Abstract Thought experiments invite us to evaluate philosophical theses by making judgements about hypothetical cases. When the judgements and the theses conflict, it is often the latter that are rejected. But what is the nature of the judgements such that they are able to play this role? I answer this question by arguing that typical judgements about thought experiments are in fact judgements of normal counterfactual sufficiency. I begin by focusing on Anna-Sara Malmgren’s defence of the claim that typical judgements about thought experiments are mere possibility judgements. This view is shown to fail for two closely related reasons: it cannot account for the incorrectness of certain misjudgements, and it cannot account for the inconsistency of certain pairs of conflicting judgements. This prompts a reconsideration of Timothy Williamson’s alternative proposal, according to which typical judgements about thought experiments are counterfactual in nature. I show that taking such judgements to concern what would normally hold in instances of the relevant hypothetical scenarios avoids the objections that have been pressed against this kind of view. I then consider some other potential objections, but argue that they provide no grounds for doubt. Thought experiments are widespread in philosophy. Hypothetical situations are described and judgements are made about them, often on the basis of mere reflection. When these judgements appear to conflict with philosophical theories, it is common to respond by rejecting or amending the theories rather than by revising the judgements. Indeed, this is one of the main ways in which philosophers evaluate their views. But how are we to make sense of this process? In particular, what is the nature of these judgements such that they are able to play this role?1 Anna-Sara Malmgren (2011) responds to this question by arguing that intuitive judgements about thought experiments are metaphysical possibility judgements whose contents straightforwardly contradict the modal implications of the theses they lead us to reject.2 In doing so, she aims to defend the view that such judgements are justified a priori, if at all, from some arguments of Timothy Williamson’s (2005, 2007a, 2007b). In what follows, I show that Malmgren’s proposal fails, and develop a variant of Williamson’s own proposal, according to which judgements about thought experiments are counterfactual in nature. First, I outline Malmgren’s route to her view, identify two problems that arise concerning misjudgements, and rule out some possible responses (§§1–2). I then make the case that these problems result from a failure to recognize that when engaging with thought experiments, we are primarily concerned with what the relevant hypothetical scenarios suffice for (§3). This brings into sharper focus what is right in Williamson’s proposal, and I show how Malmgren’s criticisms of his view can be avoided by appealing to the notion of normal counterfactual sufficiency (§4).3 I then consider two objections to this view, in response to which it is fleshed out in greater detail and provided with some extra support (§5). I end with some brief remarks concerning the bearing that this has on the apriority of such judgements (§6). 1. Proposals Following Malmgren, let us focus on the following variation on one of the original Gettier (1963) cases: …Smith believes that Jones owns a Ford, on the basis of seeing Jones drive a Ford to work and remembering that Jones always drove a Ford to work in the past. From this, Smith infers that someone in his office owns a Ford. … [F]urthermore … someone in Smith’s office does own a Ford—but it is not Jones, it is Brown. (Jones’s Ford was stolen and Jones now drives a rented Ford.) (Malmgren, 2011, p. 272) Call this ‘the Gettier Case’.4 There is a particular judgement about this case that many make if prompted. A common way of expressing the judgement in question is via the following sentence: (S) Smith has a justified true belief, but does not know, that someone in his office owns a Ford (Malmgren, 2011, p. 272). Call the judgement thereby expressed ‘the Gettier Judgement’.5 As Malmgren points out, it would be a mistake to think of this as a judgement to the effect that some particular individual, picked out by ‘Smith’, has the property of justifiably and truly believing, but not knowing, that someone in his office owns a Ford. There is no such individual, and expressing the judgement in this way need involve no confusion on that point. This presents us with an instance of what she calls ‘the content problem’ (2011, p. 267). Roughly speaking, the problem is that we typically voice judgements about hypothetical situations in just the same ways as we would if those situations were real, and yet the relevant judgements do not plausibly have contents of the same kind. So what is the content of the Gettier Judgement? Any acceptable answer to this question must allow us to make sense of the role that this judgement plays in our philosophical theorizing, namely, leading us to reject the thesis that knowledge is justified true belief. Presumably, it does so in virtue of having a modal significance that conflicts with the claim that, necessarily, someone knows a proposition p if and only if he or she has a justified true belief that p. One straightforward way to satisfy this requirement, then, is simply to identify the Gettier Judgement’s content with a suitable modal proposition, and indeed this is precisely what Malmgren does.6 The three main proposals she considers are: () Necessarily, anyone who stands to a proposition p as in the Gettier Case has a justified true belief that p but does not know that p. () If someone were to stand to a proposition p as in the Gettier Case, then he would have a justified true belief that p but not know that p. () It is possible that someone stands to a proposition p as in the Gettier Case and that he has a justified true belief that p but does not know that p.7 These can be thought of, respectively, as ways of making a little more precise the rough but straightforward suggestions that the Gettier Judgement concerns what must hold, what would hold, and what could hold, in an instance of the Gettier Case. Now, the problem for (), pointed out by Williamson (2005, pp. 6–7; 2007b, p. 185) and endorsed by Malmgren (2011, pp. 275–7), is that the Gettier Case description is rather limited in detail. As a result, it can be consistently elaborated so as to reverse the impression that Smith’s belief is justified. For example, it is consistent with the case description above that Smith has been told that Jones is renting, but that it momentarily slips his mind when he infers what he does. Someone could therefore stand to a proposition as in the Gettier Case and yet not have a justified true belief without knowledge. This entails that () is false. But the Gettier Judgement is correct. So () cannot be its content.8 () is Williamson’s alternative. Against it, Malmgren (2011, pp. 278–81) builds on the previous argument as follows. If someone were to be related to a proposition as in the Gettier Case while lacking justified true belief without knowledge, then () would be false. But the Gettier Judgement would still be correct. That is to say, our semantic intuitions seem to tell us that such ‘deviant’ instances of the case have no bearing on the correctness of the judgement, no matter what their proximity in modal space. So () cannot be the Gettier Judgement’s content either.9 Malmgren’s solution is to retreat to (). This avoids the problems above, as neither the possibility nor the actuality of apparently ‘deviant’ instances of the Gettier Case can suffice to falsify (). She goes on to address a number of objections, to which her responses are largely convincing (pp. 281–306). Taking the Gettier Judgement to be paradigmatic, then, she concludes that all of our typical judgements about thought experiments are metaphysical possibility judgements of roughly the same form.10 There are, however, serious problems for this view which Malmgren fails to consider, and to which I now turn. 2. Problems Let us examine, alongside the Gettier Judgement, another judgement that might be made about the case. A typical way of expressing the judgement in question would be via the following sentence: (S*) Smith neither has a justified true belief nor knows that someone in his office owns a Ford. Call this ‘the Gettier Misjudgement’. As the name suggests, the judgement in question is incorrect. But it is nevertheless a judgement that some make when presented with the case. So what is its content? The most obvious way of applying Malmgren’s proposal to the Gettier Misjudgement would be to assign to it the following: (*) It is possible that someone stands to a proposition p as in the Gettier Case and that he neither has a justified true belief that p nor knows that p. Problems with assigning () and (*) to the judgements expressed via (S) and (S*), however, arise immediately. I will identify two. The first problem is this. While the Gettier Judgement is correct, the Gettier Misjudgement is, as already noted, incorrect. The vast majority of epistemologists writing since Gettier have agreed on this. We are liable to think that those who utter (S*) have made some kind of mistake. But this difference in correctness is not reflected in the truth values of the candidate contents. Both () and (*) are true. (Recall that the truth of (*) was the ground for rejecting ().) On the face of it, then, Malmgren’s view would count both judgements as correct. And this is clearly the wrong result. There are, in principle, two ways in which Malmgren could respond. Either she could accept the content assignment suggested above, and attempt to explain away the apparent incorrectness of the Gettier Misjudgement, or she could deny that (*) is the content that she would assign to it. However, both options are unacceptable, for the following reasons. Regarding the former, note that the semantic appearances persist even when we have explicit knowledge of the truth of (*). Why, then, does the Gettier Misjudgement continue to strike us as incorrect? No plausible explanation seems to be forthcoming. But more significantly, any such explanation would only serve to undermine the case presented against () and (). That case relied on our semantic intuitions to tell us that the truth of (*) does not render the Gettier Judgement incorrect. But our semantic intuitions also tell us that the truth of (*) does not render the Gettier Misjudgement correct. And there is no good reason to trust our intuitions in the one case while doubting them in the other. The latter option holds that (*) is not the content of the Gettier Misjudgement, even though () is the content of the Gettier Judgement. But then what is its content to be? To rescue the view, it would have to be a false metaphysical possibility claim of some sort. But it is doubtful that any other metaphysical possibility claim in the vicinity would be as plausible a candidate as (*). And even if some such candidate could be found, it is likely, given the point above concerning the many ways in which the Gettier Case can be consistently elaborated, that it would turn out to be true as well. Whatever the prospects for finding such an alternative, however, it would do nothing to address the second problem, which builds on the first as follows. It is not merely the case that the Gettier Judgement and Misjudgement differ in correctness. They are also clearly in tension. Two subjects who assert them in turn are manifesting disagreement about the Gettier Case, where such disagreement strikes us a matter of their giving voice to inconsistent judgements. Yet if the judgements expressed have metaphysical possibility claims as contents, then the subjects are not disagreeing in the manner envisaged. This is so for the simple reason that metaphysical possibility claims—claims of the form p—cannot be inconsistent.11 So Malmgren’s view entails that the judgements in question are not in tension after all. Again, though, this is clearly the wrong result. It is worth pointing out that even if each of the proposals mentioned in §1 generates a conflict with some of our semantic intuitions concerning correctness, this second problem is unique to (). If () is the content of the Gettier Judgement, then the content of the Gettier Misjudgement should be the claim that, necessarily, anyone who stands to a proposition p as in the Gettier Case has neither a justified true belief that p nor knows that p. If () is the content of the Gettier Judgement, then the content of the Gettier Misjudgement should be the claim that if someone were to stand to a proposition p as in the Gettier Case, then he would neither have a justified true belief that p nor know that p. Of course, strictly speaking, each of these pairs is consistent. But the Gettier Judgement and the Gettier Misjudgement both take place against the background assumption that an instance of the Gettier Case is possible (see note 2 above). So on either view, the vacuous truth of the relevant pair of contents is precluded, and the mutual inconsistency of the judgements is thereby ensured. Now, Malmgren may object to all of this that her proposal was never intended to have any implications for the content of the Gettier Misjudgement. After all, her stated aim, in the service of which she presents the proposal, is to provide an account of a ‘core set’ of ‘successful thought experiments—experiments that, other things equal, … provide the thought experimenter with … a justified and reliably true belief that the theory under evaluation is false’ (2011, p. 272). And while the Gettier Judgement provides us with a justified and reliably true belief that the justified-true-belief theory of knowledge is false, the Gettier Misjudgement does nothing of the sort. So Malmgren may wish to claim that although (S) is used to express a judgement of metaphysical possibility, the judgement expressed via (S*) could have a content of an entirely different form. However, the fact that her proposal was intended to cover only a restricted range of judgements does not mean that it has no implications for judgements outside that range. Notwithstanding differences in correctness or usefulness, the Gettier Judgement and Misjudgement have a manifest similarity that makes it inappropriate to treat them as having contents of radically dissimilar form. Were one to do so, the explanations of how relevant utterances of (S) and of (S*) manage to express the judgements they do would have to proceed quite differently in each case. But given how syntactically similar the sentences are, these divergent explanations would be implausibly unsystematic. In other words, if there is no way to treat both (S) and (S*) as expressing metaphysical possibility judgements, neither should be so treated. Malmgren’s proposal must therefore be rejected.12 These issues are not particular to the judgements discussed above. Nor are they particular to the way in which Malmgren fleshes out the basic idea that judgements about thought experiments are metaphysical possibility judgements. On any such view, parallel problems will arise for many pairs of seemingly conflicting judgements. We must therefore look elsewhere for a correct account of their content. 3. Sufficiency I suggested above that there was a single issue underlying the problems just outlined. It is this. When constructing descriptions of hypothetical situations, we do so with an eye to identifying cases which strike us as sufficing, non-stipulatively, for the presence of some philosophically significant feature. And when making judgements about such cases, we are typically in the business of assessing whether or not they in fact do so. Our practice of engaging in thought experiments, then, is best construed as a search for sufficiency.13 And Malmgren’s proposal faces the difficulties it does precisely because it fails to reflect this fact. This can be brought out most clearly by considering a relatively minimal case description, such as the following impoverished version of the Gettier Case: Smith* believes that Jones owns a Ford. From this, Smith* infers that someone in his office owns a Ford. Now, there are a vast number of properties which, it strikes us as obvious, it is possible for someone in Smith*’s position to instantiate. But for any such property, picked out by , we are not thereby inclined to make or agree with the judgement that would be expressed via Smith* is . This is so even when an opposing judgement also fails to strike us as correct. To see this, just consider the judgements that would be expressed via the following sentences: Smith* has a justified [unjustified] belief that someone in his office owns a Ford.14 Smith* has a true [false] belief that someone in his office owns a Ford. Smith* knows [does not know] that someone in his office owns a Ford. Smith* works [does not work] in a large office. Smith* is [is not] over six feet tall. … If the judgements in question concerned possible co-instantiation, then our inability to see any truth in them would be inexplicable. But the reason why none strikes us as correct is simply this: the hypothetical situation has not been filled out in enough detail to warrant any of them. We look to the explicit details of a case description to provide us with a reason for taking other, further features to be in some sense (and however unintentionally) present. And the impoverished Gettier case fails to do this with respect to any of the properties predicated of Smith* above. But this is just to say that our interest is in working out what the case described suffices for. With this in mind, we can see why Malmgren’s view generated the problems it did. A genuinely possible hypothetical scenario may be consistent with the presence of either one of two incompatible features, but it cannot suffice for the presence of two incompatible features. If it suffices for the presence of one such feature, it must fail to suffice for the presence of the other. It is precisely because the Gettier Judgement and the Gettier Misjudgement both concern what the (non-impoverished) Gettier Case suffices for, then, that they are inconsistent. And it is because the case fails to suffice for Smith to lack justified true belief that the Gettier Misjudgement is incorrect. By taking these judgements to concern mere compossibility instead, Malmgren lacked the resources necessary to account for these facts. In what sense, though, can the Gettier Case be said to suffice for the presence of justified true belief without knowledge, given that it permits metaphysically possible instances in which this is absent? Well, it was Williamson’s insight that metaphysical sufficiency is not the only option here, and that the Gettier Judgement might instead concern what the Gettier Case counterfactually suffices for. But what, then, are we to make of the criticism Malmgren presses against this view? The criticism, recall, was that the correctness of the Gettier Judgement should not depend on whether there happen to be actual (or modally proximal) instances of the Gettier Case in which justified true belief without knowledge is absent. The same objection is pressed, in a slightly different way, by Ichikawa (2009) and Ichikawa and Jarvis (2009). Williamson has responded to this worry both by trying to explain away the inclination to think that the Gettier Judgement does not so depend (Williamson, 2007b, pp. 200–4) and by arguing that the critics’ proposed alternatives fare no better, as they leave the Gettier Judgement exposed to similar epistemic risks (Williamson, 2009, pp. 466–8).15 Details aside, I agree with the critics that these responses are less than satisfactory. But there is another, better response that can be given on behalf of a counterfactual proposal. 4. Normal counterfactual sufficiency We have a shared understanding of what it is for an instance of a given case to be abnormal, in this or that respect, as an instance of that case. This generalizes into a grasp of what it is for an instance of a given case to be an abnormal one simpliciter. The existence of this grasp is testified to by Williamson’s use of the the word ‘abnormal’ (2007b, pp. 200, 205) and Malmgren’s use of the word ‘deviant’ (2011, pp. 276–7 and passim) to generalize intelligibly about a range of instances of cases identified primarily as such—that is, as abnormal or deviant. But obviously we do not merely know what it means for an instance to be an abnormal one; we have a sense of whether or not particular instances of cases are abnormal, qua instances of those cases. Call this our sense of normalcy.16 This sense puts us in a position to make judgements about what would normally hold in a given case, and a moment’s reflection should reveal that this is something that we in fact do with some regularity.17 The suggestion, then, is that our judgements about thought experiments are typically judgements of just this sort—in other words, that they are typically judgements about what hypothetical scenarios normally counterfactually suffice for. Before I make this suggestion a little more precise, though, there are three closely related points that need to be made. The first is that what would be normal for a given scenario is not simply a matter of what happens to hold in the actual instances of that scenario (cf. Smith, 2010, pp. 15–16). This can be brought out in two ways. First, by pointing to the fact that our sense of what is or would be normal for a given scenario need not be undermined by the discovery that it has no actual instances. Second, by pointing to the conceivability of there being more instances of a given situation that are abnormal in some respect than ones that are normal. (Or, to make the same basic point in another way, we can imagine it being the case that genuine coincidences have happened more often than not in the actual instances of some scenario.) This idea needn’t require any notion of what is natural for a situation, in the sense of situations themselves having a nature. For it may suffice to underwrite a claim of prevalent abnormalcy that, in local modal space, instances with the (supposedly) abnormal feature are sufficiently outnumbered by instances with a (supposedly) normal feature that precludes it.18 The second is that we can be wrong about what would be normal. We can fail to realize something is normal (or abnormal), and we can think that something is normal (or abnormal) when it is not. That is to say, what would be normal for a given scenario is not simply a function of what we take to be normal for that scenario. Rather, what would be normal for a scenario is a matter of how things generally go in worlds that are like ours except in so far as they contain, or easily could have contained, instances of the relevant scenario. Now, we typically think of how things generally go in terms of the laws, norms and tendencies that we take to be in force. Keeping with this way of thinking, then, we can say that what is in fact normal for a scenario will be any feature whose absence from an instance of that scenario (in one of the worlds just mentioned) involves an exception to some relevant law, norm or tendency (that is in force across those worlds). The third is that, when engaging in thought experiments, our sense of whether or not an instance of a case is normal is not driven solely or even primarily by our sense of whether or not the properties that are of philosophical interest to us are present or absent in that instance. To take the Gettier Case as an example, our sense that certain instances of that case count as abnormal is not driven by our sense that those instances are ones in which justified true belief without knowledge is absent. If we consider an instance of the case in which the person in Smith’s position is prone to hallucinations, for example, this strikes us as abnormal precisely because he or she is prone to hallucinations, and not primarily because he or she thereby lacks justified true belief without knowledge. The relevance of these points will become clear as we proceed. For now, let us return to the suggestion mooted above. The thought was that we can characterize the content of judgements about thought experiments in terms of normal counterfactual sufficiency. In order to make the idea more precise, let us consider how to apply it to the Gettier Judgement. There are in fact two ways in which we might seek to implement this idea. One is simply to provide a new counterfactual rendering of the content of the Gettier Judgement that makes explicit the restriction to normal instances. Another is to argue that () is acceptable as it stands, on the basis that a correct semantics for () would itself involve a restriction to normal instances. As nothing of immediate epistemological import rests on the choice, I will briefly comment on the second strategy before turning to the first. According to the standard semantics for counterfactuals, () is true just in case all of the closest possible worlds in which someone is related to a proposition p as in the Gettier Case are worlds in which the relevant individual has a justified true belief that p but does not know that p.19 If the actual (or a suitably nearby) world contains an instance of the case in which justified true belief without knowledge is absent, then according to this semantics, the counterfactual is false. But this should give us pause. () is what we might call a perfectly general counterfactual. Its antecedent picks out a situation which can have multiple instances in a world, each involving distinct individuals. Indeed, it makes no reference to any particular objects, events or times. In and of itself, therefore, () does not direct us to any manageable range of particulars such that its truth turns solely on whether the modally local instances of the antecedent situation involving those particulars also satisfy the consequent. And so accordingly, in assessing (), we make no attempt at investigating the contingent circumstances of any given particulars in order to work out whether they happen to meet this condition. Now, it might well be that there is, nevertheless, some limited range of possible instances of the Gettier Case such that the worlds containing them count as the closest ones containing instances of the case. And if so, it is entirely possible that justified true belief without knowledge is absent in one of these instances. But again, in assessing (), we simply do not concern ourselves with whether or not this is the case. That is to say, we do not concern ourselves with the vagaries of happenstance, or what could have easily resulted from the vagaries of happenstance. We rely instead on our general sense of how things go—on our sense of normalcy—trusting it to put us in a position to assess (). This gives us a reason, albeit an inconclusive reason, to think that () is something the truth of which does not depend on the vagaries of happenstance, and which instead depends on what would be normal for the case specified in its antecedent. However, it is beyond the scope of this paper to provide an alternative semantics for such counterfactuals.20 Moreover, a reasonable starting point for developing such a semantics might be to seek a claim in English that is parallel to () but which makes explicit the restriction to normal instances. For these reasons, I shall now turn to the first way of implementing the idea mentioned above. What, then, is a plausible content proposal for the Gettier Judgement that would count it as concerning normal counterfactual sufficiency? The suggestion is in fact very straightforward, given what has already been said. It is this: () If someone were to stand to a proposition p as in the Gettier Case, then, normally, he would have a justified true belief that p but not know that p. There are undoubtedly significant questions concerning the precise semantics of () and how it should be formalized. But I will abstain from addressing such questions here. The sentence has a manifest sense. This puts us in a position, and makes it reasonable for us, to consider its adequacy as a characterization of the content (or modal significance) of the Gettier Judgement. For present purposes, then, our existing understanding will suffice. First, it is clear that () is both true and inconsistent with the claim that necessarily, someone has a justified true belief that p if and only if he or she knows that p (granted, of course, the background assumption that an instance of the Gettier Case is possible). It can therefore account for the role that the Gettier Judgement plays in our theorizing just as well as (), () and () can.21 Second, the present proposal avoids the problems that Malmgren’s view faced. It is false that if someone were related to a proposition p as in the Gettier Case then, normally, he would neither have a justified true belief that p nor know that p. So if this is the content of the Gettier Misjudgement, then that judgement is incorrect. Moreover, whatever turns out to be normal for the Gettier Case, it is clear that the claim just mentioned is inconsistent with (). So the Gettier Judgement and Misjudgement will be in tension in the manner required. Third, and most significantly, the present proposal is not subject to the criticism Malmgren, Ichikawa and Jarvis press against Williamson’s view. According to this criticism, if () is the content of the Gettier Judgement, then we can be right about how things would normally be in an instance of the Gettier Case, make the judgement on that basis, and nevertheless have our judgement turn out incorrect due to the existence of an actual abnormal instance. If () is the content of the Gettier Judgement, however, then this is not possible. As was noted above, the distribution of features over actual instances of a case does not settle what would be normal for that case. What normalcy turns on is the distribution of features over instances in local modal space. So a single instance of the Gettier Case in which justified true belief without knowledge is absent, whether actual or modally proximal, is not enough to falsify (). As long as instances of the Gettier Case containing justified true belief without knowledge are the norm, then the Gettier Judgement is correct. It is true, of course, that this leaves the Gettier Judgement exposed to some of the epistemic risks associated with contingency. After all, we can conceive of worlds in which it fails to be normal for instances of the Gettier Case to include justified true belief without knowledge. Consider, for example, a world in which the norms for car ownership are quite different from our own. At this world, the vast majority of car drivers rent their cars, and this is public knowledge. In such a world, seeing someone in your office drive a Ford to work and remembering that they always drove a Ford to work provides no real justification for believing that they own a Ford. So at this world, () is false. There are, however, two key points to make about the contingency of () and its associated epistemic risks. The first is that whereas it seems to us perfectly plausible that there might actually be, in our world, an instance of the Gettier Case in which justified true belief without knowledge is absent, it surely does not seem plausible to us that we inhabit a world in which (to keep with the example) the norms for car ownership are in fact vastly different from what we take them to be.22 So () is not contingent in a way that should cause us to doubt that we know it.23 The second is that if we imagine a subject in the renting world being presented with the Gettier Case, it is far from obvious that he or she would be correct to make the Gettier Judgement. For surely it would be natural for such a subject to make (what we call) the Gettier Misjudgement. And it seems, in these alternative circumstances, that this would be exactly the right judgement to make. In considering this case, therefore, we find extra support for the claim that judgements about thought experiments concern normal counterfactual sufficiency: as the facts about what would be normal for a given case vary, so too the correctness of judgements about thought experiments based on that case seem to vary.24 The foregoing makes the case for taking () to be the content of the Gettier Judgement, and so for taking typical judgements about thought experiments to be judgements of normal counterfactual sufficiency. But a question remains. Why did neither Williamson nor his critics seriously consider this option? On the part of the critics, the answer to this question is clear. Malmgren, Ichikawa and Jarvis were all concerned to defend the apriority of the Gettier Judgement. Accordingly, their fundamental problem with () as a content proposal was that, in opening up the judgement to epistemic risks associated with contingency, it threatened the judgement’s supposed apriority. On that score, () is clearly no better, and so the critics had no reason to think of it. But their argument against Williamson’s proposal was not that it left the Gettier Judgement exposed to epistemic risks associated with contingency per se; it was that it left the Gettier Judgement exposed to implausible epistemic risks. And on that score, () cannot be faulted. On Williamson’s part, there are two reasonable suggestions. The first possibility is that he saw the problems his critics presented as borne of a desire to preserve the apriority of the Gettier Judgement—a goal to which he is not sympathetic—and so failed to take them seriously. The second is that, because his aim was to formalize the Gettier Judgement so as to make explicit (and justify) the precise role it plays in rejecting the justified-true-belief theory of knowledge, the extra difficulties one would encounter in trying to formalize () made it less appealing. However, neither of these points justifies misrepresenting the content of the judgement, and it remains implausible that the judgement’s correctness depends on whether or not there just so happens to be a nearby, abnormal instance of the case in which justified true belief without knowledge is absent.25 5. Two objections I have argued that an adequate account of the Gettier Judgement must recognize that it concerns what the Gettier Case in some sense suffices for, and that taking () to be its content meets this condition while avoiding problems faced by other accounts. Before concluding that we should accept this proposal, however, we must consider two potential objections. The first, addressed in §5.1, concerns whether () might also be amended in such a way as to avoid the problem identified in §1. The second, addressed in §5.2, concerns how well the account fares when generalized to judgements about other hypothetical scenarios, and in particular to judgements about cases that are modally remote or nomologically impossible. 5.1 Amending Call any view according to which typical judgements about thought experiments have strict universal implications as contents—that is, any view according to which such judgements concern what must hold in instances of the relevant cases—a necessity view. We can think of necessity views in general as committed to the claim that the Gettier Judgement has the content: (*) Necessarily, anyone who stands to a proposition p as in Δ has a justified true belief that p but does not know that p, where ‘Δ’ serves to pick out a particular hypothetical case: the case about which the Gettier Judgement is being made. Taking Δ to be the Gettier Case, as seems most natural, (*) becomes (). But this we have already rejected. The problem, recall, was that the Gettier Case has possible instances which lack justified true belief without knowledge. This entails that () is false, and so not plausibly the content of the Gettier Judgement, given its correctness. An obvious response for one who wants to defend a necessity view, then, is to suggest that the Gettier Judgement does not really concern what must hold in instances of the case explicitly described. Rather, the response goes, it concerns what must hold in a more determinate version of that case, one richer in detail, and richer in such a way as to necessitate the presence of justified true belief without knowledge in its instances. The central question for any such view is then: what case does the judgement concern? What is Δ? If a plausible answer can be found that renders (*) non-vacuously true, then we may have no reason to prefer a counterfactual view in the first place. Anticipating a manoeuvre of this sort, Williamson claims that any attempt to emend the description of the Gettier Case so as to rule out every potentially confounding feature is bound to fail, as ‘[a]ny humanly compiled list of such interfering factors is likely to be incomplete’ (2007b, p. 185). Recently, however, this has been disputed by Grundmann and Horvath (2014a), who provide the following description of a more determinate version of the Gettier Case: Smith justifiedly believes that Jones owns a Ford, on the basis of seeing Jones drive a Ford to work and remembering that Jones always drove a Ford in the past. From this belief alone (plus the justified background belief that Jones is in his office), Smith logically infers, at time t, to the justified belief that someone in his office owns a Ford, which provides his only justification for that belief at t. In fact, someone in Smith’s office does own a Ford, so that Smith’s latter belief is true—but it is not Jones, it is Brown, and so Smith’s initial belief was false. (Jones sold his car and now drives a rented Ford.) Also, if Smith knows at t that someone in his office owns a Ford, then he knows this at t only in virtue of the facts described. (Grundmann and Horvath, 2014a, p. 530)26 Does this case description provide for a plausible defence of a necessity view? For it to do so, at least three things need to be the case. First, (*) must be non-vacuously true when Δ is taken to be the case just described. Second, it must be plausible that Δ, the case about which the Gettier Judgement is made, really is the case just described. And third, the proposal must be plausibly generalizable to other judgements about thought experiments. Let us grant, for the sake of argument, that no possible instance of this case could lack justified true belief without knowledge, and that this is so for appropriate, non-vacuous reasons. Then, if we take Δ to be the case described above, (*) will be true, and the first condition will have been satisfied. Even then, however, the other conditions are not. It is not plausible that Δ is the case above; and even if we take it to be, the resulting proposal does not plausibly generalize to cover other judgements about thought experiments. I shall bring out each point in turn. There are multiple reasons to be sceptical of the claim that the Gettier Judgement concerns the case above. To begin, note that we need some explanation of how those who make the judgement arrive at this more determinate case.27 The authors suggest that, when presented with the (original) Gettier Case description, we ‘implicitly interpret’ it in this more determinate way. It is unclear what exactly this implicit interpretation is supposed to involve. But it is, at any rate, such that the enriched description above ‘should be readily available and transparent to every professional epistemologist’ (Grundmann and Horvath, 2014a, pp. 531–2).28 This is, however, simply not the case. It takes time and intellectual effort to work out how to emend the initial description in this way. And while somewhat ad hominem, it is surely philosophically relevant to observe here that, given the significant erratum noted in Grundmann and Horvath (2014b), this way of enriching the description was apparently neither readily available nor transparent even to the authors themselves. Next, note that the truth of (*), when Δ is taken to be the case above, is supposed to be ensured by the authors’ stipulations concerning the epistemic structure of the case. This gives rise to two issues. The first is that the Gettier Judgement is naturally taken to be epistemologically substantive in two ways, both in its commitment to the absence of knowledge and in its commitment to the presence of justification. But this proposal renders the verdict that Smith’s belief is justified entirely trivial.29 In so far as we do not take our judgements concerning justification when presented with the original Gettier Case description to be trivial, then, we have reason to think that we are not interpreting it in the way the authors claim. The second concerns views of justification according to which the case above would be non-instantiable. The point here is not that such views might be correct, and that (*) might be vacuously true—we have granted for the sake of argument that it is not. Rather, the point is that, on the present account, we fail to make sense of those who hold these non-standard views of justification. Now, the authors themselves make mention of those who deny the possibility of justified false belief (Grundmann and Horvath 2014a, p. 529 n. 9; see Sutton 2007 and Littlejohn 2012). But they do not recognize that such philosophers would surely dispute the Gettier Judgement while urging the Gettier Misjudgement on us. The present proposal implies, however, that either they are misinterpreting the case due to a lack of expertise or they are, according to their own theories, making a simple mistake in failing to regard both judgements as merely vacuously true. The problem only becomes more serious once it is noticed that certain other views about justification also render the case impossible. For example, a coherentist about justification (such as BonJour 1985) might well deny that a person’s ‘only justification’ for believing something could be a single other belief of theirs. Yet such a philosopher might nevertheless demur from the Gettier Misjudgement, make the Gettier Judgement, and on its basis rule out the thesis that knowledge is justified true belief. And once again, we would be forced to attribute to them either misinterpretation or error, when neither seems appropriate. Even if Δ were the case above, though, our other problem would remain. This is that there is simply no good way to generalize the proposal under consideration to other judgements about thought experiments. Now, Grundmann and Horvath themselves voice some scepticism about whether their proposal generalizes (2014a, p. 531). But we concluded in §2 that one should not take judgements about thought experiments to have contents of inexplicably different strengths. And this conclusion, I want to suggest, should be taken to apply, not only to distinct judgements about a single thought experiment, but also to judgements about different thought experiments. That is to say, we should only assign (*) to the Gettier Judgement if a necessity view is defensible in general. So if the present proposal is to provide the blueprint for a defence of a necessity view, we must suppose that in every case in which we make correct, non-trivial judgements about thought experiments, the relevant case descriptions are being implicitly interpreted in ways that necessarily but non-trivially ensure the presence of the features that we attribute to them in our judgements.30 Yet it already seems a leap of faith to suppose that more determinate case descriptions of this kind will always be generable. To suppose that they are always in some sense present to our minds when we make the judgements we do is simply incredible. The only option that remains for one who wants to generalize this proposal is to conclude that, when engaging in thought experiments, we are guilty of making many rash and incorrect judgements of manifestly inappropriate strength. But this is a broadly sceptical result, one which we should not accept unless it is genuinely forced on us. And the availability of the proposal outlined in §4 shows that it is not. Note, then, that () does not suffer from any of these problems. It allows us to take as the case in question the one corresponding to a literal and unembellished reading of the case description provided back in §1. So the relevant case will be transparent and readily available, not only to all experts, but to anyone who understands the language. Judgements both about justification and about knowledge will be substantive. And philosophers with the various views of justification canvassed above can arrive at the same judgement about what would normally hold in instances of the case in question, or at different judgements about what would normally hold in instances of one and the same case, without being guilty of any obvious misinterpretations or mistakes. Now, the proposal just considered exemplifies only one strategy for defending a necessity view.31 Before moving on, I want to briefly consider another. The strategy I have in mind is one which invokes relevance.32 Showing what is wrong with this way of defending a necessity view will provide us with a diagnosis, made available by (), of why such views continue to strike some as appealing, despite their inadequacy. The motivating thought here is simply that when assessing whether or not a hypothetical case has certain features, we take the case description to include everything relevant to whether or not those features are present. But there are various ways to apply this rough thought in the present context. One way is to stay at the same level of generality, and claim that we take the Gettier Case description to include everything relevant to whether or not Smith has justified true belief without knowledge. Another is to make the claim a little weaker, but more specific to the case at hand, claiming instead that we take the Gettier Case description to include all of the evidence Smith has that is relevant to whether or not someone in his office owns a Ford. We can think of both of these claims as entailing that Δ can be described by the Gettier Case description concatenated with an extra sentence making explicit whichever relevance restriction we want to go with.33 Either way, the hope is that this will ensure that no possible instance of the resulting case can exhibit features rendering Smith’s belief either unjustified or knowledge. The problem with the former, more general way of implementing the idea is simply that any instance of the Gettier Case will have features that are not explicitly mentioned in its description, but which are relevant to whether or not the person in Smith’s position has justified true belief without knowledge. To see this, consider the following feature that an instance of the case might have: the person in Smith’s position having been told that the person in Jones’s position is renting. Now, the presence of this feature is clearly relevant to whether or not the person in Smith’s position has justified true belief without knowledge. But it is not explicitly mentioned in the original description of the Gettier Case. Therefore, an instance of the original case with this feature is not an instance of the case that includes the relevance restriction. So far, one may think, so good. But then, by the same token, the absence of this feature—that is, the person in Smith’s position not having been told that the person in Jones’s position is renting—must also be relevant to whether or not the person in Smith’s position has justified true belief without knowledge. Yet this too is not explicitly stated in the original description. Therefore, an instance of the original Gettier Case without this feature is also not an instance of the case that includes the relevance restriction. It follows that any instance of the enriched case must neither have nor lack this feature. In other words, it follows that there are no possible instances of the enriched case. And so a version of (*) that takes this enriched case to be Δ will only be vacuously true.34 The same basic objection can also be pressed one level up, as it were, in the following way. I claimed above that the appeal to relevance is driven by the thought that the case description can be read as including everything relevant to whether or not Smith has justified true belief without knowledge. But one thing that is relevant to whether or not Smith has justified true belief without knowledge is whether or not he has been told that Jones is renting. And the case description simply does not state whether or not Smith has been told that Jones is renting. It therefore straightforwardly fails to include everything relevant to whether or not Smith has justified true belief without knowledge, and cannot be read as doing so, contra the thought driving the appeal. The defects with the latter way of implementing the appeal to relevance are a little more subtle, but no less problematic. This relevance restriction, while weaker than the last, is still rather strong. It requires that Smith must have no evidence whatsoever that bears on whether or not someone in his office owns a Ford, other than seeing Jones drive a Ford to work and remembering that he always drove a Ford to work. But this implies that Smith has no evidence whatsoever concerning, among a great many other things, the norms of car ownership. For any such evidence would surely be relevant to whether or not someone in his office owns a Ford.35 And now the problem is obvious: without any such evidence, it is hard to see how Smith’s belief that Jones own a Ford could be justified. Unless a person in Smith’s position were in some sense aware that driving a car to work every day typically indicates ownership, what could license his or her inference? Recall that this restriction was supposed to provide us with a candidate for Δ that would render (*) true. It now looks as though this restriction will generate a case that not only fails to metaphysically necessitate the presence of justified true belief without knowledge in its instances, but which instead necessitates its absence. It therefore cannot provide for a defence of a necessity view. We are left with a puzzle. Why is it so tempting to appeal to relevance here, given the seriousness of the problems above? And what I want to suggest is that taking () to be the content of the Gettier Judgement actually allows us to provide an answer—and as far as I can see the only plausible answer—to this question. The thought is that those who would make these appeals to relevance must already have in mind a certain verdict about the case in question. That is to say, they must have already judged that Smith has a justified true belief without knowledge. For we can then understand their claim that the case description includes everything relevant as giving inchoate expression to the idea that it contains everything relevant to this judgement—that is, that it contains all that is required of a case in order for a judgement expressible in these words to be correct. And the temptation to invoke relevance when presented with certain potential further features can be understood as deriving from the fact that their presence would have been relevant, in precisely the following sense: that had one instead considered a modified version of the case with these features included, one would no longer have taken a judgement expressible in these words to be correct. But this implies that the Gettier Judgement cannot depend on ‘interpreting’ the case description in a way that is guided by any prior identification of various features as ones that would have been relevant to whether Smith has justified true belief without knowledge. For taking certain features to be ones that would have been relevant, in this sense, itself depends on the judgement. What all of this requires, of course, is that we have some account of the nature of the judgement in the first place such that it can play the required role in controlling our sense of whether or not further features would have been relevant. But this is just what () provides.36 5.2 Generalizing () The final worry I want to address concerns how well () generalizes to judgements about other cases. In particular, I will address the question of whether thought experiments involving modally remote or nomologically impossible cases pose a problem for the view that typical judgements about thought experiments are judgements about what would normally hold in instances of the relevant cases. To a first approximation, and one which will serve present purposes, the view arrived at by generalizing () can be represented as follows. If a thought experiment involves a hypothetical case, described by , and a judgement is made about it that would be expressed by (kn), then that judgement has the content expressed by If some things , …, xn were to be , then, normally, , …, xk would be .37 There are two potential problems that this view might be thought to face when picks out a modally distant or nomologically impossible case, one semantic, the other epistemic.38 The semantic worry is that the relevant normally-would counterfactuals will be at best universally vacuously true, if not universally false. The epistemic worry is that, even if the relevant normally-would counterfactuals can be non-vacuously true, we are simply in no position to make knowledgeable judgements about what would normally hold in instances of such cases. I will address these in turn. The semantic worry, I take it, is driven by the suspicion that there can’t be anything that would count, from the perspective of our world, as normal or abnormal for a case that is so thoroughly alien. But this is a mistake. Recall one of the glosses given on normalcy in §4: what is normal for a scenario is a matter of how things generally go in worlds that are like ours except in so far as they contain, or easily could have contained, instances of the relevant scenario. Now it is clear that a world which could (easily) contain an instance of a nomologically impossible scenario would have to be quite unlike our own. But this in no way precludes there being facts about how things generally go in such a world. It therefore does nothing to preclude there being facts about what such scenarios normally counterfactually suffice for. If there are laws, norms or tendencies in force at these worlds to which the absence of a certain feature from an instance of the scenario would constitute an exception, then the scenario normally counterfactually suffices for the presence of that feature. What remains unclear, however, is whether or not nomologically impossible cases can normally counterfactually suffice for anything which they do not also metaphysically necessitate. Given what has already been said, this will turn on whether or not, among the worlds that contain (or could easily have contained) instances of a given nomologically impossible case, some will count as being more like our world than others. Now, either this is so or it is not. If it is so, then there might well be laws, norms and tendencies in force in the worlds most like ours, and not in force in the others, to which the absence of a certain feature from an instance of the case would constitute an exception. In that case, the nomologically impossible scenario will normally counterfactually suffice for the presence of this feature despite not metaphysically necessitating it. If it is not so, then every world containing an instance of the case in question will count as being equally like our world. And in that case, there will be no laws, norms and tendencies in force across this entire range of worlds other than those which are metaphysically necessary, and the features whose absence would (per impossibile) be an exception to these laws, norms and tendencies are just those whose presence is metaphysically necessitated by the scenario. What it is important to see here, though, is that, either way, the relevant normally-would counterfactuals can be non-vacuously true. In the former case this is already clear: if a case normally counterfactually suffices for something without metaphysically necessitating it, then there is a non-vacuously true normally-would counterfactual. But in either case, as long as there are non-vacuously true strict implications, then there will be corresponding non-vacuously true normally-would counterfactuals. At the limit, then, normal counterfactual sufficiency simply collapses into metaphysical necessitation.39 But this cannot be grounds for objecting to the claim that typical judgements about thought experiments concern normal counterfactual sufficiency. At worst, it simply implies that typical judgements about thought experiments involving nomologically impossible cases might turn out to be correct only when corresponding strict implications are true. Finally, then, let us briefly consider the epistemic worry. We can immediately set aside this worry concerning claims of normal counterfactual sufficiency for which a parallel claim of metaphysical necessitation would be true. However it is that we are supposed to come to know such strict implications concerning modally remote or nomologically impossible cases, they can present no special problem for the present view, given that any strict implication will imply a parallel normally-would counterfactual. Any acceptable account of how we are able to knowledgeably make the former would ipso facto be an acceptable account of how we are able to knowledgeably make the latter.40 The worry, therefore, can only concern those cases in which picks out a nomologically impossible case, and picks out a feature that the case normally counterfactually suffices for but which it does not metaphysically necessitate. The worry, then, is: how can we make such judgements knowledgeably? To this I think there are three appropriate responses. The first is that, on the face of it, this is a demand to account for items of knowledge which other views of judgements about thought experiments would wish to deny us. So if there are such items of knowledge, then this is immediately a reason to deny these other accounts and accept the present one, albeit with a view to providing an answer to the question. The second is that it is a presupposition of the worry that some of the worlds containing nomologically impossible cases are more like ours than others. If so, then this will plausibly be, in part, a matter of some of the laws, norms and tendencies (however abstract and general) that are in force in the actual world still being in force in some, but not all, of these worlds. What will unite the worlds most like ours is simply their being alike in this respect. Therefore we can in principle use our sense of these abstract and general aspects of how things generally go to guide us in our judgements about the nomologically impossible scenarios. Of course, this would require that we also have some sense of what minimal differences would be required of the world in order to accommodate those scenarios. And this leads naturally to the third response, which is simply this: we really have no reason to suppose that we can make such judgements knowledgeably. After all, it might well be that, for reasons mooted above, there is nothing that would normally hold in a nomologically impossible scenario other than what that scenario metaphysically necessitates. And even if this is not so, the truths in question may simply be unknown to us. But if we cannot make such judgments knowledgably, then there is no range of judgments which this view would render problematic but which a version of the necessity view would account for. And in that case, we are provided with no reason to doubt the claim that typical judgments about thought experiments are judgments of normal counterfactual sufficiency. 6. Concluding remarks I have argued that there are no good objections to taking () to be the content of the Gettier Judgement coming either from a modification of the necessity view or from a consideration of thought experiments involving nomologically impossible cases. I want to conclude by raising two final questions. First, does taking () to be the content of the Gettier Judgement rule out the possibility of its being justified a priori? And second, to what extent can our answer be generalized to other typical judgements about thought experiments, granted that these are judgements of normal counterfactual sufficiency? Recently, Williamson (2013) has expanded his attack on the a priori/a posteriori distinction, arguing against its theoretical significance on the basis of there being considerable epistemological and psychological parallels between the geneses of certain judgements that are supposed to be paradigms of a priori and a posteriori justification. Many of the parallels he identifies seem to apply to our methods for forming judgements of normal counterfactual sufficiency. If his argument is correct, then the question of whether or not typical judgements about thought experiments can be correctly classified as a priori is simply no longer pressing. But suppose we find reason to reject his argument, and that an epistemologically significant and viable notion of the a priori can be identified. How then should we answer the questions above? On the face of it, the answer to the first may seem to be ‘yes’. The truth of () depends on what is normal for the Gettier Case. As we have seen, what is normal for this case turns on, amongst many other things, the norms for car ownership in the actual world. This is a deeply contingent matter, and so it is hard to see how the way in which our sense of normalcy underwrites the judgement could fail to undermine its classification as being justified a priori. And, again on the face of it, the answer to the second question may then seem to be that this verdict will extend to any other judgement the correctness of which depends on some contingent law, norm or tendency in the same kind of way. In the end, however, the answers we are after will turn on delicate questions concerning what exactly the a priori/a posteriori distinction is supposed to come to, questions which cannot be fruitfully pursued when abstracting away from the details of any particular account of that distinction. All that can be said at this juncture is that friends of the a priori who want to stake out a role for it in the epistemology of judgements about thought experiments will have to show one of two things: either that a priori justification is at least sometimes available for judgments of normal counterfactual sufficiency whose antecedents do not metaphysically necessitate their consequents; or that the epistemic ground of such judgments involves a significant a priori component, despite the judgments themselves failing to count as such. I do not take a stand on these deep and general issues here. Suffice it to say that for most judgements about thought experiments, we are not isolated from the epistemic risks that arise due to the contingency of normalcy. The full bearing of this fact on the apriority of these judgements is a question that must be left to another time.41 Footnotes 1 There is disagreement about the psychological and evidential bases of such judgements. In particular, some take there to be certain psychological states or episodes (‘intuitions’) the occurrence of which we typically rely on for evidence when making judgements about hypothetical cases (for example, Brown 2011, Goldman 2007, Nagel 2012, Pust 2000). Others are sceptical (for example, Cappelen 2012, Deutsch 2015, Williamson 2007b). The discussion below remains neutral on this issue. Note that the possibility of classifying certain judgements as intuitive (on some basis or other) remains even if one agrees with the sceptics. Accordingly, any reference to intuitive judgements below should be understood in a theoretically neutral way. 2 There are (at least) two kinds of thought experiment, only one of which is the focus of the present paper. One kind presents a hypothetical case and invites the thought experimenter to make a judgement about whether or not the case is possible. (Think zombies and inverted spectra.) Such judgements are manifestly metaphysical possibility judgements. Another kind, however, presents a hypothetical case, presupposes that the thought experimenter accepts its possibility, and invites him or her to make a judgement about what else holds in the case. (Think Gettier cases, trolley problems and brain transplants.) It is the latter with which we are presently concerned. 3 Gardiner (2015) has independently developed some related objections to Malmgren (2011), and likewise suggests invoking normalcy in an account of judgements about thought experiments. Gardiner’s paper came to my attention too late to include any detailed discussion of the differences between our approaches. Here I simply note that she neither offers precisely the same diagnoses of where other views go wrong as are offered below, nor develops her positive proposal specifically as a version of the counterfactual view. 4 I use ‘case’, ‘scenario’ and ‘situation’ interchangeably throughout to speak of ways for some things to be or to stand to each other: a kind of general object that I take to be specified by descriptions such as the one just given when they are put to use in thought experiments. And I use ‘the Gettier Case’ solely as a label for the case that involves all and only those features made explicit in the description above. (Note that, as is standard, ‘thought experiment’ is in fact used ambiguously: sometimes to refer to procedures that make use of cases, sometimes to refer to cases themselves. In what follows, I rely on context to determine its sense on each occasion of use.) 5 An anonymous referee suggested that one might reject the Gettier Judgement on the basis that, in this case, Smith’s stated evidence fails to justify his belief that Jones owns a Ford. But one may alter the case by enriching his evidence somewhat and, mutatis mutandis, all of the points below will stand. (That is, as long as one does not stipulate that his belief is justified; see §5.1 below.) 6 One might doubt whether modal propositions of the kinds identified below could plausibly be the contents of judgements about thought experiments. One might also doubt whether we use sentences such as (S), in the context of thought experiments, to give voice to genuine judgements at all (cf. Yablo 2001 and Eagle 2007). In either case, one can interpret most of what follows as addressing the following, less theoretically loaded question: in performing the relevant mental act given voice to via (S), what modal commitment is incurred? 7 See Malmgren (2011, pp. 274, 277 and 281). I have substituted ‘the Gettier Case’ for ‘the Gettier case (as described)’. I have also substituted ‘he’ for ‘she’, on the basis that the original case description and (S) both employed the phrase ‘his office’. 8 See §5.1 below for discussion of some potential responses to this argument. 9 Ichikawa and Jarvis (2009, pp. 225–6) present a similar objection. 10 Technically, she generalizes only to other intuitive judgements about thought experiments. But this restriction has no bearing on the issues discussed in this paper, for two reasons. First, given a plausible, non-factive characterization of what renders a judgement intuitive, all of the judgements considered below could be made intuitively. (Malmgren herself (2011, p. 268) identifies intuitive judgements as those ‘relevantly similar to’ certain paradigms, including the Gettier Judgement; in Malmgren, 2013, she provides a more explicit, albeit tentative, characterization. Neither gloss, I take it, implies that intuitive judgement is factive.) Second, the purpose of the restriction is to render the claim that such judgements are justified a-priori-if-at-all at least prima facie plausible. (The fact that judgements about thought experiments, like most judgements, can be justified on a variety of bases would render a parallel claim without the restriction to intuitive instances obviously false.) But while this restriction may have a place when it comes to a debate over justification, it has none whatsoever when it comes to a debate over content or modal significance. 11 Qualification: as long as p itself is neither a modal proposition nor a contradiction. 12 It might be suggested that the Gettier Misjudgement could have an impossibility claim as its content, and that this could restore incorrectness or inconsistency without departing too drastically in form from (). The most plausible option would presumably be something like: it is not possible that someone stands to a proposition p as in the Gettier Case and that it is not the case that he has neither a justified true belief that p nor knows that p. But the objection in the text still stands: whence the difference in form between the judgements expressed via (S) and (S*)? Where have all these negations come from? Moreover, this is logically equivalent to: necessarily, anyone who stands to a proposition p as in the Gettier Case has neither a justified true belief that p nor knows that p. What plausible explanation could there be of how or why these judgements are of such different strengths? 13 Note that the claim here is not that when engaging in thought experiments, we do so primarily in order to assess sufficiency claims, such as the claim that justified true belief suffices for knowledge (though this we do). The claim is that typical judgements about hypothetical cases are themselves judgements about what those cases suffice for. 14 This is a means of displaying two sentences: one in which the word ‘justified’ occurs and one in which the word ‘unjustified’ occurs. 15 It is puzzling that he has responded in this fashion, given his acknowledgment elsewhere that the Gettier Judgement might not be falsified by the existence of such instances (Williamson, 2007b, pp. 306–7). 16 It is easy to see how this sense could be grounded in, or part of, our capacity for counterfactual judgement, if that capacity fits the account of it sketched by Williamson (2007b, pp. 134–78). 17 It is clear that we make judgements not only about what a case would normally ensure, but also about what it would normally permit. Speaking simply of what is or would be normal for a case seems to be ambiguous between the two. (In the ensuring sense, it would be (say) normal for a T-shirt to have two armholes, and so abnormal for a T-shirt to have any number other than two. In the permitting sense, it would be normal for a T-shirt to be blue, but also normal for a T-shirt to be red, or to be white, or …) In what follows, when speaking of what is or would be normal, I have only the former sense in mind. 18 The application of the notion of outnumbering here may be suspect, as there may be an uncountably infinite number of instances (or worlds containing instances) that are close enough to be relevant to normalcy. I will simply assume that there is a way to make good the thought underlying the claim. (One suggestion, though: local modal space will presumably be the way required for there to be prevalent abnormalcy for some scenario when there are genuine, objective—though perhaps contingent and non-natural—tendencies which, for varying and disparate reasons, have failed to manifest themselves in a majority of the instances of that scenario in the actual world.) 19 See Lewis (1973) and Stalnaker (1968). In fact, it is unclear how to apply standard semantics to counterfactuals such as (), due (in part) to the presence of so-called ‘donkey anaphora’ (see Geach, 1962, King and Lewis, 2016). One aspect of the problem generated for the standard semantics by such counterfactuals is that their consequents are not propositions, and so it is unclear what it takes for an antecedent world to be a consequent world. Consider the gloss just given in the text. A nearby world in which someone is related to a proposition as in the Gettier Case may well be a world in which two people are so related to a proposition. But then who is ‘the relevant individual’ on whose epistemic state the truth of the counterfactual is supposed to depend? As it stands, no coherent condition is laid down for the world to meet. What is needed is a semantics that appeals to something more fine-grained, with a structure the account can then exploit. For discussion of how to accommodate the phenomenon, see vanRooij (2006), Williamson (2007b, pp. 305–8) and Wang (2009); cf. Fine (2012). 20 There are a number of accounts in the literature which seek to give a semantics for ‘would’ counterfactuals according to which they allow for modally local exceptions. To the best of my knowledge, such proposals typically take the antecedents and consequents of counterfactuals to be propositions, and so are of no immediate help in the present context, given the point made in the preceding footnote. Moreover, they are typically intended to apply to all ‘would’ counterfactuals, not simply what I have called general ones. But it is worth briefly abstracting away from these points to ask whether any might serve as a useful model. Schulz (2014) provides a semantics according to which the truth of a ‘would’-counterfactual depends on whether the consequent holds at a single, arbitrarily selected (via an ϵ operator), close antecedent world. However, if I understand the proposal correctly, this alone would leave the Gettier Judgement at risk of being incorrect for the wrong reasons, as bad luck could see Schulz’s ϵ operator arbitrarily selecting a world containing an abnormal instance of the Gettier Case, rendering () false. Better models for present purposes can perhaps be found in Leitgeb (2012a, 2012b), which provides a probabilistic semantics for counterfactuals, and Smith (2007), which provides a semantics for ceteris paribus conditionals that explicitly invokes a notion of normalcy. Detailed consideration of these proposals, though, is beyond the scope of this paper. 21 Regarding its truth, it is perhaps worth reiterating that normalcy is case-relative. That is, there is a distinction to be drawn between what would be normal for an instance of believing simpliciter and what would be normal for an instance of believing as in the Gettier Case. So however abnormal it might be for instances of believing (or instances of justifiably and truly believing, or instances of not knowing) to be instances of justified true belief without knowledge, this in no way precludes it from being normal for instances of the Gettier Case to involve justified true belief without knowledge. 22 It is perhaps worth noting here that claims of normal counterfactual sufficiency may exhibit a significant degree of context or occasion-sensitivity, not only in ways standardly associated with counterfactuals, but also, presumably, in ways associated with the adverb ‘normally’. 23 The present proposal therefore also avoids a criticism pressed against the counterfactual proposal by Ichikawa (2009, pp. 439–42), according to which () is not plausibly known, due to the epistemic possibility of there being a modally local violation. 24 One might object that what explains this difference in correctness is not the difference in norms for car ownership, but rather the difference in the evidence that the subjects in question have (or are naturally assumed to have) concerning those norms. As a reason to accept this alternative explanation, one might suggest that it is also plausible that were a subject in the renting world to have misleading evidence suggesting that the norms for car ownership there are more or less as they in fact are in our world, they would be correct to make the judgement that they would express via (S). Now, note that the correctness of a modal judgement about a single given hypothetical case is not sensitive to the evidence of the subject making it. (If some particular feature must—or would, or normally would, or could—be present in instances of a given scenario, this is so simpliciter, not relative to a body of evidence.) So if the foregoing accurately represents the pattern of correctness for the renting-world judgements in question, then they must strictly speaking concern different hypothetical cases. And the explanation for this difference, I take it, will be that the subjects in question are interpreting the case description in such a way that they take Smith to have the same evidence concerning the norms of car ownership that they do. The claim, then, is that in making a judgement in response to the Gettier Case description, one’s judgement really concerns whichever enriched case results from one’s evidence-dependent interpretation. There are four brief points I want to make in response to this. First, the objection implies that if someone in the actual world responds to the Gettier Case description by making a judgement that they would express by (S*), then, as long as they are appropriately deluded about the norms for car ownership due to misleading evidence, we are obliged to regard their judgement as correct. It also implies that, having made the Gettier Judgement, each of us is obliged to regard certain further judgements expressible by sentences of the form ‘Smith has evidence that p’ as trivially true, despite p concerning factors that are not explicitly mentioned in the case description. But it is not at all clear that these are obligations we are (or should be) willing to recognize. Second, and relatedly, the kind of sensitivity to evidence envisaged would suggest that individuals with different relevant evidence are likely to be making judgements about distinct cases despite being involved in a shared discourse. But there is no good reason to think that differences in evidence between subjects will frustrate public discussion of and disagreement about thought experiments in this way (cf. Williamson 2009, p. 468). These points suggest that the alternative explanation being proposed is questionable in its own right. But third, even if it is correct, it would not directly call into question the view I am proposing. It would, instead, simply undercut some extra support claimed for the view. And fourth, merely taking Smith to possess certain pieces of evidence that we ourselves possess would do nothing to avoid any of the problems facing the other views under consideration, as outlined in §§1–3 and expanded below in §5.1. To the extent that the arguments presented against those views are correct, then, they would carry over to any non-counterfactual view that adopts this alternative explanation. (I thank an anonymous referee for raising the point considered in this note.) 25 In fact, Williamson does, in a two-sentence footnote, briefly moot the possibility of taking the Gettier Judgement to concern normal instances of the Gettier Case, but dismisses it on the basis that ‘the relevant notion of normality is an epistemological one that violates the supposed neutrality of the initial description of the case’ (2007b, p. 204 n. 22). However, the only prior mention of neutrality in his discussion appears much earlier, in his characterization of a case description as being presented ‘in neutral terms, without prejudice to the target analysis’ (p. 184)—that is, in this context, without prejudice as to whether or not Smith has a justified true belief, but doesn’t know, that someone in his office owns a Ford. And on the present understanding of the Gettier Judgement, neither the case description nor its interpretation prejudges the target analysis in the relevant sense. That is, the case about which the judgement is being made perfectly well permits both instances in which the person in Smith’s position has justified true belief without knowledge and instances in which he does not. So there is no sense in which the neutrality of the initial description of the case is violated. 26 This incorporates the authors’ proposed fix for an erratum noted in Grundmann and Horvath (2014b, p. 536) and removes the underlining present in the original. 27 The demand here is for an explanation of how it is that one comes to have in mind the case on which one reflects, and about which one makes a judgement, given that its details go beyond what is provided by the initial description. This is distinct from a demand for an explanation of how it is that one arrives (perhaps justifiably or knowledgeably) at a judgement about it. Given any content proposal, there must ultimately be some account of the latter sort. But only given the present kind of proposal is an account of the former sort required. 28 One suspects that many professional epistemologists would be surprised to learn what this seems to imply, namely, that Williamson and Malmgren are not to be counted amongst their number. 29 The authors do in fact note this point (Grundmann and Horvath 2014a, p. 529 n. 8). As evidence that Gettier himself intended his original cases not to leave the presence of justification an ‘open question’, they cite the fact that he made substantive claims about the nature of justification that were intended to ‘incline us to attribute justified beliefs to Smith’. But surely this is just evidence that he took the judgement that Smith’s belief is justified to be correct? Besides, if it was intended to be stipulated, why did he not just stipulate it? 30 It would be a mistake to object to this on the basis that no general recipe has been provided for enriching descriptions so as to determine exactly which cases (expert) judgements concern. The generalized claim is simply that, when making judgements about thought experiments that are in good order, we skilfully interpret the cases in more determinate ways so as to make the relevant strict implications non-vacuously true. There is no reason to think that the plausibility of such a view depends on the possibility of spelling out in advance an exact procedure which this skilful interpretation employs. (That said, we might well wonder whether we should expect all of the expert judgements prompted by a given description to concern the same case; in this connection, see the second point raised in note 24 above.) 31 Ichikawa and Jarvis (2009) pursue another strategy that appeals to the idea that we read case descriptions in accordance with the norms for reading fiction. I will not discuss their view here; for criticism, see Malmgren (2011, pp. 303–6) and Williamson (2009, 465–9). 32 This possibility was raised by an anonymous referee. I have also encountered it numerous times in conversation. 33 In the former case, this might be the sentence ‘There is nothing else relevant to whether or not Smith has a justified true belief without knowledge’. In the latter case, it might be ‘Smith has no other relevant evidence concerning whether or not someone in his office owns a Ford’. 34 The point could have been made equally well with any epistemologically relevant feature not explicitly mentioned in the Gettier Case description, including the person in Smith’s position being a competent reasoner, having good vision, being able to easily recognize Fords, knowing the norms for car ownership, not being prone to hallucinations, having no other evidence that renders his or her belief knowledge, and so on. It would therefore be a mistake to attempt to resist this argument by pointing to the fact that the person in Smith’s position not having been told that Jones is renting is in some intuitive sense a ‘negative’ rather than a ‘positive’ feature. 35 In fact, once one begins to reflect on just how many things could have a bearing (however indirect) on whether or not someone in one’s office owns a Ford, it starts to seem plausible that even this weaker relevance restriction generates a case without possible instances. For it is doubtful that there could be a subject with only two pieces of evidence. And yet only two pieces of evidence are explicitly mentioned in the Gettier Case description. 36 It is worth noting that Malmgren considers a content proposal that has some affinities with an appeal to relevance (2011, pp. 285–90). The proposal she considers also takes the form of a strict implication, but it appeals to the ‘intended Gettier case’ in its antecedent in order to restrict the range of semantically relevant instances of the Gettier Case to ones in which justified true belief without knowledge is present. She rejects the proposal on the basis that it is unclear how the restriction can be effected other than by stipulating that ‘the intended Gettier case’ is to involve justified true belief without knowledge, thereby rendering the judgement trivial. One could, perhaps, think of the appeal to relevance as another attempt, alongside those she considers and rejects, to spell out the nature of the ‘intended Gettier case’ in a way that does not render the judgement trivial. (It is also worth noting that the supposed difficulty arises only because Malmgren wants the restriction to render the judgement plausibly justifiable a priori. Once it is accepted that the instances one intends to make a judgement about are the normal ones, however, the non-trivial viability of a proposal that restricts the range of semantically relevant instances is vindicated, albeit not in the form she envisaged.) 37 will pick out a way for n things to be, a way for k things to be; τ1 …, τn will be (typically, non-referring) singular terms. There are some superficial differences between this and the way in which () is worded, but none of substance. 38 In what follows, I will focus on nomologically impossible cases. The same points apply to modally remote cases. 39 In this respect, normal counterfactual sufficiency seems to be no different to counterfactual sufficiency simpliciter. If there is a ‘would’ counterfactual for which no antecedent world counts as being closer than any other, it will be true just in case a parallel strict implication is true. 40 Similarly, a general scepticism about the utility or reliability of judgements about nomologically impossible cases poses no special problem for the view. So let us also set aside such scepticism, however well-founded it might be. 41 Thanks to Katalin Farkas, Rory Madden, Mike Martin, Chris Meyns, Lucy O’Brien, Daniel Rothschild, Rachel Rudolph, Sebastian Sanhueza, Tom Williams, two anonymous referees, and audiences at Berkeley, UCL and CEU. References BonJour Laurence 1985: The Structure of Empirical Knowledge . Cambridge, MA: Harvard University Press. Brown Jessica 2011: ‘Thought Experiments, Intuitions and Philosophical Evidence’. Dialectica , 65( 4), pp. 493– 516. Google Scholar CrossRef Search ADS Cappelen Herman 2012: Philosophy Without Intuitions . Oxford: Oxford University Press. Google Scholar CrossRef Search ADS Deutsch Max 2015: The Myth of the Intuitive: Experimental Philosophy and Philosophical Method . Cambridge, MA: MIT Press. Eagle Antony 2007: ‘Telling Tales’. Proceedings of the Aristotelian Society , 107, pp. 125– 47. Google Scholar CrossRef Search ADS Fine Kit 2012: ‘Counterfactuals Without Possible Worlds’. Journal of Philosophy . 109( 3), pp. 221– 46. Google Scholar CrossRef Search ADS Gardiner Georgi 2015: ‘Normalcy and the Contents of Philosophical Judgements’. Inquiry , 58( 7–8), pp. 700– 40. Google Scholar CrossRef Search ADS Geach Peter Thomas 1962: Reference and Generality: An Examination of Some Medieval and Modern Theories . Ithaca, NY: Cornell University Press. Gettier Edmund L. 1963: ‘Is Justified True Belief Knowledge?’ Analysis , 23( 6), pp. 121– 3. Google Scholar CrossRef Search ADS Goldman Alvin I. 2007: ‘Philosophical Intuitions: Their Target, Their Source, and Their Epistemic Status’. Grazer Philosophische Studien , 74( 1), pp. 1– 26. Grundmann Thomas, Joachim Horvath 2014a: ‘Thought Experiments and the Problem of Deviant Realizations’. Philosophical Studies , 170( 3), pp. 525– 33. Google Scholar CrossRef Search ADS Grundmann Thomas, Joachim Horvath 2014b: Erratum to ‘Thought Experiments and the Problem of Deviant Realizations’. Philosophical Studies , 170( 3), pp. 535– 6. Google Scholar CrossRef Search ADS Ichikawa Jonathan 2009: ‘Knowing the Intuition and Knowing the Counterfactual’. Philosophical Studies , 145( 3), pp. 435– 43. Google Scholar CrossRef Search ADS Ichikawa Jonathan, Jarvis Benjamin 2009: ‘Thought-Experiment Intuitions and Truth in Fiction’. Philosophical Studies , 142( 2), pp. 221– 46. Google Scholar CrossRef Search ADS King Jeffrey C., Lewis Karen S. 2016: ‘Anaphora’. In Zalta Edward N. (ed.), The Stanford Encyclopedia of Philosophy (Summer 2016 Edition) . https://plato.stanford.edu/archives/sum2016/entries/anaphora/. Leitgeb Hannes 2012a: ‘A Probabilistic Semantics for Counterfactuals, Part A’. Review of Symbolic Logic , 5( 1), pp. 26– 84. Google Scholar CrossRef Search ADS Leitgeb Hannes 2012b: ‘A Probabilistic Semantics for Counterfactuals, Part B’. Review of Symbolic Logic . 5( 1), pp. 85– 121. Google Scholar CrossRef Search ADS Lewis David 1973: Counterfactuals . Oxford: Blackwell. Littlejohn Clayton 2012: Justification and the Truth-Connection . Cambridge: Cambridge University Press. Google Scholar CrossRef Search ADS Malmgren Anna-Sara 2011: ‘Rationalism and the Content of Intuitive Judgements’. Mind , 120, pp. 263– 327. Google Scholar CrossRef Search ADS Malmgren Anna-Sara 2013: ‘Review of Herman Cappelen, Philosophy Without Intuitions’. Notre Dame Philosophical Reviews. http://ndpr.nd.edu/news/39362-philosophy-without-intuitions/. Nagel Jennifer 2012: ‘Intuitions and Experiments: A Defense of the Case Method in Epistemology’. Philosophy and Phenomenological Research , 85( 3), pp. 495– 527. Google Scholar CrossRef Search ADS Pust Joel 2000: Intuitions as Evidence . New York: Routledge. Schulz Moritz 2014: ‘Counterfactuals and Arbitrariness’. Mind , 123, pp. 1021– 55. Google Scholar CrossRef Search ADS Smith Martin 2007: ‘Ceteris Paribus Conditionals and Comparative Normalcy’. Journal of Philosophical Logic , 36( 1), pp. 97– 121. Google Scholar CrossRef Search ADS Smith Martin 2010: ‘What Else Justification Could Be’. Noûs , 44( 1), pp. 10– 31. Google Scholar CrossRef Search ADS Stalnaker Robert C. 1968: ‘A Theory of Conditionals’. In Rescher Nicholas (ed.), Studies in Logical Theory , pp. 98– 112. Oxford: Blackwell. Sutton Jonathan 2007: Without Justification . Cambridge, MA: MIT Press. Van Rooij Robert 2006: ‘Free Choice Counterfactual Donkeys’. Journal of Semantics , 23( 4), pp. 383– 402. Google Scholar CrossRef Search ADS Wang Yingying 2009: ‘Counterfactual Donkey Sentences: A Response to Robert van Rooij’. Journal of Semantics , 26( 3), pp. 317– 28. Google Scholar CrossRef Search ADS Williamson Timothy 2005: ‘Armchair Philosophy, Metaphysical Modality and Counterfactual Thinking’. Proceedings of the Aristotelian Society , 105( 1), pp. 1– 23. Google Scholar CrossRef Search ADS Williamson Timothy 2007a: ‘Philosophical Knowledge and Knowledge of Counterfactuals’. Grazer Philosophische Studien , 74( 1), pp. 89– 123. Williamson Timothy 2007b: The Philosophy of Philosophy . Oxford: Blackwell. Google Scholar CrossRef Search ADS Williamson Timothy 2009: ‘Replies to Ichikawa, Martin and Weinberg’. Philosophical Studies , 145( 3), pp. 465– 76. Google Scholar CrossRef Search ADS Williamson Timothy 2013: ‘How Deep Is the Distinction between A Priori and A Posteriori Knowledge?’ In Casullo Albert, Thurow Joshua C. (eds.), The A Priori in Philosophy . Oxford: Oxford University Press. Yablo Stephen 2001: ‘Go Figure: A Path Through Fictionalism’. Midwest Studies in Philosophy , 25( 1), pp. 72– 102. Google Scholar CrossRef Search ADS © Geddes 2017
Mind – Oxford University Press
Published: Jan 1, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 12 million articles from more than
10,000 peer-reviewed journals.
All for just $49/month
Read as many articles as you need. Full articles with original layout, charts and figures. Read online, from anywhere.
Keep up with your field with Personalized Recommendations and Follow Journals to get automatic updates.
It’s easy to organize your research with our built-in tools.
Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera