*), whereas in fact, as I explained in Sect. 2, we need to make the opposite move, from (G.3) to (G.3*),15 and that to make CTK Gettier-proof. If Sissi forms the belief that there is a sheep in the field and her belief is true, not because what she sees is a sheep (it’s in fact a dog), but because a sheep, well-hidden behind a rock, is grazing in the field, Sissi’s cognitive faculties are relevant, but not salient enough to explain her true belief. Sissi’s true belief is a prototypical instance of uncreditworthy true belief; to accommodate for that, CTK theorists cannot content themselves with relevance, they should go for salience.16 Closing the circle: distributing credit—that is, acknowledging the many (not all salient) causal contributions to true belief—is attractive and natural, but not a cure we should expect to save CTK. At this point CTK-theorists may deny that they have forgone the salience requirement. Distribution, they may argue, doesn’t mean equal distribution. Of all causal factors affecting Sissi’s belief, Sissi’s cognitive faculties remain privileged: her knowing about the bomb is her accomplishment, it is something she has done. This is what Riggs (2009) is after, I suspect, when introducing the notion of attributability: S’s true belief shouldn’t be attributable to luck, but to S. 14 Some such objection has been raised by Greco (2007, p. 65) with respect to Lackey’s case of testimonial knowledge. The partial credit objection expressed above is in fact a simple transposition of Greco’s and Riggs’ arguments to my case of extended cognition. 15 And ultimately, from (G.3*) to (G.3**), but let us ignore that for a moment. 16 Alternatively, as an anonymous reviewer pointed out, CTK theorists might simply dispute that in Gettier

cases S’s cognitive faculties are a salient part of the causal story to begin with. If so, (G.3) remains valid, and consequently, my argument misses its target. Although this is possible, I find it hard to imagine how one can not just minimize, but just completely leave out Sissi’s faculties in a full causal explanation of her true belief. The mere fact that Sissi sees objects in front of her, does have at least some causal bearing on the truth of her judgments concerning these objects, doesn’t it? Because if not for her seeing these objects, there wouldn’t be not only no true judgments, but no judgments to begin with.

123

Synthese (2011) 181:515–529

525

Now, let me say what I find unobjectionable: Sissi’s belief formation is attributable to her, indeed. Her forming a true belief (and that’s what at stake here), however, is not. Sissi’s success is attributable to the machinery, I repeat. For suppose we were asked to explain the superior reliability of post-9/11 security systems, that is, why these yield more true beliefs regarding the contents of suitcases and the like. Here is an answer that wouldn’t do: after 9/11, baggage inspectors simply got more attentive. This explanation misses the point, since the real cause for improvement wasn’t a change in the cognitive character of the inspectors—inspectors didn’t become more vigilant at once through their own force of will—but a change in the set-up of the artifactual environment. The crucial truth-conducive work therefore has been done by Joseph, Sissi’s supervisor. He found a way to bypass an element that was out of Sissi’s control, namely her being liable to vigilance decrement, and as such increased the likelihood of Sissi forming true beliefs. Objection :: Cognitive inequivalence Opponents might demur: your argument rests on the assumption that SISSICASE and its supposed relevant counterfactual are cognitively on a par. This is incorrect. There remains a significant difference between Sissi operating SYSTEM1 , versus her operating SYSTEM2 . For one thing, she has acquired some SYSTEM2 -specific skills, procedural knowledge about how to operate the new scanning device. In case she detects a suspicious object, for instance, she has learned to suspend judgment, and wait for a confirmation message to pop up (or not). Given these differences, her operating SYSTEM1 is not the right counterfactual to SISSICASE; therefore, you have not demonstrated the causal salience of external scaffolds. Reply SISSICASE and the counterfactual indeed display differences other than in the machinery. But I simply doubt that these differences are the most salient with respect to true belief. To repeat my question: what explains the superiority of post-9/11 systems? Wouldn’t it be awkward to answer that higher reliability is due to operators having developed more truth-conducive skills? That it was a matter of them being better trained, of having learned to suspend judgment? There is at least one person who would object: Joseph.17 And the entire (cognitive) design community, for that matter. To be clear, I do not deny that Sissi’s efforts will figure in a full explanation of her true (rather than false) belief. I just think that they are not relevant enough to call them causally salient. Objection :: No knowledge Another possible objection: SISSICASE is not a case of knowledge. When Sissi notices a suspicious object on-screen, and the scanner does not return a feedback message, she might form a belief on the contents of the suitcase in question, but this is not a full belief yet. Only when she opens up the piece of luggage and sees the bomb for herself, she can be said to fully believe, and thus to know about the bomb.

17 On the other hand, in cases of failure (Sissi not finding a bomb, causing a disaster), Joseph would arguably agree with such an arrangement, for failure wouldn’t be attributable to him either; it would be Sissi’s fault, her not having trained enough.

123

526

Synthese (2011) 181:515–529

Reply If it is a matter of SISSICASE mistakenly assuming X-ray technology to allow for unambiguous bomb identification, we can easily adjust the scenario: Sissi’s task could simply be to discern whatever it is that X-ray scanners are able to pick out unequivocally (human bones, perhaps?). If so, indications on-screen are fully reliable, and Sissi can have a full belief prior to opening the suitcase.18 One could of course doubt that indirect (i.e. instrument-mediated) perception can ever result in full belief. If one goes that way, one should also be skeptical about Ferdinand’s temperature beliefs (see Sect. 3), and about the greater part of science. I will not even try to refute that sort of skepticism here, in part because there is no indication that CTK-theorists commit themselves to such a view. Objection :: No knowledge (part 2) The “no knowledge” objection might take yet another form. One might argue that knowledge attributions are dependent on the purposes of the attributor. Greco (2008), for one, explicitly endorses such a position. If so, one can reasonably wonder why we should accept that SISSICASE is a case of knowledge, for whether Sissi knows, is just dependent on the interests and purposes of the person judging Sissi’s performance.19 Reply This might be so, but to find out what caused the disparity in the truth of Sissi’s beliefs, we must work with a ceteris paribus clause. That is, to figure out the effect of the machinery, we vary the machinery (SYSTEM1 vs. SYSTEM2 ), while keeping all other things equal, including the purposes of the person judging Sissi’s epistemic achievements. So if the purposes of an attributor would make her attribute knowledge in case of SYSTEM1 , the same should hold for the case involving SYSTEM2 (i.e. SISSICASE), and vice versa. In other words, attributor contextualism sensu Greco might be attractive, but it is a view orthogonal to my argument. Objection :: E-cog is too weak Likely the most audacious counterargument is to object to the particularly weak reading of e-cog I have been endorsing. Why not agree with the extended mind theorists’ strong reading of e-cog and consider the processes going on in the machinery as genuinely cognitive, as belonging to “Sissi the extended cognitive agent”? On this condition, Sissi’s (now extended) faculties remain the most salient feature explaining her true belief, and CTK is saved. Reply This might indeed be the dilemma: to adopt the strong programme of e-cog, or to abandon CTK. I leave it to others to argue for the first option, and give some more comments on the second in the section below.

6 Prospects for credit Is there really no way, in the face of e-cog, to save the intuition that knowledge is (must be!) produced through cognitive skill? There is a way, I think, but it requires quite some tinkering.

18 Of course, if there were such a perfect, 1-to-1 correlation (say, between the presence of human bones and measured reflected radiation) Sissi wouldn’t be needed for bone detection any more. But we could still ask her, for philosophical purposes, to double-check the machine. 19 I thank an anonymous reviewer for pointing out this possible objection.

123

Synthese (2011) 181:515–529

527

Let me reconstruct how CTK was originally defined (making some simplifications for purposes of clarity): (CTK) S knows that p, iff S’s reliable cognitive faculties are the most salient part of the total set of causal factors that give rise to S’s believing the truth that p. SISSICASE undermines CTK’s salience requirement, I argued, so it is useful to make an adjustment there. We can lower CTK’s stakes, by requiring knowledge to involve S’s cognitive faculties in relevant (rather than salient) ways: (CTK*) S knows that p, iff S’s reliable cognitive faculties are an important part of the total set of causal factors that give rise to S’s believing the truth that p. (CTK*) already sits a bit more comfortable with SISSICASE. Indeed, the machinery isn’t the sole cause of Sissi’s true belief; Sissi’s perceptual skills, for instance, are relevant, though not the difference-maker yielding true belief. To make (CTK*) Gettier-proof, though, a second adjustment is needed. For suppose Sissi forms the belief that there is a bomb in the suitcase and this is true, not because what she sees is a bomb (it is in fact a hairdryer), but because behind the hairdryer a bomb is hidden. In this case, Sissi’s cognitive faculties are important to her true belief, yet this doesn’t make for knowledge. Hence, we should rewrite (CTK*) as follows: (CTK**) If S knows that p, then S’s reliable cognitive faculties are an important part of the total set of causal factors that give rise to S’s believing the truth that p. (CTK**) relaxes the conditions for knowledge a bit further. Instead of using cognitive faculties to define what knowledge is, cognitive faculties are considered a feature present just whenever knowledge is. And if (CTK*), save Gettier, accommodated for e-cog, (CTK**) will, by extension, do likewise.20 In sum, (CTK**) is plausible, but limited in scope. It points to an important characteristic of knowledge—knowledge implies credit, somehow, somewhere—but doesn’t pick out a unique property discriminating it from lucky true belief—the problem CTK-theorists set out to resolve. So although informative, (CTK**) will likely be a disappointment for those seeking the truth about knowledge. 7 Conclusion Suppose again that Franz, our lousy archer, pushes a button on a remote control, thereby activating a bow-and-arrow set, perfectly positioned and calibrated by a more capable archer. Suppose moreover that the shot hits target. Although the shot might be attributable to Franz, hitting target isn’t. 20 In a recent paper—to my mind, the only manuscript exploring the epistemological ramifications of e-cog in detail—Pritchard (forthcoming) too suggests that only a weak version of cognitive ability (à la (CTK**)) can be made consistent with e-cog. In his argument, though, he dismisses stronger versions (à la (CTK)) not on grounds particular to e-cog, but by rehearsing Lackey’s case of testimonial knowledge. Only in a second step he discusses e-cog, showing its fit with weak theories of credit. I have shown in this paper, however, that we can make Pritchard’s first step independent from Lackey; indeed, I introduced e-cog as an alternative (or complement) to her case of testimony.

123

528

Synthese (2011) 181:515–529

The same goes, I have argued, for knowledge: S may come to know that p, not through S’s own skill, but due to S’s environment being intentionally set-up such that S would come to believe the truth regarding p. In such cases of extended cognition, the most salient causal factor explaining true belief is not so much a cognitive trait of S as an external feature cognitively scaffolding her. Such scaffolding isn’t creditable to S, hence strong credit theories of knowledge must be wrong. Acknowledgements Research by Krist Vaesen was supported by the Netherlands Organisation for Scientific Research (NWO). He thanks Philip Nickel, Wybo Houkes, Martin Peterson, Anthonie Meijers, Duncan Pritchard, Andy Clark, Hans Radder, Joel Anderson, Lieven Decock, Olle Blomberg, Marieke van Holland, Auke Pols, Joel Katzav and Andreas Spahn for useful discussions on previous drafts of this paper. Open Access This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.

References Adams, F., & Aizawa, K. (2001). The bounds of cognition. Philosophical Psychology, 14(1), 43–64. Adler, J. (2002). Beliefs own ethics. Cambridge, MA: MIT Press. Audi, R. (2001). Doxastic voluntarism and the ethics of belief. In M. Steup (Ed.), Knowledge, truth, and duty (pp. 93–113). Oxford: Oxford University Press. Chang, H. (2004). Inventing temperature: Measurement and scientific progress. Oxford: Oxford University Press. Clark, A. (2007). Mementos revenge: The extended mind, extended. In R. Menary (Ed.), The extended mind. Aldershot: Ashgate. Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 1, 7–19. Greco, J. (2003). Knowledge as credit for true belief. In M. DePaul & L. Zagzebski (Eds.), Intellectual virtue: Perspectives from ethics and epistemology (pp. 111–134). Oxford: Oxford University Press. Greco, J. (2007). The nature of ability and the purpose of knowledge. Philosophical Issues, 17(1), 57–69. Greco, J. (2008). Whats wrong with contextualism?. Philosophical Quarterly, 58(232), 416–436. Hieronymi, P. (2006). Controlling attitudes. Pacific Philosophical Quarterly, 87(1), 45–74. Kvanvig, J. (2009). Responses to critics. In D. Pritchard, A. Haddock, & A. Millar, The value of knowledge. Oxford: Oxford University Press. Lackey, J. (2007). Why we dont deserve credit for everything we know. Synthese, 158(3), 345–361. Lackey, J. (2009). Knowledge and credit. Philosophical Studies, 142(1), 27–42. Menary, R. (2006). Attacking the bounds of cognition. Philosophical Psychology, 19(3), 329–344. Owens, D. (2000). Reason without freedom: The problem of epistemic normativity. London: Routledge. Pritchard, D. (2005a). Epistemic luck. Oxford: Oxford University Press. Pritchard, D. (2005b). Virtue epistemology and the acquisition of knowledge. Philosophical Explorations, 8(3), 229–243. Pritchard, D. (forthcoming). Cognitive ability and the extended cognition thesis. Synthese. Raz, J. (1999). Engaging reason: On the theory of value and action. Oxford: Oxford University Press. Riggs, W. D. (1998). What are the chances of being justified?. The Monist, 81(3), 452–473. Riggs, W. D. (2007). Why epistemologists are so down on their luck. Synthese, 158, 329–344. Riggs, W. D. (2009). Two problems of easy credit. Synthese, 169, 201–216. Rowlands, M. (2009). Extended cognition and the mark of the cognitive. Philosophical Psychology, 22(1), 1–19. Rupert, R. (2004). Challenges to the hypothesis of extended cognition. Journal of Philosophy, CI(8), 389–428. Sosa, E. (2006). Knowledge: Instrumental and testimonial. In J. Lackey & E. Sosa (Eds.), The epistemology of testimony (pp. 116–126). Oxford: Oxford University Press.

123

Synthese (2011) 181:515–529

529

Sosa, E. (2007). A virtue epistemology: Apt belief and reflective knowledge (Vol. 1). Oxford: Oxford University Press. Vicente, K. J. (2003). The human factor: Revolutionizing the way people live with technology. Toronto: A.A. Knopf Canada.

123