Synthese DOI 10.1007/s11229-008-9400-0
Confusion about concessive knowledge attributions Dylan Dodd
Received: 5 February 2008 / Accepted: 3 September 2008 © Springer Science+Business Media B.V. 2008
Abstract Concessive knowledge attributions (CKAs) are knowledge attributions of the form ‘S knows p, but it’s possible that q’, where q obviously entails not- p (Rysiew, Nous (Detroit, Mich.) 35:477–514, 2001). The significance of CKAs has been widely discussed recently. It’s agreed by all that CKAs are infelicitous, at least typically. But the agreement ends there. Different writers have invoked them in their defenses of all sorts of philosophical theses; to name just a few: contextualism, invariantism, fallibilism, infallibilism, and that the knowledge rules for assertion and practical reasoning are false. In fact, there is a lot of confusion about CKAs and their significance. I try to clear some of this confusion up, as well as show what their significance is with respect to the debate between fallibilists and infallibilists about knowledge in particular. Keywords Concessive knowledge attributions · Fallibilism · Contextualism · Invariantism · Epistemic modals 1 Introduction Concessive knowledge attributions (CKAs) are knowledge attributions of the form ‘S knows p, but it’s possible that q’, where q obviously entails not- p (Rysiew 2001). The significance of CKAs has been widely discussed recently. It’s agreed by all that CKAs are infelicitous, at least typically. But the agreement ends there. Some writers (e.g., Lewis 1996) invoke them to defend contextualist accounts of the semantics of ‘knows’, others (Rysiew 2001; Brown 2006) argue they should be explained by an invariantist theory. Some (e.g., MacFarlane forthcoming) discuss them as providing
D. Dodd (B) Arché – Philosophical Research Centre for Logic, Language, Metaphysics and Epistemology, University of St Andrews, St Andrews, UK e-mail:
[email protected]
123
Synthese
evidence for certain theses about epistemic modals. Some (Lewis 1996) think they provide powerful evidence for the truth of infallibilism (the doctrine that S only knows p if S’s evidence entails p), others think they provide evidence for fallibilism (Vogel 1999; Dougherty and Rysiew forthcoming), others seem to think they don’t provide any evidence either way (Stanley 2005a,b), and finally still others (Hill and Schechter 2007) think they not only provide evidence for fallibilism but that they also undermine the case made by DeRose (1996) and Williamson (2000) for the knowledge rule for assertion and show that knowledge accounts of practical reasoning are false. Furthermore, some think their infelicity should be explained semantically (MacFarlane forthcoming; Stanley 2005a,b), others pragmatically (Rysiew 2001; Brown 2006; Dougherty and Rysiew forthcoming). In fact, there’s a lot of confusion about CKAs. In this paper I try to clear at least some of it up. In what follows I will argue for the following theses (not in this order). First, the infelicity of CKAs does provide prima facie evidence for infallibilism. Furthermore, pace Stanley (2005a,b), the infelicity of CKAs cannot be explained semantically by fallibilists.1 Furthermore, it turns out that it follows from something virtually everyone working on epistemic modals has claimed that CKAs can be explained semantically. Stanley—and as far as I can tell everyone else—has failed to notice something, namely this. The infelicity of CKAs reveals that the dominant view in epistemology, fallibilism, is inconsistent with virtually all work on epistemic modals. Second, pace Vogel (1999) and Dougherty and Rysiew (forthcoming), the existence of so-called felicitous CKAs provides no evidence for fallibilism. Since fallibilists cannot explain the infelicity of CKAs semantically, they must attempt to explain it pragmatically. So third, I will also discuss attempts by fallibilists to amend the received view of epistemic modals and explain the infelicity of CKAs pragmatically, focusing on the most recent attempt (Dougherty and Rysiew forthcoming; I will also discuss a different attempt put forward in Rysiew 2001). I will argue that such attempts fail. Fourth, I will claim there has been an important desideratum that fallibilists also need to explain pragmatically. That they have this second explanatory burden is a fact that has been so far ignored. I will conclude by discussing how in the future fallibilists might go about giving a unified pragmatic explanation of both it and the infelicity of CKAs, and present some problems such an approach would face.
2 The infelicity of CKAs and the received view of epistemic possibility CKAs are typically infelicitous—they sound odd, even contradictory. Here’s an example. If I say ‘I know the Red Sox won’t win the next World Series’, you’re likely to dispute the truth of my claim. But if I say ‘I know they won’t win the next Worlds Series, even though it’s possible they will’, you’ll wonder what I just said. It seems that first I said I knew something, but then retracted my knowledge claim. Surely, in
1 That it can’t has been assumed without argument by Lewis (1996) and Rysiew (2001). Dougherty and Rysiew (forthcoming) are the only ones who have actually argued for this claim, but their argument differs from mine.
123
Synthese
conceding that after all it’s possible that the Red Sox will win the World Series, I’m admitting I don’t really know they won’t, right? How do we explain the infelicity of CKAs? Before discussing this question, a few preliminaries. The possibility claim in CKAs is a claim about what is epistemically possible for the utterer (and perhaps some others too, e.g., the members of his context).2 Thus, if I utter ‘I know the Red Sox won’t win the next World Series, but it’s possible they will’, I express the proposition that I know the Red Sox won’t win the next World Series & it’s epistemically possible for me that they will win the next World Series. In what follows I will focus exclusively on CKAs which express propositions where the person to whom knowledge is ascribed is among those to whom the epistemic possibility claim is (implicitly) relativized.3 The point here is that CKAs are infelicitous in normal, non-philosophical contexts—contexts where we’re discussing everyday things like who will win the world series, who will be the next president, or what the weather will be like next week. In such contexts, if you claim to know p, even though you can’t rule out the possibility that not- p, we won’t know what to you just said. It’ll seem to your audience like things aren’t so clear when we get to the crazy, far out possibilities that philosophers and pretty much no else discuss. What if you say things like ‘I know I have hands, but I can’t rule out the possibility that I’m a handless brain-in-a-vat’, or ‘I know I’m not in the Matrix, but it’s possible I am’, or ‘I know the next twenty golfers won’t get holes-in-one consecutively, even though I guess strictly speaking there’s a chance they will (and therefore, it’s possible they will)’? Unlike the cases where we’re talking about normal stuff, intuitions are probably going to be all over the place about these ‘philosophical’ examples. I suspect part of the problem here is we just don’t know how to evaluate claims about such possibilities being epistemically possible, or about our knowing that they don’t obtain. Might I be a brain-in-a-vat? Do I know I’m not? It’s hard to get a grip on how to answer such questions. Anyway, to repeat, my present claim is about our intuitions about how we talk and think in everyday life. In such contexts when we’re talking about pedestrian possibilities, CKAs are definitely infelicitous. Some writers (Vogel 1999; Dougherty and Rysiew forthcoming) have claimed that the fact that we don’t find CKAs about certain far out possibilities so infelicitous is a very significant fact. I’ll postpone discussion of this claim to Sect. 6. A fact highlighted by MacFarlane (forthcoming) and Stanley (2005b) is that the problem of CKAs can be easily explained semantically by accepting something that virtually all philosophers and linguists who work on epistemic modals tell us, namely this. A proposition q is epistemically (im)possible relative a certain knowledge base. Exactly what propositions are included in this knowledge base is contentious, but if we’re wondering whether q is possible for S, then S’s own knowledge will definitely be in the knowledge base that’s relevant for answering our question. Whatever the relevant knowledge base is, if not-q is entailed by the propositions in it, in a way that’s 2 Propositions are epistemically (im)possible only relative to a certain subject or group of subjects. As I presently sit by myself in my office, it is epistemically possible for me right now that my wife doesn’t have a headache. However, if my wife is presently suffering from a migraine, it’s not epistemically possible for her that she’s not suffering from a headache. 3 For a discussion of how to solve the problem for other sorts of CKAs, see Stanley (2005b).
123
Synthese
obvious to S (perhaps we must also add: and obvious to members of a certain group of subjects which includes S), then q isn’t epistemically possible for S. Otherwise, it is. As I said, virtually all writers on epistemic modals accept the positions just stated, thus I will henceforth call it ‘the received view (of epistemic possibility)’.4 Now, if we accept the received view, then the infelicity of CKAs can be explained as follows. I will run the explanation for an utterance by S of ‘I know that p, but it’s possible that q’ (where q obviously entails not- p), but the explanation generalizes. According to the received view, q is only epistemically possible if not-q isn’t obviously entailed by a certain knowledge base KB, and the propositions S knows are in KB. So if S knows p, p will be in KB. Since, per hypothesi, q obviously entails not- p, p obviously entails not-q. Therefore KB obviously entails not-q. Therefore, qis not epistemically possible for S if S knows p. Therefore, S’s utterance ‘I know that p, but it’s possible that q’ can’t express a truth. In fact, what S says is contradictory. So it isn’t hard to explain the infelicity of CKAs. According to the received view of epistemic possibility they’re contradictory. The problem is some philosophers— notably David Lewis 1996—have alleged that if CKAs are indeed contradictory, then the dominant view in epistemology, fallibilism, is also contradictory. If Lewis is right about this, in order to maintain their fallibilism, epistemologists must reject the received view of epistemic modals. In the next section I will argue that he is right; thus fallibilists do need to reject the received view of epistemic modals. This is the problem of concessive knowledge attributions. Contextualists like Lewis have proposed that the right way to solve this problem is by giving a contextualist semantics for ‘knows’ and its cognates. In Sect. 4 I will discuss the contextualist approach to this problem, and argue that it’s inadequate. In a recent paper Trent Dougherty and Patrick Rysiew (forthcoming) propose a solution to this problem. But in Sect. 5 I argue that their proposed solution is inadequate. In the concluding Sect. 6 I explain why another pragmatic explanation of the infelicity of CKAs—Rysiew 2001’s—is also inadequate. In the process I will introduce a second desideratum they and others are ignoring which fallibilists must also explain pragmatically. In Sect. 7 I will conclude by suggesting how fallibilists might go about giving a unified pragmatic explanation of both desiderata. 3 Is the problem of CKAs really a problem for fallibilists? In this section I’ll argue that the problem of CKAs is a problem for fallibilists. In a well-known statement of the problem of CKAs, Lewis (1996, p. 549) says this: It seems as if knowledge must be by definition infallible. If you claim S knows that P, and yet you grant that S cannot eliminate a certain possibility in which not-P, it certainly seems as if you have granted that S does not after all know that P. To speak of fallible knowledge, of knowledge despite uneliminated possibilities of error, just sounds contradictory. 4 For example, see DeRose (1991), Gillies and von Fintel (2008a,b), Hacking (1967), Kratzer (1977) and
Teller (1972). All the above claims about epistemic modals are even endorsed by recent relativist theories, see e.g., Egan et al. (2005), Egan (2007), MacFarlane (forthcoming), and Stephenson (2008).
123
Synthese
In this passage it seems that Lewis understands fallibilism to be the doctrine that 3.1 S can know p even when S’s epistemic situation isn’t such that S can rule out the possibility that not- p. Given what he says later (p. 551), it’s clear Lewis thinks 3.1 should be glossed as the doctrine that S can know p even when S’s evidence doesn’t eliminate the possibility that not-p.5 If we assume evidentialism, that the nature of one’s epistemic situation is completely characterized by what evidence one has, this yields Lewis’ gloss on 3.1. And if we also assume that if S’s evidence isn’t such that S can completely rule out the possibility that not-p, then S’s evidence doesn’t entail not-p, we get from 3.1 the following common definition of fallibilism: 3.2 S can know p even when S’s evidence doesn’t entail not- p. 6 Let’s assume 3.2 as a working definition of ‘fallibilism’. According to the received view of epistemic possibility, CKAs express necessarily false propositions. Is this really inconsistent with fallibilism? Stanley (2005a,b) thinks it’s not, though it’s unclear why. After endorsing a version of the received view of epistemic possibility, and explaining how on this view CKAs are necessarily false, he says: But now we can see that the fact that such utterances are almost always odd does not raise any problem for fallibilism. The obviously falsity of the propositions expressed by [CKAs] is consistent with the truth of the claim that one can know that p, on the basis of evidence that is logically consistent with not- p (p. 128). But, contra Stanley, the obvious falsity of propositions expressed by CKAs could pose a problem for fallibilists if the fact that they are necessarily false is inconsistent with 3.2. And I think this fact is inconsistent with 3.2. As Lewis says, when S utters a CKA, S is saying (*) S knows p, but can’t rule out a possibility in which not-p. Assuming with Lewis that it’s in virtue of our evidence that we’re able to rule epistemic possibilities out, and that our evidence rules a possibility out iff our evidence isn’t consistent with the possibility’s obtaining, (*) is equivalent to (**) S knows p, but there is a possibility in which not-p that’s consistent with S’s evidence. So, given plausible evidentialist assumptions about how we rule possibilities out, CKAs express necessarily false propositions iff (**) is necessarily false. (**) is necessarily false iff 3.2 is false. Thus it’s plausible that if CKAs express necessarily false propositions, then fallibilism is false. Thus the fact that CKAs sound contradictory does pose a prima facie problem for fallibilists. I think Dougherty and Rysiew are right to try to explain CKAs’ infelicity pragmatically, given their commitment to fallibilism. Before looking at their proposed 5 What does it mean that p is eliminated by S’s evidence? Lewis has a very particular take on this, which I shall be ignoring. He says (p. 553), “Here I say that the uneliminated possibilities are those in which the subject’s entire perceptual experience and memory are just as they actually are. There is one possibility that actually obtains (for the subject and at the time in question); call it actuality. Then a possibility W is uneliminated iff the subject’s perceptual experience and memory in W exactly match his perceptual experience and memory in actuality.” Note the internalistic understanding of evidence that Lewis is presupposing here. 6 It’s the definition of ‘fallibilism’ that both Dougherty and Rysiew (forthcoming) and Stanley (2005a,b)
are working with. Regardless of how one wants to define ‘fallibilism’, 2 states a doctrine that is worthy of our attention, and which most epistemologists will want to endorse.
123
Synthese
pragmatic explanation, we’ll first look at how a contextualist could solve the problem of CKAs. 4 Contextualism and CKAs According to contextualists about ‘knows’ and its cognates (henceforth ‘contextualism’), the proposition expressed by sentences of the form ‘S knows p’ varies from one context to another, even when the semantic contribution of ‘S’ and ‘ p’ is invariant. In other words, contextualists think ‘knows’ and its cognates are context dependent expressions. Lewis (1996) motivated his own version of contextualism by its ability to explain the infelicity of CKAs. In what follows I will discuss the plausibility of a contextualist explanation of the infelicity. I won’t focus on the details of Lewis’ own theory, but discuss the resources that contextualism in general brings to bear on this problem. In the sections that follow this one, we shall see that many problems face the attempt to explain the infelicity of CKAs pragmatically. A central advantage of a contextualist approach to the problem is that it can avoid these problems by explaining it semantically. According to contextualists like Lewis (see also Schaffer 2004, forthcoming a,b), the following is true: 4.1 CKAs always express falsehoods. That’s why they’re infelicitous. Let’s see where endorsing 4.1 takes us. A natural thought is this: 4.2 If ‘q might be true’ (as uttered by a subject S in a context C) expresses a truth, that’s because S isn’t in a position to rule all the possibilities in which q. If q is epistemically possible for you, presumably that’s because you can’t rule out the possibility that q. And that’s just what 4.2 says. Now consider an arbitrary first person CKA, ‘I know p, but q might be true’. If we endorse 4.1, we think ‘I know p’ (as uttered by S in C) expresses a proposition that entails that the proposition expressed by ‘q might be true’ (as uttered by S in C) is false. So if we endorse 4.1 and 4.2, then we think ‘I know p’ (as uttered by S in C) expresses a proposition entailing that S can rule out all the q-possibilities. q here is just an arbitrary proposition that contradicts p. Thus if we endorse 4.1 and 4.2 we’re committed to 4.3 ‘I know p’ (as uttered by S in C) is true only if S can rule out all possibilities incompatible with p. But, assuming unrestricted quantification, 4.3 is a thesis many find implausible because they think it must lead to scepticism. Here’s Lewis saying this: Let your paranoid fantasies rip—CIA plots, hallucinogens in the tap water, conspiracies to deceive, old Nick himself—and soon you find that uneliminated possibilities of error are everywhere. Those possibilities of error are far-fetched, of course, but possibilities all the same. They bite into even our most everyday knowledge. We never have infallible knowledge. (1996, p. 549)
123
Synthese
For this reason, contextualists like Lewis want to deny 4.3, at least where it’s read with unrestricted quantification. For they want many of our knowledge attributions to come out true. But how can the contextualist, who recall endorses 4.1, consistently deny 4.3, since as we’ve just seen, 4.3 seems to follow from 4.1? Here’s how contextualists like Lewis or Schaffer can avoid the implication. There are set of possibilities that are ‘live options’ in a given context.7 These are the possibilities that are relevant to the truth conditions of both knowledge and epistemic possibility attributions. On this view, 4.3 won’t be true, when read with unrestricted quantification. But this will be: 4.3* . ‘I know p’ (as uttered by S in C) is true only if S can rule out all the possibilities that are incompatible with p and which are live options in C. ‘q might be true [for S]’ (as uttered in C) is true only if S cannot rule out all the q possibilities which are live options in C. If we assume, plausibly, that any possibility that’s salient to the speaker and the members of her audience in C will be a live option in C, and that explicitly referring to a possibility in an utterance will make that possibility salient, then the possibility q referred to in a CKA will be a live option in C. It follows from the semantics of knowledge attributions, then, that S must be able to rule q out for the conjunct ‘I know p’ to be true. But in asserting the other conjunct, ‘q might be true’, the speaker explicitly denies that claim. Thus 4.1 is true according to the sort of contextualist view we’re considering. Finally, contextualists can give this semantic explanation of the infelicity of CKAs without having to endorse 4.3 (read with unrestricted quantification), and thus scepticism. Many serious objections have been raised against contextualism.8 In my view, one objection in particular is decisive against it, namely the objection that contextualists are committed to “semantic blindness”.9 In a nutshell, this is the objection. If I hear Bob say ‘I know p’, if I think he’s sincere I’ll have no hesitancy to think to myself ‘Bob believes he knows p’, and to say that to others. But if contextualism is right, this is a mistake. For it would be a mistake to assume that ‘knows’ means the same thing in my context when I ascribe this belief to Bob. But English speakers do so assume, which shows that if contextualism is right English speakers are blind to the semantics of ‘knows’—blind to the fact that it’s context dependent. And I think it’s implausible that English speakers are semantically blind in this way. Here is another problem with contextualism, which is related to the semantic blindness problem, but which has to do with CKAs. Consider the following conversations: Conversation 1 John: I know I won’t be able to afford to go on a safari this summer, even though I might be able to—I do own a lottery ticket after all. Frank: Huh? So you don’t know it, strictly speaking. Is that what you’re saying? 7 As Schaffer emphasizes, this is something that is independently plausible. See Stalnaker (1978, 1998). Schaffer also emphasizes that the semantic framework he uses in his theory of knowledge ascriptions is independently motivated by the work of Groenendijk and Stokhof on the semantics of questions. See Groenendijk and Stockhof (1984, 1989). 8 For seminal criticisms of contextualism, see Hawthorne (2004), and Stanley (2004, 2005a). 9 Schiffer (1996) and Hawthorne (2004, Sect. 2.7). For a defence of contextualism against this objection
see DeRose (2006).
123
Synthese
Conversation 2 John: I know I won’t be able to afford to go on a safari this summer. I’m out of money. Frank: You might be able to—you do own a lottery ticket after all—even though what you just said is true. We’ve just seen how the contextualist can explain semantically the infelicity in John’s utterance in Conversation 1. But the contextualist cannot explain the infelicity in Frank’s utterance in Conversation 2. In that conversation, when John says ‘I know …’ the lottery possibility isn’t salient. So we can imagine that he doesn’t have to be in a position to rule it out when he attributes knowledge to himself. His utterance expresses a proposition—call it ‘L’—that is true even though he can’t rule the lottery possibility out. But in his response when Frank says ‘what you just said’, he’s referring to the proposition John communicated, i.e., L. As we just saw, it’s compatible with the truth of this proposition that John might be able to afford the safari in virtue of his winning the lottery. So, according to contextualism, there’s nothing contradictory about Frank’s response. But it’s just as infelicitous as what John says in Conversation 1. Contextualists will need to explain why Frank’s utterance is infelicitous in Conversation 2. They’ll need to explain this pragmatically. This is a serious blow to contextualism, since it was supposed to be a chief advantage of their theory that it could explain CKAs semantically. Thus a central motivation for the theory has now been lost. In light of this explanatory insufficiency, and in light of all the other problems that others have shown contextualism has, I think we should reject the contextualist solution to the problem of CKAs.10 5 Dougherty and Rysiew’s two proposals, and the second desideratum We’ve looked at a contextualist solution to the problem of CKAs. I argued that we should reject it. Instead, we should seek a solution to the problem within an invariantist semantics for ‘knows’ and its cognates. We now turn to the most recent invariantist treatment of the problem, put forward in Dougherty and Rysiew (forthcoming). As we have seen, according to the received view of epistemic possibility, what is epistemically possible is a function of what is known. Dougherty and Rysiew reject this view as it seems fallibilists must. Instead, they propose: 5.1 p is epistemically possible for S iff not- p isn’t entailed by S’s evidence. In other words, what is epistemically possible is a function of one’s evidence rather than one’s knowledge.11 10 As far as I know, subject sensitive invariantism (see Fantl and McGrath 2002; Hawthorne 2004; Stanley
2005a) doesn’t have any special resources for solving the problem of CKAs. No defender of SSI has argued that it does, to my knowledge. Thus I won’t be discussing SSI in this paper. 11 As they recognize, if 5.1 were combined with the doctrine that one’s knowledge is a subset of one’s evidence (proper or not), it would still be the case that CKAs are necessarily false, and we would still have an account of epistemic possibility that contradicted fallibilism. So they also claim that some propositions that one knows are not part of one’s evidence. Thus they reject Williamson 2000’s doctrine that knowledge = evidence.
123
Synthese
They then propose a pragmatic, Gricean explanation of the infelicity of CKAs as follows, although as we shall see it’s based on a misunderstanding of Gricean pragmatics. Grice’s (1989) Cooperative Principle (CP) says that one should strive to make one’s assertions as relevantly informative as one can. If S says ‘It’s possible that not- p’, if we’re assuming that S is abiding by CP we’ll assume S couldn’t have said something more informative instead, such as ‘I know that p’. So an utterance of ‘It’s possible that not- p’, or of ‘It’s possible that q’ where q obviously entails not- p, will implicate that S doesn’t know that p. Thus we have an explanation of why my utterance of ‘q is possible’ communicates something that contradicts the proposition expressed by ‘I know p’ (where q obviously entails not- p). I agree with everything Dougherty and Rysiew have said so far. Their mistake is to think that all this amounts to an explanation of the infelicity of CKAs. It doesn’t. As Grice (1989) pointed out, saying in a letter of recommendation ‘This applicant has wonderful handwriting’ typically implicates that he’s not a very talented philosopher, but that doesn’t mean writing ‘This applicant has wonderful handwriting and is a very talented philosopher’ is infelicitous. What Dougherty and Rysiew didn’t seem to notice is that just because an utterance of ‘ p’ implicates that not-q doesn’t mean that an utterance of ‘ pand q’ is infelicitous. Thus showing that ‘ p’ implicates not-q shouldn’t be taken to provide an explanation of the infelicity of ‘ p and q’. So I think their proposal of how to explain pragmatically the infelicity of CKAs fails. But what about their other proposal, that what is epistemically possible is a function of evidence rather than knowledge? This seems to be a plausible claim about epistemic possibility, and at least it isn’t obvious that it entails that CKAs are all false (but see footnote 11)—the entailment fallibilists need their account of epistemic possibility to avoid. However, there is the following problem with this proposal. Speakers take ‘It’s possible that p’ to be more or less equivalent to ‘For all I [or we] know, p’ (Hintikka 1962, Chap. 2). If a mathematician told us ‘It’s possible that Goldbach’s Conjecture is true’, and we asked her, ‘So, for all you know, it’s true?’ it would be natural for her to respond with ‘Yeah, that’s what I just said’. We would regard it as bizarre for her to say ‘Hold on, I only said it’s possible that it’s true. I didn’t mean to imply that for all I know it’s true.’12 Indeed, just like CKAs, it’s infelicitous to utter sentences of the following form: 5.2 For all I know, p is true, but p is definitely false. 5.3 For all I know, p is true, but p has to be false. 5.4 For all I know, p is true, but it’s not the case that p might be true. As with CKAs, when one utters the second conjunct in the above, it feels like one has retracted what one just said in the first conjunct. The hypothesis that (EPK) what is epistemically possible is a function of what is known predicts the infelicity of 5.2–5.4, but the hypothesis that (EPE) what is epistemically possible is a function of the evidence does not. This can be seen as follows. The left conjunct denies that p is inconsistent with the speaker’s body of knowledge. 12 As before, I’m talking about what’s the case when we’re speaking about everyday things in everyday contexts. Things may get less clear when we’re talking about the esoteric, bizarre, far out possibilities that philosophers focus on. To repeat, I’m postponing discussion of these sorts of possibilities to Sect. 6.
123
Synthese
And then the right conjunct retracts that claim by saying p’s not epistemically possible, and thus affirms that it’s inconsistent with one’s body of knowledge. Contradiction—if what’s epistemically possible is a function of what is known, but not if it’s not. So here’s a point in favor of (EPK) over (EPE) (where, controversially, these are taken to be mutually exclusive; see footnote 11). Furthermore, the phenomena here appears to be similar to CKAs; thus it would be desirable to give a unified treatment of the infelicity of CKAs and of 5.2–5.4. An important advantage the advocate of (EPK) has over the EPE theorist is that they’re able to give the above simple unified explanation of the infelicities of both CKAs and 5.2–5.4. All this points to a significant advantage an advocate of (EPK) has over Dougherty and Rysiew. On the other hand, if we want to be fallibilists, we’ve seen that it seems we just won’t be able to say that what is epistemically possible is a function of our knowledge, as the received view would have it. So what is epistemically possible has to be a function of some other feature of our epistemic situation. It seems that all fallibilists must simply bite the bullet here. They’re going to have to explain both the infelicity of 5.2–5.4 and the infelicity of CKAs pragmatically. And if we think it’s something besides knowledge that determines what is epistemically possible, why not think that it’s our evidence? Dougherty and Rysiew may have the given the most plausible alternative to the received view of epistemic possibility that is available to fallibilists. Before concluding by discussing how fallibilists might explain these two things pragmatically, I will discuss something that Dougherty and Rysiew think is evidence that CKAs should be explained pragmatically rather than semantically, namely the existence of felicitous CKAs. 6 Is it significant that there are felicitous CKAs? We have seen that the received view of epistemic modals has CKAs all expressing contradictions. Dougherty and Rysiew claim that a disadvantage of such an explanation of their infelicity is that it leaves unexplained the fact that competent speakers do sometimes utter felicitous CKAs (forthcoming, p. 6). They think this is evidence that CKAs don’t express contradictions. Let us look at some of Dougherty and Rysiew’s examples (pp. 7–9) of allegedly felicitous CKAs (they take 6.1 and 6.2 from Hawthorne 2004; for other examples see Vogel 1999; Hill and Schechter 2007; 6.4 is my own): 6.1 I know they’re going to lose, but I’m going to carry on watching just in case. 6.2 I know it’s not going to rain, but still, I’m going to take an umbrella just in case. 6.3 A referee suggests making a small point explicit, and then says, ‘I know no one will get confused about this, but it’s possible someone will’. 6.4 I know I have hands, even though it’s possible I’m a handless brain in a vat. I’ll grant for the sake of argument that 6.1–6.4 can all be uttered felicitously. Dougherty and Rysiew think this shows that CKAs don’t express contradictions. Why this is supposed to follow will be discussed presently. First, it should be pointed out that 6.1 and 6.2 aren’t felicitous CKAs—that’s because they’re not CKAs—they don’t explicitly say anything about something being
123
Synthese
epistemically possible. Why then, are they presented as evidence that there are felicitous CKAs? I think it must be that when we interpret what the speaker of 6.1 and 6.2 is saying, she might as well have uttered a CKA. Let me explain. When the speaker utters 6.1, she says she’s going to keep watching the game just in case her team comes back and wins. She could be read as saying something like this. There’s a slight chance or possibility that her team will come and win; just in case that possibility obtains, she’s going to keep watching. So she might as well have said that she knows that her team will lose, but it’s possible that it will win. Now let’s examine 6.2. The speaker who asserts it says he’s going to take an umbrella just in case it rains. He is acting as if its raining were a possibility. And so we can charitably interpret him as believing that it’s epistemically possible (for him) that it will rain, even though he says he knows it won’t.13 Just as with 6.1, then, the speaker of 6.2 can be interpreted as believing that a certain not- ppossibility is epistemically possible, even though they ascribe to themselves knowledge of p. And in both cases the above-mentioned authors have the intuition that there’s nothing infelicitous about speakers saying such things. Finally, in 6.3 and 6.4 the speaker explicitly utters a CKA. Once again, some have the intuition such utterances can be felicitous. We’re assuming 6.1–6.4 can be uttered felicitously, and thus that speakers can felicitously utter CKAs. Dougherty and Rysiew think that this is powerful evidence that CKAs can be true, and thus that S knows p doesn’t entail the negation of q is epistemically possible for S[where q is obviously a not- p possibility]. They seem to be assuming that the fact that a CKA can be uttered felicitously means it doesn’t have a contradictory content. This is what I deny. Speakers can and do utter contradictions felicitously all the time.14 Examples like the following are common in everyday discourse: ‘It was the best of times, and it was the worst of times’; ‘I love you, and I hate you’; ‘He’s a genius, but he’s stupid’; ‘I’m happy but I’m not happy’.15 A similar phenomenon is that speakers frequently say things that are completely trivial— e.g., ‘Tomorrow’s another day’16 —felicitously, in spite of Grice’s Maxim of Quantity (Grice 1989), which requires speakers to be informative. A recently published scholarly essay begins with this sentence: “Climate change may—or may not—be a central issue for the world economy” (Dell et al. 2008, p. 1). Just because speakers can felicitously utter a certain sentence S, doesn’t mean we shouldn’t also think S has a content
13 It’s unclear whether this is the best interpretation. An anonymous reviewer thought the speakers of 6.1 and 6.2 would be better interpreted simply as admitting doing something irrational. 6.1’s speaker knows her team will lose, but she irrationally carries on watching as if they might. 6.2’s speaker knows it won’t rain, but irrationally brings an umbrella as if it might. Personally, I’m not sure what to think of 6.1 and 6.2, and have some sympathy for the referee’s interpretation. If it were right, as opposed to a more charitable interpretation like the one I offered above, then the felicity of 6.1 and 6.2 provide no evidence that there are true CKAs, contrary to what Dougherty and Rysiew think. In this section I will grant Dougherty and Rysiew’s position that the speakers of these sentences should be interpreted charitably. I will go on to argue that the felicity of 6.1 and 6.2 still doesn’t provide any evidence that there are true CKAs. 14 This was emphasized to me by an anonymous referee. 15 This last example was from the anonymous referee. 16 This example is from Dougherty and Rysiew (forthcoming).
123
Synthese
that’s obviously contradictory.17 We should ask whether, e.g., utterances of 6.3 and 6.4 are cases of felicitous assertions of contradictions, or whether they’re felicitous because they’re not contradictory. Unfortunately for Dougherty, Rysiew, Vogel, and the like, a good case can be made that the best explanation of the felicity of 6.1–6.4 is not that they’re felicitous because true, and therefore not contradictory. When I’m in a frame of mind where I’m disposed to utter one of 6.1-6.4, I’m just as apt to utter one of the following: 6.1 . I’m sure they’re going to lose, but I’m going to carry on watching just in case. 6.1 . There’s no way they’re going to lose, but I’m going to carry on watching just in case. 6.2 . I’m sure it’s not going to rain, but still, I’m going to take an umbrella just in case. 6.2 . There’s no way it’s not going to rain, but still, I’m going to take an umbrella just in case. 6.3 . The referee says, ‘I’m sure no one will get confused about this, but it’s possible someone will’. 6.3 . The referee says,‘There’s no way anyone will get confused about this, but it’s possible someone will. 6.4 . I’m sure I have hands, even though it’s possible I’m a handless brain in a vat. 6.4 . There’s no way I don’t have hands, even though it’s possible I’m a handless brain in a vat. To me all of these are just as felicitous as 6.1–6.4. However, I think they’re clearly contradictory, or at least it’s clear their speakers aren’t committed to their literal truth. In 6.1 , 6.2 , 6.3 and 6.4 the speaker is saying s/he’s sure—i.e., certain—that a possibility p won’t obtain, even though s/he acknowledges p might obtain. To be certain that p won’t obtain, is to be certain that no not- p possibility will obtain—i.e., to think there’s no chance a not-p possibility will obtain.18 But if she thinks that a not- p possibility might obtain, she thinks there’s a chance a not-p possibility will obtain.19 So in uttering 6.1 , 6.2 , 6.3 and 6.4 the speaker seems to say that there is no chance that a not- p possibility obtains, while seemingly acknowledging that there is chance a not- p possibility obtains. Contradiction. And the contradiction seems even more explicit in 6.1 , 6.2 , 6.3 and 6.4 . To say that there’s no way something will happen is just another way of saying it’s not possible it will happen. So in 6.1 , 6.2 , 6.3 and 6.4 the speaker is apparently saying that a certain possibility p isn’t possible, while at the same time acknowledging that p is possible. Thus although 6.1–6.4 may all be felicitous, I don’t take that to be good evidence that they’re not also contradictory. There are other assertions that are very similar to 17 Likewise, from the fact that speakers can felicitously utter S, and Grice’s Maxim of Quantity, we shouldn’t necessarily infer that S doesn’t have a trivial content. 18 I’m going to talk about the speaker thinking that there is [not] a chance that a possibility will/won’t obtain. The real thought, though, is about the subject’s credence in a particular possibility. I do this for ease of exposition—nothing turns on it. 19 This follows if we assume that it’s not the case that not- p is an epistemic possibility that has an epistemic probability of 0. Strictly speaking, there are such epistemic possibilities, but they’re rare and involve very unusual scenarios. I’m assuming that we can ignore them when interpreting speakers who utter 6.1 –6.4 .
123
Synthese
them which are contradictory, but just as felicitous. Indeed, I think we even have at least some evidence that 6.1–6.4 are contradictory here. It’s not just that, say, 6.1 is very similar to 6.1 and 6.1 . As I said earlier, when I’m in a frame of mind where I’m likely to utter 6.1, I would also be likely to utter 6.1 or 6.1 . In a real life situation, I would take any of the three as just as good as any of the others as getting across to my audience what I want to get across. But the literal semantic content of 6.1 and 6.1 is a contradictory proposition. I think this is evidence that 6.1 is too. 7 Conclusion: where should fallibilists go from here? I’ve argued we shouldn’t resort to contextualism to solve the problem of CKAs. I also argued that Dougherty and Rysiew’s attempt to reconcile fallibilism with the infelicity of CKAs fails, but that they’re also right to regard the infelicity of CKAs as a problem for fallibilism. Fallibilists should follow Dougherty and Rysiew and try to explain the infelicity pragmatically, and with an invariantist semantics for ‘knows’ and its cognates. I will conclude by suggesting how future fallibilists might try to give such a pragmatic explanation. We have seen that there are two things for fallibilists to explain pragmatically. First, there’s the infelicity of CKAs. The problem here is that when we treat what is epistemically possible as a function of what is known, as the received view of epistemic possibility says we should, we get the result that CKAs are contradictory. And this seems inconsistent with fallibilism. Given this fact, I think fallibilists should follow Dougherty and Rysiew in saying that what is epistemically possible is a function of evidence rather than knowledge. This leads to the second thing fallibilists must now explain, namely the fact that 5.2-5.4 are infelicitous too. What this points to is that speakers treat ‘For all I know, p’ and ‘ p might be true’ as more or less equivalent, or very close to one another in meaning. This can’t be explained if we say that what’s epistemically possible is a function of evidence, and not a function of knowledge. Fallibilists must explain this second thing too. When we keep in mind the fact that both of these things need to be explained, a constraint on how our explanation of the infelicity of CKAs should go suggests itself. The implicature in the CKA should come not from the conjunct ascribing knowledge, but from the conjunct that claims something is epistemically possible. We can only hope that our explanation of the infelicity of CKAs will generalize to explain the second desideratum too if it meets this constraint. For if we want to explain why we regard ‘ p is possible’ as nearly equivalent to ‘For all I [we] know, p’, the fact that an ascription of knowledge typically pragmatically conveys non-semantic information is not going to be able to do any work for us, since neither sentence is a knowledge ascription. The only thing both desiderata have in common is that both contain a claim about something’s being epistemically possible. So a unified treatment of both facts is going to have to have this feature do the pragmatic work. Incidentally, this suggests a fatal flaw in the earlier well-known attempt by Rysiew 2001 to give a pragmatic explanation of the infelicity of CKAs. For in this earlier theory, Rysiew had the knowledge ascription convey the key non-semantic information. Rysiew’s account went like this. When we ascribe knowledge of p to S, we pragmatically convey that S can rule out all the salient alternatives to p, even though
123
Synthese
it’s not part of the truth-conditions of knowledge ascriptions that S is in a position to do this. And when S utters ‘I knows p, but it’s possible that q’, S obviously will make the possibility that q salient. So if q is obviously an alternative to p (i.e., it obviously entails not- p), then S will implicate that s/he can rule out the possibility that q, even though at the same time S is claiming that q is possible for her. Since this account has the knowledge conjunct do all the pragmatic work, it won’t generalize to explain why 5.2–5.4 are infelicitous, and so the fallibilist should look for a different pragmatic explanation of the infelicity of CKAs. A more promising approach would start with the thought that an utterance of ‘ p is possible’ implicates that p is an epistemic possibility that the members of the context of utterance need to take seriously (as Dougherty and Rysiew point out, p. 9). Such an implicature should follow from Grice’s Cooperative Principle (CP), for it seems it wouldn’t be relevantly informative to mention that p is an epistemic possibility unless it was a possibility worth taking seriously. Given that epistemic possibility claims carry this implicature, an assertion of ‘I know p, but it’s possible that q’ (where q obviously entails not- p) would implicate that there is an alternative to p that needs to be taken seriously. But that this conflicts with the claim that the utterer knows p doesn’t conflict with fallibilism. For it also conflicts with a fallibilist theory that claims that in order to know p one needs to be able to rule out all the alternatives to p which need to be taken seriously, but not all the alternatives to p simpliciter. Turning to the second desideratum, again appealing to CP we point out that ‘For all I know, p’ pragmatically conveys the information that the fact that p is consistent with our knowledge is significant and worthy of our consideration. Perhaps in conveying this information, we would thereby also convey the information that the fact that p is epistemically possible is a fact worthy of consideration (more on this shortly)—in other words, we would implicate that p is a possibility that needs to be taken seriously. And this, we have seen, is also conveyed by an assertion of ‘It’s possible that p’. Thus we also have an explanation of the fact that speakers take ‘For all I know, p’ as expressing more or less the same thought as ‘It’s possible that p’. If fallibilists are going to give a successful unified pragmatic explanation of both facts, it really seems to me that the explanation would have to be something like the one I just sketched. But I will now close by presenting two misgivings I have about the approach. I don’t think these objections are decisive. But they do present important problems someone giving an explanation along these lines would need to respond to. First, in explaining the second desideratum I started by pointing out that by CP ‘For all I know, p’ should pragmatically convey the information that it’s a significant fact that p is consistent with my knowledge. So far so good. But then I said, “Perhaps in conveying this information, we would thereby also convey the information that the fact that p is epistemically possible is a fact worthy of consideration.” In saying this I was assuming that from it’s a significant fact that p is consistent with my knowledge we can infer that it’s a significant fact that p is epistemically possible. But this inference would only be valid if what is epistemically possible was really a function of our knowledge, which is the very view of epistemic possibility we have seen fallibilists must deny. Note what the premise says is a significant fact: that p is consistent with my knowledge. From this we conclude that it’s also a significant fact that p is epistemically possible But we’ve seen fallibilists need to deny the validity of this inference.
123
Synthese
The above problem seems to me the really serious one. But here’s a second misgiving. I don’t know what the right semantics for ‘strictly speaking’ are. However, ‘Strictly speaking, the hangar’s not empty’ seems to convey that although by loose standards we might count the hangar as being empty, it’s not empty by strict standards. In many contexts ‘The hangar’s empty’ will be considered true if there were no airplanes in it, even though there were tools in it and stuff like that. Similarly, ‘Strictly speaking, p is possible’ conveys the information that the p-possibility may be very remote, but it’s still a possibility nonetheless. Prefixing ‘Strictly speaking’ to ‘ p is possible’, it seems to me, encourages one’s audience to look for remote possibilities that make the possibility claim true. Doing so encourages one’s audience to look for remote possibilities that are epistemically possible, not just those epistemic possibilities that we should take seriously. So it seems that prefixing ‘ p is possible’ with ‘Strictly speaking’ should have the tendency to cancel any implicature that p is a possibility that needs to be taken seriously. But now note that just as speakers will typically take ‘ pis possible’ as basically equivalent to ‘For all we know, p’, they will also take assertions of the following as being basically equivalent (even if the second is cumbersome): 7.1 Strictly speaking, p is possible. 7.2 Strictly speaking, for all we know, p. That speakers will take 7.1 and 7.2 to be basically equivalent is a problem for the current proposal. When speakers are encouraged to focus on the fact that p is merely strictly speaking epistemically possible, rather than the claim that p is a possibility that they need to take seriously, they are still just as willing to think that this claim is equivalent to the claim that p is consistent with what they know as they were before. So the approach just sketched does have some difficulties. Nevertheless, it seems like the most promising approach for fallibilists. I, at least, can’t think of another way fallibilists could give a unified explanation of the two facts we’ve seen they need to explain pragmatically. I am personally doubtful that fallibilists can give such a unified explanation. However, I hope to have provided some help to a fallibilist who might try to accomplish this feat in the future, and to have shown why fallibilists need to see accomplishing it as something that needs to be done.20 References Brown, J. (2006). Contextualism and warranted assertability manoeuvres. Philosophical Studies, 130, 407–435. doi:10.1007/s11098-004-5747-3. Dell, M., Jones, B. F., & Olken, B. A. (2008). Climate change and economic growth: Evidence from the last half century. NBER Working Paper No. 14132. DeRose, K. (1991). Epistemic possibilities. The Philosophical Review, 100, 581–605. doi:10.2307/2185175. DeRose, K. (1996). Knowledge, assertion and lotteries. Australasian Journal of Philosophy, 74, 568–580. doi:10.1080/00048409612347531. 20 Some of the arguments I give in this paper I presented at a seminar at the University of St Andrews in December 2007, and I am grateful to all the participants there for their critical feedback. The feedback from Torfinn Huvenes and Crispin Wright was especially helpful. I am also grateful for the comments of my two anonymous reviewers from this journal, which led to significant improvements.
123
Synthese DeRose, K. (2006). “Bamboozled by our own words”: Semantic blindness and some arguments against contextualism. Philosophy and Phenomenological Research, 73, 316–338. doi:10.1111/j.1933-1592. 2006.tb00620.x. Dougherty, T., & Rysiew, P. Fallibilism, epistemic possibility, and concessive knowledge attributions. Philosophy and Phenomenological Research (forthcoming). Egan, A. (2007). Epistemic modals, relativism, and assertion. Philosophical Studies, 133, 1–22. doi:10. 1007/s11098-006-9003-x. Egan, A., Hawthorne, J., & Weatherson, B. (2005). Epistemic modals in context. In G. Preyer & G. Peter (Eds.), Contextualism in philosophy. Oxford: Oxford University Press. Fantl, J., & McGrath, M. (2002). Evidence, pragmatics, and justification. The Philosophical Review, 111, 67–94. Gillies, A., & von Fintel, K. (2008a). CIA leaks. The Philosophical Review, 117, 77–98. doi:10.1215/ 00318108-2007-025. Gillies, A., & von Fintel, K. (2008b). An opinionated guide to epistemic modals. In T. Gendler, et al (Eds.), Oxford studies in epistemology (Vol. 2). Oxford: Oxford University Press. Grice, H. P. (1989). Logic and conversation. In H. P. Grice (Ed.), Studies in the way of words. Harvard University Press. Groenendijk, J., & Stockhof, M. (1984). Studies on the semantics of questions and the pragmatics of answers. Ph.D. Dissertation, University of Amsterdam. Groenendijk, J., & Stockhof, M. (1989). Type-shifting rules and the semantics of interrogatives. In G. Chierchia, B. Partee, & R. Turner (Eds.), Properties, types and meanings. Volume 2: Semantic issues. Kluwer Academic Publishers. Reprinted in Portner, P., & Partee, B. (Eds.). (2002). Formal semantics: The essential readings. Blackwell Publishing. Hacking, I. (1967). Possibility. The Philosophical Review, 76, 143–168. doi:10.2307/2183640. Hawthorne, J. (2004). Knowledge and lotteries. Oxford: Oxford University Press. Hill, C., & Schechter, J. (2007). Hawthorne’s lottery puzzle and the nature of belief. Philosophical Issues, 17, 102–122. doi:10.1111/j.1533-6077.2007.00125.x. Hintikka, J. (1962). Knowledge and belief. Cornell University Press. Kratzer, A. (1977). What ‘must’ and ‘can’ must and can mean. Linguistics and Philosophy, 1, 337–355. doi:10.1007/BF00353453. Lewis, D. (1996). Elusive knowledge. Australasian Journal of Philosophy, 74, 549–567. doi:10.1080/ 00048409612347521. MacFarlane, J. Epistemic modals are assessment-sensitive. In A. Egan, et al (Eds.), Epistemic Modality (forthcoming). Rysiew, P. (2001). The context-sensitivity of knowledge attributions. Nous (Detroit, Mich.), 35, 477–514. doi:10.1111/0029-4624.00349. Schaffer, J. (2004). From contextualism to contrastivism. Philosophical Studies, 119, 73–103. doi:10.1023/ B:PHIL.0000029351.56460.8c. Schaffer, J. The contrast-sensitivity of knowledge ascriptions. Social Epistemology (forthcoming a). Schaffer, J. Knowledge in the image of assertion. Philosophical Issues, 18 (forthcoming b). Schiffer, S. (1996). Contextualist solutions to scepticism. Proceedings of the Aristotelian Society, 96, 317–333. Stalnaker, R. (1978). Assertion. In Syntax and semantics (Vol. 9). Academic Press. Stalnaker, R. (1998). On the representation of context. Journal of Logic Language and Information, 7, 3–19. doi:10.1023/A:1008254815298. Stanley, J. (2004). On the linguistic basis for contextualism. Philosophical Studies, 119, 119–146. doi:10. 1023/B:PHIL.0000029353.14987.34. Stanley, J. (2005a). Knowledge and practial interests. Oxford: Oxford University Press. Stanley, J. (2005b). Fallibilism and concessive knowledge attributions. Analysis, 65, 126–131. doi:10.1111/ j.1467-8284.2005.00537.x. Stephenson, T. (2008). Judge dependence, epistemic modals, and predicates of personal taste. Linguistics and Philosophy, 30, 487–525. Teller, P. (1972). Epistemic possibility. Philosophia, 2, 303–320. doi:10.1007/BF02381591. Vogel, J. (1999). The new relevant alternatives theory. Philosophical Perspectives: Epistemology, 13, 155–180. Williamson, T. (2000). Knowledge and its limits. Oxford: Oxford University Press.
123
