This is a pre-peer-reviewed draft of a paper forthcoming in Pacific Philosophical Quarterly; please cite the official version. Counterfactuals and Epistemic Basing Relations This paper is about the epistemic basing relation, which is the relation which obtains between a belief and the reason for which it is held. It is sometimes said that the basing relation is the relation that obtains between a (doxastically) justified belief and the reasons which justify it. This second claim, which relates justification to the basing relation, may be correct, but it is not definitional, as the first claim is. It is not inconsistent to think that there can be doxastic justification even in the absence of a basing relation, but it is inconsistent to think that there can be a relation of believing for a reason even in the absence of a basing relation. In thinking about how to analyze the basing relation, we must be careful to keep questions of basing and justification distinct. Still, even granting that justification might not always require proper basing, it is clear that beliefs are very often justified by virtue of being based upon good reasons, so without an account of the basing relation, we cannot have a complete account of justification. I have two main goals in this paper. The first is to argue that counterfactuals have a central role to play in the basing relation; to that end, in section 1, I go through some recent accounts of the basing relation, pointing out the importance of counterfactuals along the way. My second goal is to argue for two related analyses, each of which has a plausible claim to be called an epistemic basing relation. The analyses each have a causal core, and they incorporate other elements which I pick up in my survey of recent accounts. They also incorporate what Marshall Swain calls a “pseudo-overdetermination” condition. This condition Counterfactuals and Epistemic Basing Relations 2 has been widely, and I think mistakenly, rejected. I defend the pseudo-overdetermination condition from objections in section 2. Section 3 is where I set out my positive accounts of the basing relation. 1. The importance of counterfactuals 1.1. Causation and counterfactuals I think that the most plausible way to think of the relation which obtains between distinct singular actual events C and E, where C is a cause of E, is in counterfactual terms: roughly, an event C is a cause of a distinct later event E just in case either E counterfactually depends on C, or there is a chain of counterfactually dependent events leading back from E to C.1 If this approach to causation is correct, then counterfactuals feature in the causal account of the basing relation, which is more or less the default way to think about the basing relation. The core of the causal account is that a belief B is based on a reason R just in case R non-deviantly causes B. I do not mean to argue for the counterfactual approach to causation here; I assume that it is plausible, and that it is therefore also a plausible way to think of the basing relation.2 Even if the Lewisian counterfactual approach to causation is misguided, however, it is clear enough that counterfactuals are central to several different accounts of basing. 1.2. Interventions 1 This rough account is from Lewis (1973). It does not take account of complications arising from questions about how to characterize the essential features of events, or the problem of pre-empting causes. A full account of causation in counterfactual terms would require that those things be spelled out, and apparent counterexamples would need to be explained away. But that is a whole project to itself, and for my purpose here, the required complications to the account make no difference, so I will work with Lewis’s early account to keep the exposition simple. 2 One reason for preferring the counterfactual approach to causation over rivals such as the energy-transference theory, though, is that the counterfactual approach is more general: it will work in possible worlds which function according to different physical laws from ours. Counterfactuals and Epistemic Basing Relations 3 In his (2012), Kevin McCain adopts James Woodward’s (2003) interventionist account of causation, in order to give a causal account of the basing relation which avoids problems posed by causal deviance and overdetermination. The idea behind the interventionist account of causation is that event/factor C is a cause of E in a system V just in case, if we were to perform an intervention on C, and hold all other variables in V fixed at their current values, there would be a change in E. To perform an intervention is to change the value of a variable in a system, in an idealized experimental setting: a setting where we would be in fact capable of changing the value of the variable, and of observing the result, and of holding all other variables in the system fixed. Intervention is an idealized operation; it is not constrained by reality or by empirical limitations. The interventionist account of the basing relation makes use of two causal notions. The first, the relation of being a direct cause, is a relation between types of variables: A necessary and sufficient condition for X to be a direct cause of Y with respect to some variable set V is that there be a possible intervention on X that will change Y (or the probability distribution of Y) when all other variables in V besides X and Y are held fixed at some value by interventions. (Woodward 2003, p.55) The other causal relation, that of being an actual cause, is a relation between token values of variables. Actual causes are defined in terms of directed causal paths and a redundancy range. There is a directed causal path from variable X to Y in a system “if and only if each variable starting with X and ending with Y is a direct cause of the variable that immediately succeeds it” (McCain 2012, p.361). The idea of a redundancy range is as follows: in a system V, where there is at least one direct causal path from X to Y, and Vi is a variable which does not lie on a path from X to Y, values v1...vn are on the redundancy range for Vi with respect to the path from X to Counterfactuals and Epistemic Basing Relations 4 Y just in case interventions on the values v1…vn, while holding the actual value of X fixed, would not result in any changes in the actual value of Y (ibid. p.361). To be clear, to say that Vi is not on a path from X to Y is not to say that there is no path from Vi to Y; Vi can be causally relevant to Y. The values v1…vn are on the redundancy range for Vi with respect to the path from X to Y if and only if changing the value of Vi from one of those values to another would not affect the value of Y, provided that we hold the value of X fixed. With these definitions in mind, scanting the details, we can say that “when X is an actual cause of Y it is possible to change the actual value of Y by changing the actual value of X while holding all other direct causes of Y that are not part of a chain of direct causes… leading from X to Y at a value within their redundancy range” (ibid. p.362). McCain defines the epistemic basing relation in interventionist terms, as follows: IB-R: S’s belief that p at t is based on her reasons, R, if and only if at t: (1) Each ri ∈ R is a direct cause of S’s believing that p AND (2) Each ri ∈ R is an actual cause of S’s believing that p AND (3) It is not the case that intervening to set the values of all direct causes of S’s believing that p, other than the members of R, to 0 will result in S’s not believing that p when every ri ∈ R is held fixed at its actual value. (ibid. p.364) This is a counterfactual account of the basing relation. The interventionist account of causation is explicitly counterfactual (Woodward 2003, p.196): it is explained in terms of the changes that would occur in the value of one variable, if we were to change the value of another one. My point here is not that I have a problem with McCain’s analysis of the basing relation; it takes a Counterfactuals and Epistemic Basing Relations 5 particular kind of counterfactual approach to causation, and it gives a causal account of the basing relation. Except for the lack of a pseudo-overdetermination condition, it is similar in spirit to the analyses I will end up with. But I do not take up the interventionist way of talking about causation and basing, because it seems unnecessarily metaphorical, referring as it does to what we would observe if we were to intervene and change the values of variables in idealized experimental settings. A more straightforward counterfactual approach to causation can dispense with talk of idealized experimental settings, and it can dispense with the idea of changing the values of variables; the straightforward counterfactual account of causation is just about whether the effects would be different if the (alleged) causal conditions were different. We can say everything we need to say about basing, I think, in more straightforward counterfactual terms. 1.3. Dispositions I Ian Evans identifies five “data” which must be accommodated by accounts of the basing relation (2013, pp.2946-2947): (1) Mutual basing: beliefs can be at least partly based on each other.3 (2) Multiple basing: beliefs can be simultaneously based on multiple, distinct reasons. (3) Backwards basing: beliefs can acquire new bases – we are capable of acquiring new reasons upon which we can base a belief, after we’ve already formed the belief for some other reason. (4) Basing termination: we can lose a reason for a belief, as when we discover that the reason is a bad one. We need not lose a belief when we lose a reason, though: we might have distinct reasons which sustain the belief. (5) Unconscious basing: we can have reasons for our beliefs of which we are unaware. 3 For example, the answers to crossword puzzles have to fit together, and when we see that the answers to clues 1 and 2 fit, it makes sense to think that the belief that the answer to clue 1 is X is at least partly based on the belief that the answer to clue 2 is Y, and vice versa. Or, for another example, consider cases of explicitly circular reasoning. Imagine a case where someone is arguing that he is not a brain in a vat, and as proof he puts forward his knowledge that he has hands, and then in support of his belief that he has hands, he puts forward his knowledge that he is not a brain in a vat. Perhaps reasoning in such a circular fashion is irrational, or perhaps it is legitimate – but certainly it is possible, and the analysis of the basing relation must allow it. Counterfactuals and Epistemic Basing Relations 6 For Evans, these five data must be accommodated by any adequate theory of the basing relation. He argues that his account of the basing relation accommodates all five of them, while rival accounts fail on one or more of them. According to an alternative, naïve sort of causal theory, for example, belief B is based on reason R just in case R is one of the actual causes of B. As everyone recognizes, this account fails to handle cases of causal deviance. Even more importantly, for Evans, this crude causal account cannot accommodate the phenomenon of basing termination. If R is among the reasons which caused S to have belief B, that fact will not be changed even if S later loses R. But it seems clear enough that, if S loses R, then R should no longer count as a basis for B. Causal sustainer theories – theories according to which B is based on R just in case R is among the reasons which causally sustain B – do a better job than the naïve causal account. Causal sustainer theories can accommodate the phenomenon of basing termination, because R can cease to sustain B, even if R did sustain B at some previous time. But, Evans argues, sustainer theories face the problem of “overabundant sustainers.” The idea is that there are many things which are causally responsible for S’s continued holding of belief B, which are clearly not reasons on the basis of which S holds B. For example, my continued breathing is a causal sustainer of all of my beliefs; if I stop breathing, I will die, and lose my beliefs. A causal theorist can reply to counterexamples of that sort, however, by pointing out that my continued breathing is not a potential reason for belief. The class of things that can count as potential reasons upon which beliefs can be based ought to be restricted to the kinds of things that we take into account when thinking about what to believe. Swain, for example, restricts the class of potential reasons to beliefs and other mental states, including desires, hopes, and nonpropositional states such as sensations. Imposing such a restriction is not a controversial move. Counterfactuals and Epistemic Basing Relations 7 In response to such a move, however, Evans points out (2013, p.2949) that there can still be cases where a mental state causally sustains a belief, even though the two are not appropriately related. In effect, Evans’s objection boils down to the fact that causal sustainer theories face causal deviance problems. For example, I could have the belief that it is sunny outside today, and the belief that the Maple Leafs won their hockey game last Saturday. It is also possible that I have a neurological condition which is such that my belief that it is sunny outside causally (but non-inferentially) sustains my belief that the Leafs won their game. These beliefs clearly have nothing to do with each other, and the only reason they are connected is because of a problem with my brain, but they meet the counterfactual causal sustaining condition. So, at the very least, a causal sustaining account of the basing relation must specify which sustainers count as bases for beliefs, if this kind of account is going to work. Doxastic theories of the basing relation, Evans argues, also do not succeed. Doxastic theories can say that in order for S’s belief B to be based on reason R, it is necessary that S have a meta-belief to the effect that R is a good reason for B. (Or they can say that the meta-belief is sufficient for basing.) But, Evans argues, such a meta-belief requirement violates datum (5): it is possible for one’s belief to be based on a reason even if one does not realize it, and even if one would not take that reason to be a good reason for the belief. One might, without realizing it, hold a belief about the quality of a colleague’s work at least partially on the basis of her gender, for example. Evans’s own account of the basing relation is the Dispositional Theory: DT: S’s belief that p is based on m iff S is disposed to revise her belief that p if she loses m. (2013, p.2952) Counterfactuals and Epistemic Basing Relations 8 This account is motivated, for one thing, by the following connection: “If one’s belief that p is really based on one’s belief that q, one responds to a loss of the belief that q by revising one’s belief that p” (ibid.) Evans argues that DT accommodates that connection better than a rival straightforward counterfactual account can (we will return to this point shortly). He also argues that DT accommodates all five data that an account of the basing relation should. The only datum that looks like it might be trouble for DT is multiple bases: if S’s belief has several different bases R1, R2, etc., each of which is sufficient for sustaining it, then it doesn’t look like S would be disposed to revise her belief if she lost one of them. Evans replies to this problem by pointing out that dispositions can be masked: one can be disposed to φ even if one does not φ, when something masks the disposition. A glass can be disposed to break if struck, even if it doesn’t break when struck (say, if it is packed in styrofoam). Similarly, a subject can be disposed to give up a belief in the face of the loss of a reason on which the belief is based, even if she doesn’t revise the belief when the reason is lost, because the presence of other sustaining reasons can mask that disposition. It bears mentioning, now, that we can be happy with most of what Evans has to say, while remaining committed to a causal/counterfactual approach to the basing relation. The main problem Evans raises for causal sustaining accounts, again, was the problem of overabundant sustainers, which boils down to a causal deviance problem. But, as we will see in the next section, there is a way to handle causal deviance problems, and it can be built into causal sustaining accounts of the basing relation. The main objection Evans raises against non-sustaining causal theories (in addition to the causal deviance problem), recall, is that, qualify them however we might, they don’t seem to be able to handle the phenomenon of basing termination. For a straightforward causal account, a Counterfactuals and Epistemic Basing Relations 9 belief B is based on a reason R just in case R caused (was a cause of) B – and if R caused B at some point in the past, then even if a subject loses R at some point, that will not change the facts of B’s causal history. This does appear to be a problem for a simple causal account of the basing relation. However, we need to distinguish two kinds of basing termination. One variety looks like a phenomenon that any theory of the basing relation needs to be able to accommodate; the other is an optional addition for a theory of the basing relation, but it is certainly not a requirement. The first kind of basing termination is: Basing termination 1: a subject S has a belief B that was caused by reason R1 at t1; at t2 S gains another reason R2 for B; at t3 S explicitly rejects R1 (or S explicitly rejects a support relation between B and R1), but retains B because of R2. The possibility of this kind of basing termination looks like something that any theory of the basing relation needs to allow for. (Indeed, I will explicitly build in a condition in the final analyses in section 3 in order to accommodate this requirement.) Once S explicitly rejects either R1 or a support relation between B and R1, it doesn’t look like R1 could be a reason for which she believes B any longer. And, although a naïve causal account of the basing relation does not allow for this kind of basing termination, it is easy enough to construct a subtler causal account. We could say, for example, that S holds B based on R at time t just in case R was a cause of S’s holding B at or prior to t, and S does not explicitly reject R at t (i.e. S does not disbelieve R at t, nor does he disbelieve that R supports B at t). This subtler causal account is still too crude for a complete account of the basing relation, but it handles the kind of basing termination that any account of the basing relation should handle. The other kind of basing termination is: Counterfactuals and Epistemic Basing Relations 10 Basing Termination 2: a subject S has a belief B that was caused by reason R1 at t1; at t2 S gains another reason R2 for B; at t3 S loses R1, perhaps simply forgetting about R1, but retains B because of R2. This kind of basing termination need not be accommodated by a causal account of the basing relation: we can, but need not, say that S no longer believes B on the basis of R1. An access internalist about justification would no doubt want to build this kind of basing termination requirement (or, better, justifier-termination requirement) into the theory of doxastic justification, which is fine with me; but there is no need for us to turn a substantive view about doxastic justification into a restriction on the epistemic basing relation. There is nothing obviously wrong with saying that forgotten reasons can remains reasons for which a belief is held, even if they are incapable of justifying the belief after they’re forgotten. So it is not obvious that any theory of the basing relation needs to accommodate Basing Termination 2. It also bears pointing out that we can construct a counterfactual analysis which can do everything that Evans’s DT does. A crucial step in the argument for the DT account of the basing relation is that it is supposed to be superior to a straightforward counterfactual account of the connection between a belief’s being based on a reason and of the fact that a subject would give up or revise the belief if she lost the reason. The straightforward counterfactual attempt to account for that connection is: CFT: S’s belief that p is based on m iff S would revise her belief that p were she to lose m. (Evans 2013, p.2951) CFT is inadequate because it is quite possible for S’s belief B to be based on a reason R, without its being the case that, were S to lose R, S would revise B. This happens in cases where there are multiple independent sustaining reasons upon which a belief is based. DT, on the other hand, can Counterfactuals and Epistemic Basing Relations 11 handle such cases: when S’s belief B is overdetermined by reasons R1 and R2, it can still be the case that S is disposed to revise B if he loses R1; it’s just that that disposition is masked by the presence of R2. DT is therefore superior to CFT. All of that is correct, but CFT can be revised to handle beliefs with multiple bases. One step in that direction is: CFT*: S’s belief B is based on each reason Ri in a set of reasons R, just in case were S to lose the entire set of reasons R, S would revise B. CFT* allows that beliefs can have multiple bases: any time there is a belief and a set of reasons, which are related such that if the subject lost all of the reasons, then he would lose the belief, the belief is based on that set of reasons. But CFT* is inadequate as it stands, because it allows in all sorts of completely irrelevant reasons into the basing relation. According to CFT*, for example, my belief that coffee is delicious is based on both of the following sets of reasons: R1 = {a certain kind of repeated sensory experience}; R2 = {a certain kind of repeated sensory experience, the sound of ocean waves}. While R1 is fine as a set of reasons for my belief, R2 is clearly not a set of reasons for my belief about the taste of coffee, because its second member is irrelevant to that belief. However, R2 is still related to my coffee-belief, such that if I lost both of its members, then I would revise my belief. So what we need is a principle like CFT*, but which does not allow irrelevant reasons to enter into the basing relation. Here is such a principle: CFT**: S’s belief B is based on each reason Ri in a set of reasons R, just in case each Ri contributes causally to S’s having B, and if S were to lose the entire set of reasons R, S would revise B. Counterfactuals and Epistemic Basing Relations 12 CFT** allows multiple bases, and excludes causally irrelevant factors from the set of reasons, so it is not vulnerable to the counterexample to CFT*. CFT** does the same work as DT, while remaining in an explicitly counterfactual framework.4 Of course, if talking about dispositions is just a convenient shorthand way to talk about complicated counterfactuals, then perhaps CFT** and DT aren’t really rivals. If dispositions are really just complicated counterfactuals, then the two definitions probably turn out to be equivalent; the counterfactual in CFT** which explicitly refers to what would happen if each of the overdetermining bases for a belief are absent is just making explicit the fact that dispositions can be masked, which is precisely the feature of dispositions which Evans makes use of in arguing that DT accommodates overdetermining reasons. Taking dispositions to be complicated counterfactuals in this way would be fine with me. But Evans does not want to link dispositions with counterfactuals; at least, he leaves it open whether they should be reduced to counterfactuals or not. My own view is that we should take dispositions to be reducible to counterfactuals; if we do not, then dispositions turn out to be really a mysterious sort of thing. We do of course have an intuitive grasp on when an object has a certain disposition, and that intuitive grasp is enough for our ordinary purposes in dealing with dispositions. But our intuitive grasp of dispositions surely relies on our understanding of what objects would do, if subjected to certain conditions – which is to say, our intuitive understanding of dispositions surely relies on our grasp of certain relevant counterfactuals. If we want to divorce dispositions from counterfactuals, then CFT** is superior to DT, since at least we know what CFT** means. 1.4. Dispositions II 4 CFT** employs the concept of causation, which some people will not take to be a counterfactual relation; still, even if they are right, the second conjunct, “if S were to lose the entire set of reasons R, S would revise B,” is explicitly counterfactual. Counterfactuals and Epistemic Basing Relations 13 John Turri’s recent account of the basing relation also employs dispositions, though in a different way than Evans does. The core of Turri’s account is the Necessity of Causation thesis: NC: R is among your reasons for believing Q (at time t) only if R causes or causally sustains your belief (at t). (2011, p.385) Turri’s argument for NC is that it is the only account which is capable of accounting for two important facts about the basing relation. The first is that reasons are difference-makers: when X is a reason for Y’s obtaining, then X makes a difference as to whether Y obtains. The second fact is that basing is not a “brute” relation: whenever a belief B is based on a reason R, there is always some further relation obtaining between B and R in virtue of which B is based on R. NC explains these two features: “Causation provides the metaphysical underpinning of basing, which explains why it isn’t brute. And causes are difference-makers, which explains why reasons are difference-makers” (ibid.). In defense of the claim that only NC can adequately account for these features of the basing relation, Turri argues against two rival accounts. The first is Marshall Swain’s. According to Swain, a belief B is based on a reason R just in case either R is a cause of B, or R would have been a cause of B in an appropriate set of counterfactual circumstances. Turri argues that Swain’s counterfactual (pseudo-overdetermination) approach does not respect the fact that reasons are difference-makers. He also reminds us of an apparently fatal sort of counterexample due to Tolliver (1982). We will come back to these objections to Swain’s account in the next section. Turri also argues against doxastic theories, which are a type of non-causal account of the basing relation. According to doxastic theories, “if you believe Q, and you believe P, and you judge that Q is good evidence to believe P, then your belief that Q is thereby among your reasons Counterfactuals and Epistemic Basing Relations 14 for believing P” (Turri 2011, p. 386). Turri objects, in my view correctly, that the doxastic approach to the basing relation “entails that it is impossible to judge that you have two good reasons to believe P but believe for only one of them” (ibid., italics in original). Since it is (perhaps irrational but) possible to be completely unaffected by a reason that one judges to be a good reason for believing, the doxastic theory is committed to an unacceptable consequence.5 NC, which says that causation is necessary for basing, therefore looks like a plausible core for an account of the basing relation. It is not, however, a complete account. An unsatisfactory way to turn it into a full account of basing would be to strengthen it into a biconditional, so that B would be based on R if and only if R is a cause of B. Such a causal analysis of basing fails to handle deviant causation.6 Turri’s proposal to handle that problem is to build a non-deviance condition into the causal analysis: CA: R is among your reasons for believing Q if and only if R non-deviantly causes your belief. (2011, p. 390) And to explain the non-deviance condition, Turri invokes the idea of the manifestation of cognitive dispositions. So the picture is that the basing relation consists of a reason’s being a non-deviant cause of a belief; deviance is ruled out by the condition that the causation must manifest a subject’s cognitive traits; and cognitive traits are understood as dispositions or habits 5 A related problem with doxastic theories is that they tend to run together questions of justification and questions of basing. Keith Korcz, for example, opens his paper, in which he defends a version of the doxastic theory, with the following general characterization of the basing relation: “The epistemic basing relation is the relation which must hold between a person’s belief and the adequate reasons for holding that belief if the belief is to be epistemically justified by those reasons” (2000, p.525). We need to keep in mind that it is one thing to hold a belief for a particular reason, and another thing for one’s beliefs to be justified by reasons. Some epistemologists (e.g. Ginet 1985) hold that a reason can justify a belief, if the subject judges that it is a good reason for holding the belief, even if it’s not the case that the belief is held for that reason. Now, perhaps that view about justification is right, or perhaps only reasons for which a belief is held are capable of acting as justifiers for it; we do not need to take a stand on the relation between justification and proper basing here. The point is that that is a debate which must be had only after we’ve already got an account of the nature of the basing relation in hand – the connection between basing and justification ought not to be taken as a datum which constrains the account of the basing relation itself. 6 An account that simply turns NC into a biconditional would also fail to handle Basing Termination 1, if the causal relation Turri has in mind is a relation of causal sustaining. In any case, the important thing here is the way Turri proposes to handle the causal deviance problem. Counterfactuals and Epistemic Basing Relations 15 to form or sustain beliefs in response to certain evidential circumstances. Turri’s ultimate analysis of the basing relation is: Causal-Manifestation Account (CMA): R is among your reasons for believing Q if and only if R’s causing your belief manifests (at least some of) your cognitive traits. (2011, p.393) Turri takes the manifestation-relation to be an undefined primitive in his account. After all, we are readily able to distinguish cases where a disposition is manifested, from cases where an object is disposed to have a type of effect, and the effect happens, but not because of the disposition. For example: a glass is packed in Styrofoam; it is dropped to the floor; the fall doesn’t break the glass, but someone simultaneously fires a gun into the package, and the glass breaks. The glass breaks when it falls, which is something that glasses are disposed to do, but the glass’s breaking does not manifest the disposition to break when it strikes the floor. It happened for another reason. We are just as capable of distinguishing the manifestation of cognitive traits from their non-manifestation. Consider the following case:7 Ed sees a car accident happening in the road ahead of him. He swerves to avoid the vehicles, spilling hot coffee on his lap in the process. Now he believes that he has spilled hot coffee on his lap (call this belief B2); and, in the causal history of this belief is Ed’s previous belief that a car accident was happening (call this belief B1). But clearly, Ed’s belief that he has spilled hot coffee in his lap is in no way based on his previous belief that a car accident was happening, and it is fairly clear that B1’s being a cause of B2 does not manifest any cognitive traits. There is a cognitive trait manifested in Ed’s forming belief B2, the belief that he’s spilled hot coffee in his lap: the formation of that belief manifests the cognitive trait of believing what his senses tell him. But that is not to say that B1’s being a cause 7 Plantinga (1993, p.69) gives a similar example. Counterfactuals and Epistemic Basing Relations 16 of B2 manifests a cognitive trait. By contrast, in a normal case, where I look in my cupboard and see a full coffee tin, and I form the belief that I have plenty of coffee in the cupboard, that beliefformation manifests the cognitive trait of believing that medium-sized objects are there when they appear to be there. So, since distinguishing cases where a disposition is manifest is something we are good at, it is reasonable to take the manifesting of dispositions as primitive in giving an account of the basing relation. However, a natural worry that might come up now is that perhaps this appeal to manifesting a cognitive trait will render the analysis of the basing relation circular: perhaps the reason we are able to distinguish cases where a reason’s causing a belief manifests a cognitive trait from cases where is does not is that we are responding to prior intuitions about whether the belief is based on the reason. The idea is that if we see that a belief is based on a reason, then we judge that it the reason’s causing the belief manifests a cognitive trait. If intuitions about basing underlie intuitions about the manifestation of cognitive traits, then it is circular to appeal to the manifesting of cognitive traits in giving an analysis of basing. But there are two responses to this worry. For one thing, it is not clear that the intuitive response to cases works in this way. After all, it could easily be that we judge that a belief is based on a reason because we see that the reason’s causing the belief manifests a cognitive trait, rather than the other way around. But more importantly, it seems to me that it is possible for cognitive traits to be manifested when a reason causes a belief, without instantiating an epistemic basing relation. For example, it seems to me that forgetfulness is a kind of cognitive trait (certainly it is a mental trait). Now, in a case where a reason causes a belief, and the only cognitive trait which that causal relation manifests is the trait of forgetfulness, then we will likely Counterfactuals and Epistemic Basing Relations 17 want to rule it as a case where the belief is not based on the reason. Here is an example of what I have in mind: Forgetful Jimmy Jimmy has discovered that Superman is Clark Kent. However, Jimmy is a forgetful person, and what with his deadlines at work, he forgets about Superman’s true identity. Now, Jimmy has a neurological condition which causes him to believe that, unless they are superheroes, all of his co-workers who wear suits to work are weak-limbed. (Jimmy reflectively rejects the connection between wearing a suit and being weak-limbed, but he just cannot help believing, of each of his co-workers who wear suits, that they are weak.) One day, he sees Clark wearing a suit to work, which, together with his neurological condition, causes him to believe that Clark is weak-limbed.8 In this case, Jimmy’s seeing Clark in a suit is a partial cause of his belief that Clark is weaklimbed. Now, it seems to me, the causal relation which obtains between Jimmy’s seeing Clark in a suit and his belief that Clark is weak manifests the cognitive trait of forgetfulness: had Jimmy not forgotten that Clark is Superman, he would not believe that Clark is weak. (The causal relation between what Jimmy sees and what he believes obtains because of, and not in spite of, his forgetfulness, which is why I think we must say that this causal relation manifests his forgetfulness.) Still, in spite of the fact that Jimmy’s seeing Clark in a suit is something that could be a reason for belief (visual experiences can be reasons), and that that potential reason is a Ok, but could Jimmy really forget that Clark is Superman? Maybe that’s too implausible. If so, we could easily just change the case, so that Jimmy never discovered that Clark is Superman; he just saw Clark do something one day that only a very strong person could do, but then later on he forgot about Clark’s strength, and reverted to his belief that Clark is weak-limbed. 8 Counterfactuals and Epistemic Basing Relations 18 cause of Jimmy’s belief about Clark’s strength, and that that causal relation manifests a cognitive trait (forgetfulness), this still appears to be a case of causal deviance. Jimmy does not believe that Clark is weak on the basis of the fact that he saw Clark wearing a suit. Two things are worth emphasizing about cases like this. The first is that, because they are possible, it follows that the recognition of the manifestation of a cognitive trait does not presuppose prior recognition of the instantiating a basing relation, so it does not render the analysis of the basing relation viciously circular to employ the manifestation of cognitive traits in the account. The second thing to say is that we need to specify just which cognitive traits are capable of giving rise to a basing relation, in order to be able to fully handle the problem of causal deviance. I have no principled proposal to offer, but it seems that we should rule out forgetfulness, absent-mindedness, and other “negative” traits like these. Perhaps this proposal is a bit ad hoc, but it is only a small addition to an otherwise quite plausible approach to the basing relation, and with this specification of Causal Manifestation condition, it looks like Turri’s account no longer faces any clear counterexamples of this sort. 2. Pseudo-overdetermination So far, we have seen that some plausible recent proposals for an account of the basing relation employ concepts that are, at bottom, counterfactual in nature. I don’t think that anything I’ve said so far is terribly controversial. If counterfactuals are an appropriate way to think of causation and dispositions, then they are of central importance for understanding the basing relation. And it is after all reasonable to think that counterfactuals are essential to causation and dispositions. We have also seen two conditions that I think we should build into the analysis of the basing relation. The analyses I propose will explicitly incorporate conditions to accommodate Basing Termination 1 and the Causal Manifestation condition. Counterfactuals and Epistemic Basing Relations 19 I want to move on now to consider Swain’s pseudo-overdetermination condition. This condition has been widely, and I think mistakenly, rejected in the literature on basing. In this section, I will briefly explain Swain’s analysis of the basing relation, as well as why pseudooverdetermination is an intuitively plausible addition to the analysis, and then I answer some important objections to it. 2.1. Swain’s analysis The rough idea that Swain aims to capture is that a belief B is based on a reason R just in case either R appears in B’s actual causal history, or else if some set of events in B’s actual causal history had not happened, then R would have caused B. Swain captures that rough idea in the following definition: (DB) S’s belief that h is based upon the set of reasons R at t = df. (1) S believes that h at t; and (2) For every member, rj, of R, there is some time, tn (which may be identical with or earlier than t), such that (a) S has (or had) rj at tn; and (b) there is an appropriate causal connection between S’s having rj at tn and S’s believing that h at t. (1981, p.74) Potential members of R are limited to beliefs and other mental states (not necessarily propositional ones). The appropriate causal connection in clause (b) is that either rj appears in h’s causal history at tn, or else rj is a pseudo-overdeterminant of an item which appears in h’s causal history at tn (1981, p.86-87). Pseudo-overdetermination is the relation which obtains between two occurrent events C and E, when C is not an actual cause of E, but if some set of events in E’s actual causal history had not happened, and both C and E had happened anyway, then C would Counterfactuals and Epistemic Basing Relations 20 have been a cause of E.9 Pseudo-overdetermination is not a real causal relation, because C does not appear in E’s causal history. However, whether C happens is relevant to whether E happens, in close counterfactual circumstances. It is a cause-like relation, very similar to the relation of overdetermination. The reason for including the pseudo-overdetermination condition in the analysis of the basing relation is to handle cases like Lehrer’s gypsy-lawyer. That case, though complicated, is by now a familiar one in discussions of the basing relation. It will be worth it to go through the case here, in order to see the importance of the pseudo-overdetermination condition, and also to make sense of Korcz’s modified, doubtful gypsy-lawyer cases. The original gypsy-lawyer case is meant to show that causal accounts of the basing relation are too weak, i.e. that they count cases of basing as cases of non-basing. Korcz’s doubtful gypsy lawyer cases are meant to show that if we add the pseudo-overdetermination condition to a causal account, then the resulting analysis is too strong, i.e. it will count cases of non-basing as cases of basing. In the original gypsy-lawyer case, there is a lawyer who is also a gypsy, and who has full faith in the reading of his cards. The lawyer is defending a man who stands accused of eight murders. The case is very emotionally charged, and everyone wants to believe that the defendant is really guilty of all the murders. The lawyer is subject to the same emotional factors as everyone else, and he believes that his client is guilty. He proceeds to do a reading of the cards, however, and they indicate that his client is innocent of the eighth murder. Because of his faith in the cards, the lawyer forms the belief that his client is innocent of that murder. He goes on to reexamine all of the evidence, and he realizes that there is a complicated line of argument proving There is a wrinkle that accounts like Swain’s have to incorporate. Kvanvig (1985) points out that one counterfactual may be “buried” by another counterfactual, which Swain’s account does not make room for. I will ignore this wrinkle here and in what follows; it can be handled, but it complicates things further. See Korcz (1997) for discussion. 9 Counterfactuals and Epistemic Basing Relations 21 his client’s innocence of the eighth murder. He sees the connection between the evidence and his client’s innocence, and he bases his belief on that connection. However, he also recognizes that, if not for his faith in the cards, the emotional factors involved in the case would cloud his judgment, so that he would not be able to see the connection between the evidence and his client’s innocence. What appears to be happening in this case is that the lawyer is basing his belief in his client’s innocence of the eighth murder both on his reading of the cards, and on the complicated line of argument from the evidence which he later discovered. It does not seem right to say that there is a causal path from the evidence to the lawyer’s belief in his client’s innocence, however. His faith in the cards causes his belief in his client’s innocence. If the argument from the evidence was also an actual cause of his belief, then it would be an overdetermining cause. But it is not an overdetermining cause, because it is not the case that, if the lawyer lost his faith in the cards, he would still believe in his client’s innocence on the basis of the complicated argument from the evidence. The lawyer would lose both his recognition of the force of the evidence, and his belief in his client’s innocence. However, as Swain points out (1981, p.90), although the argument from the evidence does not genuinely overdetermine the lawyer’s belief in his client’s innocence, it does pseudooverdetermine his belief: holding constant the lawyer’s belief in his client’s innocence and his recognition of the evidential argument in support of that belief, and removing his faith in the cards, it would be the case that the evidential argument causes the belief.10 Tierney and Smith (2012) argue that the gypsy lawyer case, as well as Lehrer’s later case of Raco the racist doctor, shows that Swain’s account is mistaken. Their objection only considers the actual-causal-history aspect of Swain’s account, including genuine overdetermination, however. It does not take into account the pseudooverdetermination condition, which handles these cases. 10 Counterfactuals and Epistemic Basing Relations 22 So that is Swain’s account and the rationale for including a pseudo-overdetermination condition. Let’s move on now to look at some objections which have been raised against it. 2.2. Objection 1: the doubtful gypsy lawyer Keith Korcz has argued that the pseudo-overdetermination condition is too strong. He argues that Swain’s account yields the wrong result in the case of the doubtful gypsy lawyer, which is a variant on the original gypsy lawyer case. The relevant features of the doubtful gypsy lawyer case are: the lawyer performs a reading of the cards, which indicates that, for reasons to do with her client’s character, her client could not have committed the eighth murder. There is a good line of reasoning leading from the evidence to her client’s innocence of the eighth murder, which has to do with the fact that her client did not have access to the murder weapon. The lawyer dismisses that line of reasoning, though, because the cards said that her client is innocent because of her character.11 And the lawyer knows that she never has any uncaused doubts.12 So the relevant features of the set-up in question are: (1) the card reading occurs (2) the lawyer doubts the complicated line of reasoning showing her client’s innocence (3) the lawyer’s doubt described in (2) is based on the card reading (4) the lawyer has no uncaused doubts, and knows that she has no uncaused doubts (Korcz 1997, p.178). 11 But surely, it might reasonably be objected, the lawyer could easily believe both that the murder was inconsistent with her client’s character, and that her client didn’t have access to the murder weapon? I am sympathetic to this objection. To make sense of the case, we need to add something like: the cards indicated that her client did have access to the murder weapon, so any apparently good line of argument indicating otherwise must be mistaken. 12 This last point is only important for evaluating the comparative distance of possible worlds, if we tried to argue that the complicated line of argument does not pseudo-overdetermine her belief. We can ignore it here, as I will not be making that kind of argument. Counterfactuals and Epistemic Basing Relations 23 The idea is that it is clear that the lawyer’s belief in her client’s innocence of the eighth murder is not based on the complicated line of reasoning, since she explicitly doubts that line of reasoning. However, Korcz claims, it looks like the complicated line of reasoning pseudo-overdetermines her belief in her client’s innocence, since if the card reading had not occurred, the complicated line of reasoning would cause her to believe in her client’s innocence.13 I think that we should agree that the complicated line of reasoning pseudooverdetermines the lawyer’s belief in her client’s innocence, in the case as Korcz describes it. Even so, we should not agree that Swain’s account entails that her belief in her client’s innocence is based on the complicated line of argument, because the line of argument is not a reason that the lawyer has. As the case is set up, the lawyer is explicitly doubtful of the line of reasoning, so we cannot count it among her reasons. For Swain, the set of things that can be reasons includes a subject’s beliefs and other non-belief mental states, but nowhere does he say that we can include things that a subject doubts as potential reasons. (And there is no reason why we should add the claim that things that people doubt are among their reasons for belief to Swain’s account.) We can count the complicated line of reasoning which the lawyer doubts as among the reasons that she would have had, if she had not performed the card reading, but we cannot count it as among the reasons that she actually had. So, in spite of the fact that the complicated line of argument pseudo-overdetermines her belief in her client’s innocence, the line of argument is not a basis for her belief, on Swain’s account.14 13 This is an important difference from the original case of the gypsy lawyer, where the emotional factors of the case are so strong that the lawyer would not recognize the complicated line of argument if he had not done the card reading. 14 In his (2000, p.531), Korcz gives a slightly modified doubtful lawyer case, where the lawyer does believe the complicated line of reasoning; she just doesn’t believe that the line of reasoning supports the belief in her client’s innocence. This revised case is too strange to trust our intuitions, if we even have any intuitions about it, since it asks us to imagine a lawyer who recognizes that her client had no access to the murder weapon, but who explicitly doubts that that fact is relevant to whether her client committed the murder. Such a lawyer does not seem to have coherent doxastic attitudes, which makes it difficult to judge pre-theoretically whether there is a relation of believing Counterfactuals and Epistemic Basing Relations 24 2.3. Objection 2: the pendulum case Joseph Tolliver (1982) gave what is probably the most influential objection to Swain’s account. His counterexample is a genuine, but not insurmountable, problem for the pseudooverdetermination condition. The objection is that there can be cases where a belief B1 is explicitly held on the basis of another belief B2, and B2 is not held on the basis of B1, but B1 pseudo-overdetermines B2. His example is the pendulum case. In this case, a student learns that we can calculate the period of a pendulum from its length, and that we can calculate the length of a pendulum from its period. The student observes a pendulum with a length L, and calculates that it has period P. In this case, the student’s belief that the pendulum has period P is based on his belief that the pendulum has length L. His belief that the pendulum has length L is not based on his belief that it has period P; the calculation went the other way around. The problem for Swain’s theory is that the belief that the pendulum has period P does pseudo-overdetermine the belief that it has length L. That’s because, if we take away the actual cause of the belief that the pendulum has length L, and we leave the belief that it has period P in place, then the student’s belief that the period has length L would be caused by his belief that it has period P. (Any other explanation of the student’s having both beliefs, such as that his teacher told him both the length and the period, can be ruled out by stipulation.) We cannot handle the pendulum case, as we handled the case of the doubtful gypsy lawyer, by saying that the belief that the pendulum has period P is not among the set of potential for a reason present in this case. We can imagine a lawyer who thinks that there is a powerful reason for thinking her client is innocent of the murder, which is unrelated to whether he client had access to the murder weapon, as in Korcz’s case which I’ve discussed in the main text – but I doubt that we can really imagine a lawyer who thinks that access to the murder weapon is irrelevant to her client’s innocence. Korcz’s revised doubtful lawyer case does not, therefore, give us a good reason to reject the pseudo-overdetermination condition. Counterfactuals and Epistemic Basing Relations 25 reasons that the student has. It is a belief, and beliefs are the clearest type of thing that must be counted as potential reasons. A tempting response to the pendulum case is to add a temporal priority condition in the analysis. Such a response ultimately fails, but it is useful to see why it fails. The condition is: Temporal Priority: S’s belief B is based on R at t only if B was not already a cause or pseudo-overdeterminer of S’s having R before t. If we were to add a condition like Temporal Priority to Swain’s analysis, it would no longer be vulnerable to the pendulum case. The student’s period-belief is first caused by his length-belief, and only once that causal relation has been established does it become the case that the student’s period-belief pseudo-overdetermines his length-belief. However, there are two obvious problems with Temporal Priority. The first is that it rules out cases where S believes B on the basis of R at t1, but then realizes that she’s got things backwards, so at t2 she ceases to believe B on the basis of R; instead, she believes R on the basis of B. For example, suppose that Norma believes that her friend Marvin is untrustworthy when it comes to financial advice, though she doesn’t have any good evidence for that belief. One day, Marvin advises Norma to withdraw all her money and bury it in a hole in the ground. Because Norma believes that Marvin is no good with money matters, she believes that his most recent piece of advice is a bad one. But then she reflects that she hasn’t really had any good evidence for thinking that Marvin is untrustworthy, so she ceases to base her belief that his recent piece of advice is bad on her belief in Marvin’s untrustworthiness. But Norma also realizes that burying all her money in a hole is a bad idea, considered on its own merits. So she proceeds to base her belief in Marvin’s untrustworthiness on her belief that his most recent piece of advice was a bad Counterfactuals and Epistemic Basing Relations 26 one. This kind of basis-switching seems to be possible, but Temporal Priority entails that Norma cannot switch to believing in Marvin’s untrustworthiness on the basis of his bad recent advice. The second problem for Temporal Priority is that it excludes most cases of circular basing (which we saw above, as Evans’s first datum for an account of the basing relation, the datum of Mutual Basing). Suppose that S believes B1 on the basis of B2, and S believes B2 on the basis of B3,… and S believe Bn on the basis of B1. Perhaps this kind of case is epistemically legitimate, or perhaps not, but surely it is possible for such a series of circularly-based beliefs to occur. But Temporal Priority entails that Bn cannot be a reason for which B1 is held, because B1 was already a cause of Bn. A better principle, which handles the pendulum case without excluding the kinds of cases that Temporal Priority does, is the following. Again, if we are considering whether a belief B is based on reason R at time t, the condition will be: Causal Priority: S’s belief B is based on R at t only if: if B was an actual cause of R before t, then R is (not a pseudo-overdeterminer, but) an actual cause of B at t. Causal Priority says that pseudo-overdetermination of B by R establishes that B is based on R only if B was not already an actual cause of R. The idea that Causal Priority aims to capture is that actual causes take precedence over pseudo-overdeterminers in determinations of whether a basing relation obtains. Adding Causal Priority to Swain’s analysis allows it to handle the pendulum case: the student’s period-belief was actually caused by his length-belief, so Causal Priority entails that the fact that the student’s period-belief pseudo-overdetermines his lengthbelief does not establish a basing relation of his length-belief upon his period-belief. Causal Priority also allows both basis-switching and circular basing. Both types of case involve actual causation of B by R at some time t1, and then actual causation of R by B at some Counterfactuals and Epistemic Basing Relations 27 later time t2. The only way we could generate a counterexample to Causal Priority would be to come up with a case where (1) B is non-deviantly caused by R, which establishes that B is based on R, and (2) where B pseudo-overdetermines R, and (3) where it would have to be clear that B’s pseudo-overdetermining R clearly generates a basing relation of R upon B. But Tolliver’s pendulum case is precisely a case with features (1) and (2), and the very point of Tolliver’s case is that it is obvious that R is not based on B by virtue of B’s pseudo-overdetermining R. It looks like Causal Priority is therefore not vulnerable to counterexamples, so we can add it to Swain’s account to handle pendulum cases. 2.4. Objection 3: reasons as difference-makers Turri objects to Swain’s pseudo-overdetermination condition on the ground that reasons are difference-makers. Pseudo-overdeterminers do not appear to be difference-makers, though, because they do not enter into the actual causal histories of what they pseudo-overdetermine. Turri gives a persuasive example, where a very good pitcher sits out a whole game; another pitcher started the game and threw a shutout, and helped his team to victory. The starting pitcher was an important causal factor in bringing about his team’s victory: he was a difference-maker. The one who sat out, however, made no difference at all – even supposing that, were he to have pitched, he would also have helped his team to victory. He made no difference, because the pitcher who actually started pitched the whole game, and it was the starting pitcher who is responsible for the win. We can agree with Turri that reasons are difference-makers. But we do not need to agree that reasons must be actual difference-makers; they can be difference-makers in close counterfactual situations. Pseudo-overdetermining factors do not provide us with actual causal explanations – they are not actual causes – so if an explanation of why an event has occurred, Counterfactuals and Epistemic Basing Relations 28 given in terms of the actual causal chain(s) which led up to its occurrence, is what we mean by “reason,” then pseudo-overdeterminers are not reasons. But I see no reason to say that reasons upon which a belief is based must always be actual causes; close counterfactual causes of belief, of which a subject is aware (as in the gypsy lawyer case), seem like they ought to count as reasons for belief. Recall that in the gypsy-lawyer case, the lawyer recognizes that there is a complicated line of reasoning which supports his belief that his client is innocent of the eighth murder. The lawyer also recognizes, however, that he is vulnerable to emotional factors involved in the case, and if it were not for his faith in the cards, he would be unable to see the connection between his client’s innocence and the complicated line of reasoning. Still, he does recognize that the line of reasoning supports his client’s innocence, and if he were able to do so, he would believe in his client’s innocence on that basis alone. In this case, it seems that we should say that the complicated line of reasoning is a reason for which the lawyer believes in his client’s innocence. It is because of such cases that we need a pseudo-overdetermination condition in a causal analysis in order to handle cases like the gypsy-lawyer: although the complicated line of reasoning does not cause the lawyer’s belief in his client’s innocence, it does pseudooverdetermine his belief, so we can say that a basing relation obtains in such cases, if we include a pseudo-overdetermination condition in the analysis. 2.5. Objection 4: against gypsy-lawyer cases A final objection to consider is that gypsy-lawyer cases trade on the conflation of questions of justification and questions of basing, so there is no need to adopt a pseudo-overdetermination condition in response to them. Mittag (2002, p.549) argues that all discussions of gypsy-lawyer cases assume that the lawyer is justified in some sense in his belief, by the complicated line of reasoning, and then they use that intuition to generate the verdict that his belief must be based on Counterfactuals and Epistemic Basing Relations 29 the complicated line of reasoning. It follows that if we simply reject the claim that the lawyer’s belief is justified (which some epistemologists want to do), or if we reject the view that justification requires proper basing, then we have no reason to accept that his belief is based on the line of reasoning. I do not find this objection convincing. The standard distinction between propositional and doxastic justification is that between beliefs for which a subject is in possession of good reasons (or whatever other good-making feature of beliefs you like) are propositionally justified, while beliefs which are held on the basis of good reasons are doxastically justified. This is not a universally accepted distinction, but it is very common. Now, it is true that the lawyer in the gypsy-lawyer case appears to be justified in some sense in his belief, by the complicated line of reasoning. Clearly he has propositional justification, in the form of the complicated argument from the evidence. And it also seems that he has doxastic justification on that basis; he recognizes that he has good reason for his belief, so he is believing exactly as he ought. But, it seems to me, the judgment that the lawyer has doxastic justification is a result, not a cause, of the intuition that his belief in his client’s innocence is based on the complicated line of reasoning.15 The intuition about the lawyer’s belief being based on the line of reasoning is not dependent on the intuition about his justification, so it is not an intuition that we can simply reject by rejecting the intuition that his belief is justified, or by rejecting the essential connection between doxastic justification and proper basing. 3. Two counterfactual causal basing relations 15 At least, that is the way I myself have understood the case. At first, I found the case so complicated that I did not want to say that the lawyer was doxastically justified in his belief; but, in re-reading and thinking about the case, I came to think that he really does seem to base his belief on the line of reasoning. It was only after that, that I came to think that his belief is also doxastically justified. Counterfactuals and Epistemic Basing Relations 30 We have now seen that, if counterfactuals are in fact central to both causation and dispositions, as I think, then counterfactuals are implicitly central in several recent attempts to characterize the basing relation. We have seen that we need Swain’s pseudo-overdetermination condition, or something very much like it, to allow causal accounts of basing to handle gypsy-lawyer cases. And we have seen that the pseudo-overdetermination condition can be saved from the objections which have been thought to be fatal to it. What I want to do now is to set out two versions of what I take to be the most plausible characterization of the basing relation. One is a causal-history relation, and the other is a causalsustainer relation. They both have a claim to be called epistemic basing relations, and their independent appeal is probably part of what has driven the internalism-externalism controversy in epistemology. Both accounts draw on the plausible aspects of Swain’s, Turri’s, and Evans’s accounts. From Swain, we can take the formulation of the causal relation and the pseudooverdetermination relation. As it stands, however, Swain’s causal-or-pseudo-overdetermination account is vulnerable to causal deviance problems. This point is rarely remarked upon, since causal deviance is such a ubiquitous type of problem, and since the really controversial element in Swain’s analysis is the pseudo-overdetermination condition, but we do need to add a condition to avoid causal deviance. For that purpose, we can adopt Turri’s causal-manifestation condition. Furthermore, from Evans, we need to recognize the importance of accommodating the phenomenon of basing termination, or at least the kind of basing termination identified in Basing Termination 1. We can also take up Evans’s observation that if a subject holds a belief on the basis of a reason, then she is disposed to give up the belief if she loses the reason. This Counterfactuals and Epistemic Basing Relations 31 observation lends support to a causal-sustainer account of the basing relation. Finally, as a response to Tolliver’s pendulum case, we can incorporate the Causal Priority condition. So the elements we have are: a causal approach to the basing relation, which can be understood either in a causal-history sense or a causal-sustaining sense; a pseudooverdetermination condition; a cognitive-manifestation condition, to handle problems of deviant causation; a basing termination condition; and a causal priority condition. Putting these together, we get the following two accounts of the basing relation: Basing-History: S holds belief B for reason R at time tn just in case: (i) S believes B at tn; and (ii) Either: Both: S has R at tn, and R is a direct cause of B or a pseudo-overdeterminer of B; or Both: S had R at tm and R was either a causal ancestor of B at tm or a pseudo-overdeterminer of a causal ancestor of B at tm; and (iii) R’s being a direct cause, or a pseudo-overdeterminer, or a causal ancestor, or a pseudo-overdeterminer of a causal ancestor, of B manifests a cognitive trait of S’s; and (iv) Either R is an actual cause or causal ancestor of B, or B is not an actual cause or causal ancestor of R; and (v) It is not the case that S has explicitly eliminated R from her reasons for believing B at tn or at some point between tn and tm. Condition (i) is obviously necessary. Condition (ii) makes this a causal-history-or-pseudooverdetermination analysis. Condition (iii) is Turri’s way to block causal deviance. Condition (iv) is a version of the Causal Priority condition from section 2.3, formulated as a disjunction Counterfactuals and Epistemic Basing Relations 32 instead of a conditional, and modified to include the ancestor-relation. The point of this condition is that pseudo-overdetermination does not always establish a basing relation: if R is an actual cause of B, and B is a pseudo-overdeterminer of R, then B is based on R, and R is not based on B. We need this condition to rule out pendulum cases. Condition (v) accommodates Basing Termination 1. Although Swain himself explicitly rejects anything like Basing Termination 1 in his own analysis (1985, p.89), we need such a condition in a causal account of the basing relation. What we want in an analysis of the basing relation, after all, is an account of the reasons for which a subject S holds a belief at a given time; once S has explicitly excluded R from her reasons for believing B, R is no longer causally relevant in any sense to whether S believes B at that time. If she continues to believe B anyway, it is because of some other reason. But bear in mind that condition (v) does not exclude R from the set of reasons for which S holds B if S loses R in some way other than by explicitly excluding it from her reasons for B, such as by forgetting about it. When a reason is forgotten, it is no longer a causal sustainer of B (the job of causal sustaining will most likely go to S’s seeming to remember that B), but it is still relevant to a causal explanation of why S holds B. Consider, for example, Goldman’s (1999) case of Sally. Sally believes that broccoli is healthy to eat; she learned that from reading an article in a reputable newspaper. Later, Sally forgets the source of her belief. Still, Goldman thinks, her belief continues to be justified by the fact that it came from a reputable source. We can explain his intuition that the source of the belief still justifies the belief by the fact that the source of Sally’s belief is still part of the causal explanation of why she holds her belief. The second account of the basing relation is: Basing-Sustaining: S holds belief B for reason R at time t just in case: Counterfactuals and Epistemic Basing Relations 33 (i) S believes B at t; and (ii) S has R at t and R is either a causal sustainer of B at t or a pseudo-overdeterminingsustainer of B at t; and (iii) R’s being a direct cause or a pseudo-overdeterminer of B manifests a cognitive trait of S’s; and (iv) Either R is an actual cause of B, or B is not an actual cause of R. This analysis is just like Basing-History, with the exception that condition (ii) only allows that current causal sustainers or pseudo-overdeterminers can be bases for beliefs. Basing-Sustaining does not need to explicitly include anything like condition (v) from the Basing-History analysis, to handle basing termination, because that is already entailed by conditions (ii) and (iii) of Basing-Sustaining: if R is a non-deviant causal or pseudo-overdetermining sustainer of B for S at t, then it is not the case that S has explicitly excluded R from her reasons for believing B at t. (Non-deviance is guaranteed, recall, by the cognitive-manifestation requirement in condition (iii).) Both Basing-History and Basing-Sustaining have a claim to be called basing relations. They both give an account of what it is for a reason to be a reason for which a belief is held. But it bears emphasizing that Basing-History and Basing-Sustaining need not be seen as competing accounts of the basing relation; they are simply two different relations. They only become competitors when it comes to deciding what sort of basing relation must obtain between beliefs and the reasons which justify them. Basing-History, it seems to me, is more amenable to externalist views of doxastic justification, since R can enter into B’s causal history without being accessible to S at a later time (S might forget about R). Basing-Sustaining seems to be more amenable to internalist views about doxastic justification, since it restricts the relata of the basing Counterfactuals and Epistemic Basing Relations 34 relation to beliefs and reasons which a subject has a given time and which causally sustain or pseudo-overdetermine her belief at that time. That means that only internally accessible reasons are reasons for which beliefs are held, in this sense. 4. Conclusion We have seen that counterfactuals are central to the account of the basing relation. If we take causation and dispositions to be best understood in counterfactual terms, then counterfactuals are implicit in several recent analyses of the basing relation. Whether or not counterfactuals are the right way to think of causation and dispositions, though, we need to build a pseudooverdetermination condition into the analysis, to handle gypsy lawyer cases, and the pseudooverdetermination condition is explicitly counterfactual. The analyses which I have proposed here are perhaps somewhat complicated, but each of the conditions which I have included is well-motivated. The causal core of the account is simply the default view of the basing relation, and we also saw Turri argument in support of it, which seemed reasonable. We needed to add the pseudo-overdetermination condition to handle gypsylawyer-types of objections to causal analyses, and we needed to add the Causal Priority condition to handle pendulum-types of objections to the pseudo-overdetermination condition. We also needed to add Turri’s cognitive manifestation condition, to handle causal deviance, as we needed to add a condition to accommodate Basing Termination 1. Finally, it seems to me that we should allow that there are two important but distinct kinds of causal relation which can give rise to a basing relation: the relation of being a causal sustainer of a belief, and the relation of being in a belief’s causal history. There is a case to be made for thinking of each sort of relation as a basing Counterfactuals and Epistemic Basing Relations 35 relation. The result of combining these various elements is, I hope, two well-motivated and counterexample-free analyses of epistemic basing relations. References Evans, Ian (2013). The Problem of the Basing Relation. Synthese. vol.190, pp.2943-2957. Ginet, Carl (1985). Contra Reliabilism. The Monist. vol.68, no.2: Knowledge, Justification, and Reliability (Part II), pp.175-187. Goldman, Alvin (1999). Internalism Exposed. Journal of Philosophy. vol.96, no.6, pp.271-293. Korcz, Keith (1997). Recent Work on the Basing Relation. American Philosophical Quarterly. vol.34, no.2, pp.171-192. (2000). The Causal-Doxastic Theory of the Basing Relation. Canadian Journal of Philosophy. vol.30, no.4, pp.525-550. Kvanvig, Jon (1985). Swain on the Basing Relation. Analysis. vol.45, no.3, pp.153-158. Lewis, David (1973). Counterfactuals. Oxford: Blackwell. McCain, Kevin (2012). The Interventionist Account of Causation and the Basing Relation. Philosophical Studies. vol.159, pp.357-382. Mittag, Daniel (2002). On the Causal-Doxastic Theory of the Basing Relation. Canadian Journal of Philosophy. vol.32, no.4, pp.543-559. Plantinga, Alvin (1993). Warrant and Proper Function. Oxford: Oxford University Press. Swain, Marshall (1981). Reasons and Knowledge. Ithaca, NY: Cornell University Press. Counterfactuals and Epistemic Basing Relations 36 Tierney, Hannah, and Smith, Nicholas (2012). Keither Lehrer on the Basing Relation. Philosophical Studies. vol.161, pp.27-36. Tolliver, Joseph (1982). Basing Beliefs on Reasons. Grazer Philosophische Studien. vol.15, pp.149-161. Turri, John (2011). Believing for a Reason. Erkenntnis. vol.74, pp.383-397. Woodward, James (2003). Making Things Happen. Oxford: Oxford University Press.