Knowledge and epistemic necessity

Philos Stud (2012) 158:493–501 DOI 10.1007/s11098-012-9960-1 Knowledge and epistemic necessity John Hawthorne Published online: 6 June 2012  Springer Science+Business Media B.V. 2012 Abstract Claims of the form ‘I know P and it might be that not-P’ tend to sound odd. One natural explanation of this oddity is that the conjuncts are semantically incompatible: in its core epistemic use, ‘Might P’ is true in a speaker’s mouth only if the speaker does not know that not-P. In this paper I defend this view against an alternative proposal that has been advocated by Trent Dougherty and Patrick Rysiew and elaborated upon in Jeremy Fantl and Matthew McGrath’s recent Knowledge in an Uncertain World. Keywords Knowledge
Epistemic modality
Epistemology 1 Introduction Claims of the form ‘I know P and it might be that not-P’ tend to sound odd. One natural explanation of this oddity is that the conjuncts are semantically incompatible: in its core epistemic use, ‘Might P’ is true in a speaker’s mouth only if the speaker does not know that not-P. This semantical property can also form the basis for an explanation of why claims of the form ‘P but it might be that not-P’ sound odd, at least assuming that knowledge functions as a norm for assertion. In this case, the conditions for the truth of (and hence felicitous assertion of) the right conjunct are incompatible with the conditions for felicitiously asserting the left. This approach to ‘might’ claims—call it the standard approach—is one that many, including myself, are very sympathetic too. Recently, Trent Dougherty and Patrick Rysiew have advocated a competing idea, one that has been endorsed and elaborated upon in Jeremy Fantl and Matthew McGrath’s recent Knowledge in an Uncertain World (Fantl and McGrath 2009). J. Hawthorne (&) Magdalen College, University of Oxford, Oxford OX14AU, UK e-mail: john.hawthorne@philosophy.ox.ac.uk 123 494 J. Hawthorne Suppose that knowing that P is, pace the remarks above, compatible with the epistemic possibility that not-P. But suppose, in asserting that it might be that P, one typically communicates (without semantically expressing) the claim that there is a significant chance that not-P (since it otherwise wouldn’t be worth bothering to say that there is a chance that P) Suppose that, further, knowledge is not compatible with a significant chance that not-P.1,2 In that case, while claims of the form ‘I know P and it might be that not-P’ are often true, they are still not assertible.3 In this note I shall explain why a story of this sort is unlikely to prove satisfying. None of the considerations that I offer below are decisive. But they seem to me to provide a pretty strong case for the standard approach. 2 Some ground-clearing Those who run a semantics for ‘might’ in terms of knowledge ought and typically do recognize that there are various other uses of that expression that merit a different treatment. Most obviously, certain uses of ‘might’ combined with ‘have’ express metaphysical possibility.4 But even if we focus on simple combinations of ‘might’ with a verb, a variety of treatments are required suppose there is a bucket with a mix of single and double-headed coins. I pick one out, not knowing what coin I have picked. I bet on heads. I say ‘I don’t know whether I might lose’. Now, quite obviously, for all I know, the coin will land tails. The uncertainty I have expressed, makes no sense if ‘I might lose’ is tantamount to ‘for all I know I will lose’. Similarly, suppose I have a crushing position in a chess match and announce ‘I can see at least three winning combinations that I might choose. But I know which combination I’m going to choose—it involves a flashy, crowd pleasing, rook sacrifice.’ In these settings, ‘might’ requires a different treatment. Those who think that the core epistemic use of ‘might’ is connected to knowledge will thus 1 This account will naturally try to play up analogies to standard cases of implicature: just as there wouldn’t typically be any point saying ‘There is a gas station nearby’ if one know that it was closed, there wouldn’t typically by an point in satiying ‘There is a chance that not-P’ if one knew that the chance was not realized in actuality. 2 I am not going to fuss too much about the question how, in detail, ‘significant’ is to be construed. At one point Dougherty and Rysiew gloss it this way: ‘one has real grounds for supposing that not-p might be the case and one (therefore) isn’t confident that p’. This strikes me as unsatisfactory for a variety of reasons. For one thing it uses ‘might’ in the analysis. If ‘might’ expresses ‘significant chance’ the analysis is circular, meanwhile, if ‘might’ expresses bare chance, then, by the authors lights, grounds for thinking that not-p might be the case will be grounds for thinking that one’s evidence does not logically entail the conclusion. But such grounds need hardly render one less than confident that p. For as the authors are thinking about things, a failure of evidence to entail P is commonplace even when one knows P. Fantl and McGrath gloss on significance is arguably more promising—a chance is significant when it is not idle, which in turn means that it renders it inappropriate to ‘put p to work’ as a basis for belief or action (op. cit., pp. 22–23). However it is arguable that the best gloss for the purposes of the proposal is even more flatfooted: a chance that P is significant when one is not in a position to know that it is not actualized. 3 Assuming knowledge is the norm of assertion, this idea can (as Fantl and McGrath emphasise) also account for the unassertability of ‘P and it might be that not-P’. 4 Some uses of ‘might have’ pertain to temporal distance, others to modal distance. The latter uses involve ‘fake past tense’ (See Iatridou 2000). 123 Knowledge and epistemic necessity 495 distinguish what we shall call ‘flat-footed’ uses of ‘might’, ‘possibly’ and ‘could’, where the truth turns on what the speaker knows, from a range of other uses which require a different treatment. Those who do not gloss epistemic possibility in terms of knowledge will likely make a similar distinction, though they will tell a different story about how facts about epistemic perspective determine truth in the default— what I have called ‘flat-footed’—use. One such story—endorsed by Dougherty and Rysiew but not by Fantl and McGrath—is that ‘Might P’ is true at a context iff the speaker’s evidence does not obviously entail not-P. Now one thing that makes it hard to adjudicate debates about the standard package rather tricky is that there is no surefire diagnostic for ‘flat-footed’ uses. Consider the following speech, made in a context where we are trying to figure out which movie theatre to search in order to find a certain couple. ‘There are five movies showing at the movie theatre. So lets consider in turn the five movies that it is possible that they are at. One of them is Nightmare on Elm Street 7, and we know that they wouldn’t go to see that….’ The speech sounds perfectly acceptable. But it is a speech in which one says it is possible that the couple is at a certain movie and also that one knows the couple are not at it. The standard package will almost certainly want to say that this is not a flat-footed use of ‘possible’. But if one adopted an alternative construal of flat-footed uses, then it may be open to one to adopt a flat-footed construal in this case. What all this shows is that, owing to the semantic flexibility of ‘might’ and ‘possibly’, the standard package is not going to be undone simply by pointing to sample speeches where one says something tantamount to ‘I know P and possibly/it might be that not-P’. 3 ‘Might’ and ‘just in case’ There are contexts is where it is pretty clear that one is rather half-hearted, or at least less than fully serious, when one says ‘I know P’. One kind of case of this sort is one where one uses ‘I know P’ as an expression of pessimism: ‘I know this isn’t going to work’ can be said in a tone where it is pretty clear that one isn’t seriously claiming to know that the undertaking isn’t going to succeed. Here one is merely expressing pessimism, albeit in a slightly exaggerated way. (One might similarly exaggerate using ‘I’m sure’: ‘I’m sure this isn’t going to work). Perhaps one very briefly lets one’s pessimism lull one into thinking one knows the relevant content. But such conviction is obviously at best half-hearted. An interlocutor will know that if one presses the speaker as to how they know and whether they really know and so on, the speaker will very quickly retract. Another sort of context where one’s commitment to know is less than fully serious is one where one uses ‘know’ to express reassurance. It is easy enough to imagine cases when uses ‘I know you are going to be all right’ to reassure a sick friend. Here again it is pretty clear that one isn’t very serious about the claim to know. Perhaps one’s efforts at reassurance induce a kind of mutual optimism that lulls both speaker and hearer into briefly taking themselves to know everything will be all right. But even so, the speaker is obviously not seriously undertaking to defend the claim that she knows the audience will be all right. These kinds of cases—where the claim to know is very half- 123 496 J. Hawthorne hearted—are ones where it is particularly easy to transition to claims that signal that one’s all-things considered judgment is that the relevant content might be false. ‘I know this isn’t going to work. But in case it does….’ ‘I know the operation will be a success. But just in case it isn’t….’ But it is worth noting that even in these halfhearted contexts, an explicit claim that the relevant content might be false sounds extremely awkward. ‘I know this isn’t going to work. But in case it does…’ and ‘I know the operation will be a success. But just in case it isn’t…’ sound a lot better than ‘I know this isn’t going to work. But it might work’ or ‘I know the operation will be a success. But it’s possible that it won’t be.’ The standard approach has a very natural explanation of this contrast. It would defeat the point of the pessimistic or reassuring speech to explicitly contradict that speech with one’s very next sentence. One cannot keep up the combination of a half-hearted commitment to know P combined with an all things considered judgment that one does not know P when one outright asserts something incompatible with knowing P. Meanwhile, the ‘just in case’ speech does not outright assert anything incompatible with a claim to know, and so it can leave the relevant combination of attitudes intact. By contrast, I don’t see that the Dougherty/ Rysiew story has a natural account of the contrast in felicity between ‘just in case’ and ‘might’. On their story, neither the ‘might’ follow-up nor the ‘just in case’ follow-up explicitly contradicts the knowledge speech; both at best pragmatically communicate something incompatible. So why the marked contrast in felicitly conditions? 4 Some spurious data Whatever we think about the relationship between ‘know’ and ‘might’, there is no disputing the fact that, holding the context fixed, ‘might p’ and ‘must not-P’ are incompatible, as are ‘might p’ and ‘it couldn’t be that P’. If it is claimed that the acceptability of a certain kind of speech is problematic for the standard package, one should look at analogous speeches where ‘know P’ is replaced by ‘must P’ or by ‘It couldn’t be that not-P’. If the resulting speech is still acceptable, then the original data is likely to be dubious. This provides one test that discounts certain pieces of putative evidence against the standard approach. Thus consider the following speech, which Fantl and McGrath offer as prima facie evidence against the standard account: 1. The possibility that not-p is ridiculous and not worth considering. I know that p.5 To be fair to them, they are ambivalent about the acceptability of these speeches. But what they don’t notice in any case, is that to the extent that they sound natural, speeches like the following are also natural: 5 Op. Cit. p. 21. 123 Knowledge and epistemic necessity 2. 3. 497 The possibility that he is not in the pub is ridiculous and not worth considering. He must be in the pub. The possibility that he is not in the pub is ridiculous and not worth considering. He couldn’t be anywhere else. Given that no one will be tempted to use 2 and 3 to try to show that (at a context) ‘might not-p’ is compatible both with ‘must p’ and ‘couldn’t be that not-p’, we shouldn’t be using 1 against the standard package. Consider also their 4. Of course there is some chance that I’m wrong, anything is possible, but I know P.6 One can certainly get away with this speech, especially with intonational focus on ‘know’ (inserting a slight hedge—as in ‘I reckon I know’—makes it sound even better). But one can also get away with 5. 6. The answer has to be/must be 52. Of course there is some chance that I’m wrong, anything is possible. Of course there is some chance that I’m wrong, anything is possible. But that view can’t be right for reasons x, y and z. (Analogously, 6 is improved by intonational focus on ‘can’t, and ‘But I reckon that view can’t be right’ is even better.) 5 Cancellation data The Dougherty/Rysiew proposal predicts that making it explicit that the chance of P is insignificant should drastically improve certain speeches in which one admits to a chance of not-P while claiming knowledge of P. Here the implicature of ‘might’ that is integral to the proposal would be cancelled. If the infelicity of ‘I know P and might not-P’ turns on such an implicature, this cancellation should induce felicity. So, for example, one would expect ‘There is a chance that he will not show up, but it’s not significant—it is very small indeed. I know he will show up’ to be just fine. But to my ear, at least, it still sounds dicey. Once one has made salient that there is a chance, the fact that it is insignificant or not worth worrying about doesn’t appear to put one in the comfort zone when it comes to making a knowledge claim. I am of course perfectly well aware there are contexts where ‘It is possible that not-P’ and ‘I know P’ are coassertable. One obvious way they might be coassertable at a context is because ‘possible’ is being used in a non-flat-footed way that is neutral on whether one knows P. The point I am pressing is that in a context where ‘It is possible that P’ and ‘I know not P’ are not coassertable, matters are not much improved by adding ‘but the probability is very small indeed’. (While I don’t not wish to make too much of the point, it is even worth remembering that there are situations where we wish to say ‘it might be that P’, 6 Op. Cit. p. 21. 123 498 J. Hawthorne where the chance of P is zero, but where reminder of the latter fact does nothing to alleviate a sense that one does not know not-P. Suppose someone is throwing a dart at random at a dartboard with continuum many points. It is natural to say ‘It might hit the bulls eye’. And if the proclivities of the dart thrower are not weighted in any way, then we are compelled to say that the chance of hitting any particular point is zero.7 But still we are not comfortable saying in such a context that we know it won’t hit the bulls eye.) 6 Inferences about absence of epistemic possibility According to the picture under consideration, knowledge is typically accompanied by an epistemic possibility that not-P, albeit an insignificant possibility. Quite obviously this picture has to explain why we are typically quite willing to say ‘It couldn’t be that not-P’ in a situation where we take ourselves to know P. In response to an earlier observation of mine to that effect, Dougherty and Rysiew offer the natural suggestion that such speeches, while literally false, are ‘communicatively effective hyperbole’ (p. 130). The idea is that in saying ‘It couldn’t be that P’ one semantically expresses the false proposition that there is no epistemic chance that P while communicating the true proposition that there is no significant epistemic chance that P. One worry about this suggestion is that it fails to explain the naturalness of certain inferences. In particularly, we are perfectly willing to assert ‘It couldn’t be that P or Q’ on the heels of ‘It couldn’t be that P’ and ‘It couldn’t be that Q’. The picture at hand would in no way predict any such willingness. The problem isn’t that the inference is semantically invalid on such a view.8 The problem is that if the point of saying ‘It couldn’t be that P’ is in generally to communicate that there is no significant chance that P, then we ought be to be rather hesitant to move from ‘It couldn’t be that P and ‘It couldn’t be that Q’ to ‘It couldn’t be that P or Q’ in rather the way that we would be hesitant to move from ‘There is no significant chance that P’ and ‘There is no significant chance that Q’ to ‘There is no significant chance that P or Q’. Suppose that there is no significant chance that P and no significant chance that Q but there is a significant chance that P or Q’. Then the speech ‘There is no chance that P or Q’, would not merely be false, but it would not even have the virtue of communicating the truth that there is no significant chance that P or Q’. (Granted, there are defensive manoevers that could be made by the view I am attacking: for example, it might be said that we are typically disposed to make the inference because we are either typically blind to the fact that insignificant chances add up to significant ones or else because it is rare that we pay 7 Note in this connection that is problematic to claim that if the epistemic chance that P is one, then it must be that P. For this would make the construction ‘it must be that P’ non-factive. 8 I should note in passing though that one of Dougherty and Rysiew’s glosses of epistemic possibility— where q’s being epistemically possible for S’ is glossed as ‘q as non-negligible probability on S’s total evidence’—renders this inference semantically invalid, at least if certain tiny values [0 count as negligible. They also don’t notice that if epistemic necessity is to be factive then in certain infinite cases, we shall have to allow that q is epistemically possible while having epistemic probability zero. 123 Knowledge and epistemic necessity 499 attention to a pair of insignificant chances that add up to significant ones. But at the very least, the view has some explaining to do here.) 7 Some logically complex constructions Dougherty and Rysiew’s picture is not multi-premise closure friendly. Quite obviously, whatever threshold is picked out by ‘significant chance’ there can be a significant chance that a conjunction is false even though there is no significant chance that one of the conjuncts is false. One might thus naturally expect cases to arise where one knows each of a set of propositions but, even having deduced their conjunction, one does not know their conjunction. For many of us this is a significant cost of their kind of story. But even if we are not wedded to multi premise closure, there is a related worry about their package. They offer an explanation of why ‘I know P and it might be that not-P’ is unassertable. But they do not look at a range of logically more complex constructions. Of particular relevance here are constructions in the vicinity of ‘I know that P and I know that Q but it might be that Not (P or Q).’ Note to begin that claims like this sound just as infelicitous as ‘I know P and it might be that not-P’. Consider for example the awkwardness of ‘I know he will show up and I know she will show up but one of them might not’ or ‘I know of each of the people on the list that they are innocent but it might be that one of them is guilty’. It seems to me that their story does not provide a good explanation of these data. For any threshold above zero that ‘significant chance’ might denote, it may be that there is a significant chance that one of the people on the relevant list is innocent but that for each person, there is no significant chance that that person is innocent. Hence, in communicating that there is a significant chance that one of the people on the list is innocent one does not thereby communicate that there is a person such that there is a significant chance that that person is innocent. Hence the story would seem to predict that in communicating the proposition that there is a significant chance that someone on the list is guilty, one would not thereby challenge the truth of the claim that for each person, one knows that person is innocent. The infelicity of the more complex constructions that I have been looking at remains unexplained. (Again, there are defensive manoevers available. The proponent of the view might say that ordinary people think that multi premise closure is true even though it isn’t and that the awkwardness of ‘I know he will show up and I know she will show up but one of them might not’ is to be explained by the implicature of significant chance in combination with that faulty commitment.) 8 A questionable impulse If the core notion of epistemic necessity isn’t to be defined in terms of knowledge, what is its proper grounding? We have already encountered Dougherty and Rysiew’s idea—that it is determined by the subject’s evidence. (As they are aware, if it is to draw different lines from the standard package, this proposal needs to be 123 500 J. Hawthorne supplemented by a rejection of Timothy Williamson’s hypothesis that one’s evidence consists of all and only the propositions that one knows.) For their part, Fantl and McGrath’s are guided by some notion of being maximally justified, which in turn would underwrite cavalier rational betting behavior: When one is maximally justified that P it is rational to bet on P at any odds.9 And one can well imagine various other demanding standards: For it is to epistemically necessary that P it needs to be absolutely clear that P. Or: For it to be epistemically necessary that P it needs to be clear and distinct that P. In each case, if we hold the supplementary hypothesis that knowledge does not require that the demanding standard be met, we get a new candidate grounding for epistemic necessity. I suspect that the temptation to shift to one of these alternative groundings for epistemic necessity is produced by impulses that, once acquiesced in, produce no coherent resting place. It bears emphasis here that whatever, say, maximal justification comes to, it is unlikely that its presence and absence will be luminous.10 Now in any case where someone is maximally justified but doesn’t know it, we will vacillate about whether it is rational for them to bet the farm against a penny. But even more importantly, if they are maximally justified, know they are maximally justified, but are so constructed that in various not so far off scenarios where they are not maximally justified they are unable to know that they are less than maximally justified, then we will still vacillate in our judgments about how they ought to act in the cases where they are in fact maximally justified. Indeed, I think that whatever the strength of one’s epistemic position with respect to P, recognition of the anti-luminosity of the absence of that strength will produce an impulse to say ‘there is a chance that I am wrong about whether P’.11,12 Suppose, for example, it is clear and distinct that P (using Descartes’ idiom) or absolutely clear that P (using Unger’s) but that I am a being who can’t always tell when something is less than clear and distinct or less than absolutely clear. Recognition of that liability will inevitably produce some impulse to say ‘There is a chance I’m wrong’.13 If one wants to build a notion of epistemic chance around that impulse, ‘maximal justification’, ‘absolutely clarity’ and ‘clearness and distinctness’ will not serve to 9 Op. Cit., p. 13. Of course, as they are aware, no ‘internalist’ non-factive justification condition can be sufficient for epistemic necessity, since ‘Must P’ is factive. 10 Here I am using ‘luminous’ in Timothy Williamson’s sense: a condition is luminous just in case, whenever it obtains one is in a position to know that it obtains. See Williamson (2000). 11 Some philosophers talk as if we can make good sense of a scale of strengths of epistemic position, where even people that all know P can be ranked according to their strength of position with respect to P. I am very skeptical of the existence of such a scale, but I will not pursue the matter here. 12 And since doubts generated on this basis seem always at least excusable it does not seem promising to definite the relevant ideal as ‘case where a doubt is excusable’. 13 Note that even if one it is absolutely clear and it is absolutely clear that it is absolutely clear, one will find an impulse to say ‘There is a chance that I am wrong’ once one has taken stock of the fact that one is so constructed that there are none too exotic cases where a proposition is false, where it is less than absolutely clear, where it is not absolutely clear that it is less than absolutely clear, and indeed where one is extremely confident in that ‘bad’ case that it is absolutely clear. 123 Knowledge and epistemic necessity 501 underwrite epistemic necessity any more than knowledge.14 One upshot is that it is misguided to try craft a notion of epistemic necessity according to which, if it is epistemically necessary that P then any doubt about P would be inexcusable or utterly unreasonable. But a more general lesson may be that the impulses that lead one to think that knowledge is not good enough for epistemic necessity or maximal epistemic probability are dangerous impulses. There may be no cogent conception of a condition that satisfies the demands that such impulses invite. References Fantl, J., & McGrath, M. (2009). Fallibilism, epistemic possibility and concessive knowledge attributions,’ philosophy and phenomenological research, In Knowledge in an uncertain world. Oxford: Oxford University Press. Iatridou, S. (2000). The grammatical ingredients of counterfactuality. Linguistic Inquiry, 31(2), 231–270. Williamson, T. (2000). Knowledge and its limits, Chapt. 3. Oxford: Oxford University Press. 14 One other consideration that militates towards the same conclusion is that ‘must’ is a modal concept and as such will have whatever logical structure is common to all uses of ‘must’. Arguably this will, inter alia, require it to satisfy at least the minimal principles of normal modal logic, according to which if something is a logical truth, then it must be that P, and if must (P . Q) then (must P . must Q). But this will generate all sorts of epistemic necessities where a doubt is perfectly reasonable. For it might be perfectly reasonable or excusable to doubt some logical truths—say the law of excluded middle; and in a case where one knows that P, knows that P . Q but has not yet performed the inference from those two premises to Q, a doubt about Q at that point may be perfectly understandable. Granted many will react to this by claiming that the epistemic ‘must’ does not satisfy the minimal principles of normal modal logic. But it may be that this risks making ‘must’ too unwieldy and gerrymandered (in ways analogous to the suggestion that epistemic probabilities do not satisfy the Kolmogorov axioms). Obviously there is much more to be said here. 123