1. What the evidence actually supports vs. what one takes one's evidence to support: the need for weakening the impossibility-claim

Epistemologists shouldn't need reminding that, at least pace certain very radically subjective Bayesians, there is a distinction between what one's evidence actually supports and what one takes it to support.Footnote ⁵ I might take the fact that I drew a tarot card showing The Hanged Man to be strong evidence that some terrible fate will befall me tomorrow, but in fact it isn't strong evidence of this. Moreover, not all failures to correct identify what one's evidence supports involve stupidity or superstition or gross, manifest irrationality in the way that this first example (on most precisifications) does. For example, whatever doxastic attitudes the evidence justifies having about various complex philosophical questions – like whether we have free will, or whether the mental is reducible to the physical – it is highly non-obvious that it does so, and some intelligent, thoughtful people are misidentifying what the evidence does support.Footnote ⁶ Such failures also need not involve self-deception or lack of awareness of one's own mental states.

Given the obviousness – and clear epistemological importance – of the difference between what one's evidence supports and what one takes it to support, it is odd that in debates about topics like believing at will, the possibility of responding to non-evidential reasons, and the like, it seems to go out the window. So, for example, Adler (Reference Adler2002a: 8) makes the astounding claim that “believing in opposition to one's evidence is motivationally unintelligible”, when presumably he meant instead to make the less astounding (though still bold) claim that believing in opposition to one's own judgment about the evidence is motivationally unintelligible.Footnote ⁷ (His paper is about the (alleged) impossibility of epistemic akrasia, which is usually understood as involving defiance of one's own judgment about what the evidence, not just a failure to believe what the evidence in fact supports.) Examples of people failing to believe what their evidence actually supports are clearly intelligible and prevalent. This phenomenon need not come about because of their willfully defying what they recognize the evidence to support (as in cases of epistemic akrasia), but can happen in a more anodyne way when people are just mistaken about what their evidence supports.

Other examples abound. Here are just a few:

• • Shah (Reference Shah2003: 462) writes that “one particular belief-forming process, reasoning, is regulated solely by evidential considerations” (his italics). This isn't right, at least if an “evidential consideration” is a consideration that is actually evidence rather than one that the agent takes to be evidence.Footnote ⁸ Considerations that one takes to be, but aren't actually, evidence for p can certainly influence one's reasoning about whether p. At most it's considerations that one doesn't take to be evidence that can't feature in reasoning.

• • Noordhof (Reference Noordhof2004: 247) writes that “if we are consciously attending to the question of whether p, we cannot help but make a judgment in line with what the evidence gives us grounds for believing true”. This isn't right. Lots of people consciously attend to the question of whether p and end up with a judgment that is not in line with what the evidence gives them grounds for believing true; uncontroversially, this can happen when they misevaluate their evidence. At most it's true that such people can't help but make a judgment in line with their own evaluation of what the evidence gives them grounds for believing true.

• • Gendler (Reference Gendler2008: 565) writes that “belief aims to ‘track truth’ in the sense that belief is subject to immediate revision in the face of changes in our all-things-considered evidence.” This isn't right. A change in my all-things-considered evidence may have absolutely no effect on my belief if I don't recognize it as evidence. At most, belief is subject to immediate revision in the face of perceived changes in our all-things-considered evidence.

The list goes on. You may think I'm being pedantic or uncharitable here, and that the quoted authors obviously meant to make the calmer, more careful claims about what one takes the evidence to support. But at the very least they are failing to be clear about the distinction I'm insisting on (and I find it genuinely puzzling why the same mistake keeps repeating itself). Moreover, as we'll see over the next few sections, brushing over the distinction has very real effects on the debate about evidentialism.

Before that, I want to note a slightly distinct, though related, debate where inattention to the distinction also causes substantive errors. In the debate about whether we have responsibility or control over our beliefs, it's often said that our beliefs are “controlled by the evidence,” “determined by the evidence,” or “at the mercy of the evidence”; belief is something that “happens to us, not something we do”; it's not “up to us.”Footnote ⁹ That's supposed to set a puzzle for how we can have control over them or responsibility for our beliefs (and, to some, definitively shows that we don't have such control or responsibility). But the initial premise is wrong: our beliefs aren't just controlled or determined by the evidence, because the effect of our evidence on our beliefs is mediated by our evidential standards: what we take to be evidence for what. Plausibly, since our evidential standards belong to us, this grounds a sense in which our beliefs are (partially) products of our agency and are not just imposed from without by the evidence.Footnote ¹⁰

The lesson of this section is just the one that I previewed in the Introduction. We are going to be considering arguments for evidentialism that rely on the premise that it is impossible to respond (directly and consciously) to non-evidential consideration in belief formation. Taken on its most natural reading, ‘non-evidential considerations’ with respect to some proposition p are considerations that don't in fact constitute evidence for or against p. But read this way, the arguments will not be sound, because it is possible to respond directly and consciously to considerations that don't in fact constitute evidence for or against p, when one takes those considerations to be evidence for or against p. The only hope for these arguments is to employ a weaker impossibility-claim, which says (roughly) that it's impossible to respond (directly and consciously) to considerations that one takes to be evidentially irrelevant. The question to be investigated in the next three sections is whether weakening the impossibility-claim in this way allows the arguments to go through.

2. Arguments from a motivational constraint on normative reasons

Kelly (Reference Kelly2002) and Shah (Reference Shah2006) offer similar arguments for Evidentialism-Reasons. The argument goes roughly as follows:

In this statement of the argument, I have (following Kelly) formulated the argument in terms of what one believes on the basis of.Footnote ¹¹ One might instead use the language of motivating reasons, or the reasons for which one believes.Footnote ¹² The impossibility-claim would then be that non-evidential considerations cannot serve as motivating reasons for belief – the reasons for which one believes – and the constraint on (normative) reasons would be that some consideration is a normative reason for Ф-ing only if it can serve as a motivating reason for Ф-ing. I take this difference to be largely cosmetic, since I take the notion of a motivating reason for Ф-ing (or a reason for which one Ф's) to be roughly equivalent to the notion of a consideration that one Ф's on the basis of. In any case, any subtle differences between these notions shouldn't, I think, be very important for what follows.

Though the Kelly–Shah argument has been much-discussed,Footnote ¹³ I think the most glaring weakness in the argument has been missed, which is that at least prima facie, the argument equivocates on ‘non-evidential consideration’. As we saw in the last section, Impossibility-Claim is only at all plausible if we take ‘non-evidential considerations’ to mean something like ‘considerations that the agent takes to evidentially irrelevant’.Footnote ¹⁴ Let's make that explicit by precisifying Impossibility-Claim a bit:

But in Evidentialism-Reasons, which is supposed to be the conclusion of the argument, the phrase ‘non-evidential considerations’ is most naturally interpreted as referring to considerations that aren't actually evidence for p. After all, the evidentialist stance is that only (genuine) evidence for p constitutes a reason to believe p, not that only considerations that the agent takes to be evidence for p constitute a reason to believe p. Again, let's make this precise:

So, once both of those claims have been made precise, we're left with the following argument:

This argument is patently invalid.

At first glance, what Impossibility Claim-Precise and Constraint on Reasons actually seem to entail is:

Could the evidentialist be content with having established Subjective Evidentialism-Reasons? I think not; it is both too weak and too strong for the evidentialist's purposes. Let me explain.

First, it is too weak for the evidentialist, because it doesn't establish that some non-evidential consideration can't be a normative reason for belief as long as the agent is taking it to be a reason. So for example, suppose that you're offered $1,000,000 to believe p, and suppose that you (falsely and irrationally) think that the offer of money is evidence that p – perhaps because you wrongly (falsely and irrationally) believe that there's an angel presiding over you, who ensures that you're only ever offered money to believe true propositions. It seems that the argument does not rule out the fact that you've been offered money from counting as a normative reason for you to believe p.

The evidentialist might respond by pointing out that Constraint on Reasons on its own need not rule out all the non-reasons. In order for something to be a normative reason for you to believe p, it has to (i) be something you can believe p on the basis of (i.e., meet the Constraint on Reasons) and (ii) be something that counts in favor of believing p. But for this reply to have any force, the evidentialist has to be presupposing that the fact that you're offered $1,000,000 to believe p doesn't count in favor of believing p (or doesn't count in favor of it in the right way to be a reason to believe p). And that assumption is clearly question-begging in this context. The Kelly–Shah argument was supposed to provide a non-question-begging argument for evidentialism by arguing from a constraint on reasons that has independent motivation and plausibility. If that constraint isn't what's really doing the work in ruling out the offer of money from counting as a reason, then we've given up on the ambitions of the argument.

Second, and just as importantly, Subjective Evidentialism-Reasons is too strong (both for the evidentialist, and on its own intuitive merits). For it seems to rule out some consideration that is evidence for p, but which the agent takes not to be evidence for p, from constituting a normative reason to believe p. But then agents will be let off the hook for failing to respond to their evidence whenever they fail to recognize what it supports. For example, suppose that the Orioles’ terrible losing record last year plus their lack of any free-agent acquisitions over the offseason together provide decisive support for believing that they will perform badly this year. But suppose also that, though I'm aware of the losing record and the lack of free-agent acquisitions, I don't recognize that the evidential force of these considerations, claiming that it's a new season and last year's results are irrelevant. If considerations that I don't take to be evidentially relevant to whether p are ruled out from being normative reasons for me to believe p, then we are forced to say that the Orioles’ losing record and their lack of free-agent acquisitions aren't reasons for me to believe that the Orioles will perform badly this year. But this is surely not right; they are reasons for me to believe that the Orioles will perform badly this year, notwithstanding my failure to recognize that fact. And in particular, the evidentialist should want to say that. If I fail to believe that the Orioles will have a bad year – or, even worse, positively believe that they'll have a good year – I'm failing to respond to my evidence correctly, and thus failing to believe as I ought to; the evidentialist should want to capture that. But she can't if she's committed to restricting our reasons to the considerations that we take to be evidence.

Given this second, “too strong” problem, Subjective Evidentialism-Reasons seems implausible. But it did seem that the above argument from Impossibility Claim-Precise and Constraint on Reasons to Subjective Evidentialism-Reasons was valid. Do we, then, have a reductio of Constraint on Reasons?

Only on one interpretation of Constraint on Reasons, I think. Notice that Constraint on Reasons is not relativized to an agent. One interpretation of Constraint on Reasons would say that some consideration C constitutes a normative reason for A to Ф only if it's possible for A to Ф on the basis of C (holding fixed certain background mental states of A's, such as her basic desires, or what she takes to be evidence for what).Footnote ¹⁶ This would indeed seem to suggest that if I don't take some consideration to be evidence for p, then it can't be a reason for me to believe p, which yields the bad result in the Orioles case and many other cases with the same structure. But it might plausibly be suggested that this is not the right way to understand Constraint on Reasons. Rather, it might be said, this constraint should allow that C can constitute a normative reason for A to Ф so long as someone (with the right psychology, which might include background mental states different to A'sFootnote ¹⁷) could Ф on the basis of C. Hence, the most plausible precisification of Constraint on Reasons is:

Constraint on Reasons-Precise doesn't rule out the Orioles’ losing record from last year, plus their lack of free-agent acquisitions, being a reason for me to believe that they will perform badly this year. It doesn't matter that I don't recognize the force of these considerations; the point is that someone who did would be able to believe that the Orioles will perform badly this year on this basis.

With this clarification made, though, we can see that the situation for the Kelly–Shah argument is even worse than it first appeared. We now have the following two premises:

But what do these two claims jointly entail? Nothing (relevant), it seems. They don't entail Evidentialism-Reasons; they don't even entail Subjective Evidentialism-Reasons. (Thus even the argument for Subjective Evidentialism-Reasons, once its premises have been made plausible and precise, turns out to be invalid after all.) Moreover, they don't seem to rule out many (if any) would-be reasons for belief, since it seems that for any consideration that you take to be evidentially irrelevant to whether p, it's always going to possible to imagine someone who does take it to be evidentially relevant to whether p, and is thus able to believe p on the basis of it.

To make this more vivid, return to the case where you're offered $1,000,000 to believe p. As I noted above, by the lights of Subjective Evidentialism-Reasons, there's nothing to rule this offer of money out from counting as a normative reason for you to believe p if you take the offer of money to be evidence that p. But now we can see that, at least by the lights of the precisified constraint on reasons alone, there's nothing to rule out the offer of money from counting as a normative reason for you to believe p even if you don't take the offer of money to be evidence that p. For it's still true that someone (specifically, someone who did take the fact that they've been offered $1,000,000 to believe p to be evidence for p) could believe p on the basis of the fact that they've been offered $1,000,000. And that's enough for this not to be ruled out by Constraint on Reasons-Precise even as a reason for you to believe p.

Although I seem to have shown that, when we make the premises and conclusion of the Kelly–Shah argument precise, it's not valid, one may have the sense that there's something fishy about the way I'm resisting it. Suppose for the sake of argument that the fact that you've been offered $1,000,000 does constitute a (pragmatic, non-evidential) reason to believe p. I've been insisting that it is possible to believe p on the basis of this fact, but only by mistaking this fact for an evidential reason to believe p. Isn't there something at least a bit odd about the idea that this fact is a pragmatic reason to believe p, but that one can only base one's belief in p on this fact by mistaking it for an evidential reason to believe p?

I agree that there's something that calls out for explanation there, and I'll turn to this point later, in §4. For our purposes in this section, the question is whether there's a way to save something of the form of the Kelly–Shah argument itself in the vicinity here. I can see two strategies for claiming that there is.

So far I've been appealing to the undifferentiated notion of a “consideration”, where the fact that you've been offered $1,000,000 to believe p is the same consideration, whether you're thinking of it as evidence that p, or merely as something that shows believing p to be pragmatically valuable. One might wonder if there's a more fine-grained notion than that of a consideration, that will be sensitive to whether this consideration is serving as (or being thought of as?) evidence for p, or whether it is serving as (or being thought of as?) something that shows believing p to be pragmatically valuable. We'd then need to revise the Constraint on Reasons so that it requires it to be possible to base one's belief on this more fine-grained thing, in order for the corresponding reason to be present. Then, the thought would be that while it is possible to base your belief on the fine-grained entity (whatever it is) that corresponds to the fact that you've been offered $1,000,000 qua (apparent?) evidence that p, it isn't possible to base your belief on the different fine-grained entity (whatever it is) that corresponds to the fact that you've been offered $1,000,000 qua (apparent?) thing-that-shows-believing-p-to-be-pragmatically-valuable. And so the latter cannot constitute a normative reason.

While this is an intriguing strategy, a lot more has to be said to make it clear and to vindicate it. Most obviously, we're owed an account of what the fine-grained entity is, exactly. Additionally, the view seems to be committed to saying that both motivating reasons (or bases) and normative reasons are constituted by this fine-grained entity, rather than by a coarser-grained one. This requires a significant rethinking of orthodoxy, since the standard view (which Kelly and Shah do not seem to depart from in the way they speak) is that reasons are facts, which are clearly too coarse-grained to do the job (the fact – that believing p will get you $1,000,000 – is the same in both cases).Footnote ¹⁸ Other prominent views, such as that reasons are propositions, or that they are mental states (likely, when it comes to epistemic reasons, beliefs in those propositions), seem likewise too coarse-grained to do the job. This rethinking of orthodoxy is a major task to be executed, and it seems ad hoc if it cannot be motivated independently of the need to save evidentialism.Footnote ¹⁹

A different way of pursuing a proposal in the same broad spirit – this is the second strategy mentioned above – is to try to build the fine-grainedness not into our account of what a reason is, but rather into Constraint on Reasons.Footnote ²⁰ So, one might try something like the following:

The thought here is that in order for the fact that you've offered me $1,000,000 to believe p to constitute a genuine, practical reason for me to believe p, it has to be possible for me (or at least, someone) to believe p on the basis of this monetary offer under the guise of a practical consideration (rather than, say, by taking this fact to be evidence for p). That is – assuming for reductio that this consideration does practically count in favor of believing p – I (or someone) have to be able to believe p on the basis of the monetary offer through appreciating the way in which it counts in favor of believing p – viz., practically. Put together with the premise that we can't believe on the basis of practical considerations under the guise of practical considerations, this yields Evidentialism-Reasons.

I will be honest here: I think this is the best strategy for reviving the argument from a motivational constraint, and it is certainly worthy of consideration. That said, once again, this option is so far completely unexplored in the literature. In particular, I know of no-one who has argued for Constraint on Reasons-Revised, as opposed to the (weaker) Constraint on Reasons-Precise.

Here's one challenge for this strategy. Most ordinary agents do not have sophisticated views, when they act or believe on the basis of some consideration, about exactly how this consideration counts in favor of the act or belief in question, or about what categories of reasons there are (viz. moral, prudential, instrumental, aesthetic, epistemic, and so on) and what the boundaries between them are, in a way that allows them to determine what kind of reason the consideration that they are acting or believing under the guise of is, such as to fix what kind of reason they are acting or believing under the guise of. Even philosophers find these questions very difficult to answer, let alone ordinary people. So a lot of the time it is going to be seriously indeterminate under what “guise” someone is acting under when they act on some consideration.

This is most obvious in the case of reasons for action, where it seems frequently indeterminate whether someone is acting on some consideration under the guise of a moral reason or under the guise of some other reason (prudential, instrumental, aesthetic, etc.). But it can also arise in the case of belief as well. Consider (a version of) the sort of case discussed by Stroud (Reference Stroud2006) and others, where friendship impinges upon our doxastic practices. Suppose that Tamara is accused of a crime, and protests her innocence to her friend Smruti. Smruti then reports that she believes that Tamara is innocent “because she's my friend” – not an unusual sort of thing for a person to say. It seems reasonable to take Smruti at her word when she says she believes Tamara because Tamara is her friend. But under what guise is Smruti conceiving of this consideration? She could be taking the fact that Tamara is her friend (someone she knows well, trusts, and so on) to be evidentially indicative of the fact that she wouldn't lie to her; or she could be taking the fact that Tamara is her friend to be some kind of practical (moral? sui generis friendship-based?) reason to believe her. Or – and I suggest this is plausibly true in a wide range of real-life cases – it may simply be indeterminate whether Smruti is believing on the basis of her friendship under the guise of an evidential reason or under the guise of a practical reason. She may not have separated out, in her mind, the two kinds of (putative) ways that her friendship with Tamara might count in favor of believing Tamara – or considered in which way (or both) she takes it to so count in favor.

This makes it hard to definitively say whether Smruti counts as someone who can believe on the basis of her friendship with Tamara under a practical guise, and thus makes it hard to see whether Constraint on Reasons-Revised rules out friendship-based practical reasons for belief or not.Footnote ²¹ Now, the proponent of Constraint on Reasons-Revised might insist that to settle whether it is possible for anyone to genuinely believe (or act) on some consideration C under the guise of a reason of some particular kind K, we should focus only on agents who have the distinction between the K-reasons and the non-K-reasons clearly in view, and see whether they can believe (or act) based on C under the guise of a K-reason. But such people are, in the main, sophisticated, professional philosophers. And it seems rather ad hoc – not to mention rather far from anything that motivated the Constraint on Reasons in the first instance – to insist that what kinds of reasons ordinary people can have ultimately turns only on how people quite unlike them, namely professional philosophers, are motivated.

While I have raised some initial challenges for the two strategies for saving the Kelly-Shah-style argument – the fine-grained view of reasons, and the appeal to different guises – I admit that these challenges are not entirely conclusive, and may ultimately be answerable. So I have to make do with the conditional conclusion that the Kelly–Shah-style argument can't be made to work unless one of these two strategies can be vindicated. It's in any case interesting if it turns out that the only way to preserve the argument is to embrace one of these views. In any case, I think it's clear that Kelly and Shah themselves didn't have either of these views in mind, and did not see the problem that requires them to adopt one of them. They simply weren't attending sufficiently to the difference between considerations that the evidence actually supports and considerations that one takes it to support, and the significance of this difference for their arguments.

3. Arguments from ought implies can

I next want to discuss another argument that is in the neighborhood of, but not identical to, the one we just considered – though the two are often not well-distinguished. I haven't found a clear statement of this argument as distinct from the first, but it seems to me to be “in the air” in discussions of evidentialism and pragmatism.Footnote ²² The argument turns on an “ought implies can” principle. An initial rough statement of it might be as follows.

1. 1. If there were non-evidential reasons for belief, then it would sometimes be the case that you ought to believe p when your evidence doesn't support p.

2. 2. But you cannot believe p when your evidence doesn't support p (impossibility-claim).

3. 3. You ought to Ф only if you can Ф (ought implies can).

   So,

   Evidentialism-Reasons. There are no non-evidential reasons for belief (all reasons for belief are evidential).

But, for reasons that are now familiar, premise (2) is wrong as it stands: what is true at most is that you cannot believe p when you don't take your evidence to support p. With premise (1) also adjusted to preserve the argument's validity, the result is:

• 1*. If there were non-evidential reasons for belief, then it would sometimes be the case that you ought to believe p when you don't take your evidence to support p.

• 2*. But you cannot believe p when you don't take your evidence to support p.

• 3. You ought to Ф only if you can Ф (ought implies can).

  So,

  Evidentialism-Reasons. There are no non-evidential reasons for belief (all reasons for belief are evidential).

The problem with this argument, though, is that it seems to overgeneralize. For even if there aren't non-evidential reasons for belief, it is sometimes the case that you ought to believe p even though you don't take your evidence to support p. This is so by evidentialist lights in cases where your evidence does decisively support p, but you don't recognize this. But given premises (2*) and (3), it seems that it will follow that it isn't true that you ought to believe p in such cases. Thus premises (2*) and (3) seem to threaten evidentialism itself.Footnote ²³ Something in premises (2) and (3) seems to require toning down.

I suggest that the problem is this. (2*) is scope-ambiguous, being able to be interpreted either of the following ways:

I think we should deny that (2*-Narrow) is true, at least on any reading of ‘cannot’ that vindicates (3) (i.e., ought implies can).Footnote ²⁴ Suppose you don't take your evidence to support p. In the relevant sense of ‘can’ that vindicates ‘ought implies can’, you still can believe p – by (reasoned) change of mind about whether the evidence supports p. That's why, in a case where your evidence does support p, but you don't think it does, we can still say that you ought to believe p, without any violation of (a plausible version of) ought implies can. The relevant sense of ‘can’ for ought implies can doesn't treat all your existing background psychological states as unalterably fixed, for this would be too restrictive. (Compare: you don't care about anyone except for yourself. In a weak sense of ‘can’ that treats this not caring as unalterably fixed, this means that you can't act from altruistic motives. But it doesn't follow that it's false that you ought to act from altruistic motives. In the relevant sense, you can start caring about others, and thus start acting out of altruistic motives.)

That leaves (2*-Wide), which is more plausible (though still highly contestable). What it says is that you can't simultaneously (i) believe p and (ii) not take your evidence to support p. Suppose for the sake of argument that this is true. If we interpret (2) as (2*-Wide), we need to make a corresponding adjustment to (1) to preserve the argument's validity, giving us yet a third version of the argument:

• 1*- Wide. If there were non-evidential reasons for belief, then it would sometimes be the case that you ought to (believe p while simultaneously not taking your evidence to support p).

• 2*- Wide. But you cannot (believe p while simultaneously not taking your evidence to support p).

• 3. You ought to Ф only if you can Ф (ought implies can).

  So,

  Evidentialism-Reasons. There are no non-evidential reasons for belief (all reasons for belief are evidential).

But the anti-evidentialist who accepts (2*-Wide) and (3) can, and should, simply deny (1*-Wide). She should hold that when you have decisive non-evidential reasons to believe p, you should not only believe p, but also believe that your evidence supports p. After all, if (2*-Wide) is true, then believing that your evidence supports p will be a necessary means to succeeding in believing p.Footnote ²⁵ Indeed, ever since Pascal, pragmatists about reasons for belief have counseled those who have pragmatic reason to believe things to do so by convincing themselves that there's sufficient evidence for these claims.Footnote ²⁶ But if this is right, it won't be true that, in such cases, you ought to (believe p while simultaneously not taking your evidence to support p). Thus, (1*-Wide) will be false. Thus, ultimately, one way or another, the argument fails.

4. Arguments by inference to the best explanation

A third kind of argument from an impossibility-claim to evidentialism can be found in Adler's book Belief Own Ethics (Adler Reference Adler2002b: Ch. 1). Though Adler doesn't quite explicitly present it this way, I think the argument can be presented as proceeding via inference to the best explanation. The idea is that it is a fact that we can't respond to non-evidential considerations, and that the best explanation of this fact is that we all tacitly recognize and accept the truth of Evidentialism-Ought (which, to recap, says that one ought to believe p only if one has adequate evidence for p).Footnote ²⁷

If this style of argument works, it may also provide support for the claim that Evidentialism-Ought is a constitutive norm for belief. It is not completely clear what it is for a norm on some mental state to be constitutive of that mental state, as opposed to being a genuine, necessary, norm on that mental state that is nevertheless not constitutive of it. But here is an initially appealing thought: for many norms N on some mental state kind K, it doesn't follow merely from N's being a norm on states of kind K that everyone tacitly recognizes and accepts it as a norm on states of kind K. So maybe what distinguishes the constitutive norms on states of kind K are that they are ones that anyone who counts as having a mental state of kind K – or, perhaps, anyone who has the concept of mental state kind K, such that they're capable of thinking of their own mental states as falling under kind KFootnote ²⁸ – must tacitly recognize and accept, and thus be to at least some extent disposed to comply with. If this is the right way of drawing the distinction between constitutive and non-constitutive norms, then anyone who is committed to a claim of the form:

is also committed, given that they think that N is a constitutive norm for S, to:

Given this account of the difference between constitutive and non-constitutive norms, it makes sense for someone who accepts the Adlerian argument to also accept that Evidentialism-Ought is a constitutive norm. For, according to Adler, the best explanation of the fact that we can't respond to non-evidential considerations is that we all (or, at least, those of us who have the concept of belief) tacitly recognize and accept the truth of Evidentialism-Ought. And if that's so, then by this account of constitutivity, Evidentialism-Ought is a constitutive norm: it will be constitutive of being in the state of belief both that this state is subject to Evidentialism-Ought, and that (perhaps just insofar as one conceives of the state as a belief) one is disposed to regulate it in compliance with Evidentialism-Ought.Footnote ³⁰

As with the previous arguments, there is a question about how to interpret the claim that we can't respond to non-evidential considerations, as it appears as a premise in the argument by inference to the best explanation: as referring to considerations that aren't actually evidence, or as referring to considerations that we don't take to be evidence. And as we've already seen in §1, Adler is fairly uncareful about moving between these two sorts of claims. Nevertheless, at least some of the time, he presents the datum to be explained as being the fact that we can't respond to considerations we don't take to be evidence, or that we can't defy our own judgments about what the evidence supports.Footnote ³¹ And as we've seen, this is how we need to interpret the datum if it's to be at all plausible. (Indeed, we may want to weaken it further, to say, for example, that we're merely strongly disposed not to defy our own judgments about what the evidence supports.)

Now, we might hold out some hope that, unlike the arguments of the last two sections, the argument by inference to the best explanation still works well even when this is taken account of. My recognition and acceptance of the truth of evidentialism sounds like a good candidate for explaining why I can't respond to considerations that I don't take to be good evidence, since I will (normally) take responding to them to involve having to violate evidentialism, a norm that I accept and am disposed to comply with. This evidentialist explanation reissues the challenge that I mentioned toward the end of §2: if there are, contrary to this explanation, non-evidential reasons for belief, what can explain why we only respond to them by mistaking them for evidential considerations?

However, once we make clear that what we need to explain is not a disposition to believe what the evidence actually supports, but a disposition to believe what one takes the evidence supports, I think a competing explanation comes into view. As a number of philosophers have suggested, there is a coherence requirement on doxastic attitudes, distinct from evidentialism itself, that requires us to line up our beliefs with our judgments about what the evidence supports:

Inter-level coherence (ILC). Rationality requires that you do not simultaneously:

1. (i) {Believe that a doxastic attitude D is inadequately supported by the evidence, have D}; or

2. (ii) {Believe that a doxastic attitude D is decisively supported by the evidence, fail to have D}Footnote ³²

Failing to distinguish Inter-Level Coherence from Evidentialism-Ought is a mistake very similar to failing to distinguish what one takes one's evidence to support from what is actually supports, since Inter-Level Coherence demands conformity of one's beliefs with what one takes one's evidence to support rather than with what it actually supports.Footnote ³³ Despite their distinctness, though, Evidentialism-Ought and Inter-Level Coherence are at least prima facie consistent, due to the latter's “wide-scope” character. Suppose you falsely judge that your evidence supports believing p. In that case, Evidentialism-Ought requires you not to believe p. But Inter-Level Coherence doesn't categorically demand that you do believe p, since you could also satisfy it by giving up your belief that your evidence supports believing p.Footnote ³⁴

As far as I can tell, one could mimic Adler's explanation of why we can't respond to considerations that we take to be non-evidential in belief-formation using Inter-Level Coherence instead of Evidentialism-Ought, and not much is lost from the explanation. The suggestion would be that we all tacitly recognize and accept the irrationality (or incoherence) of simultaneously holding, for example, that the evidence doesn't support p, while also believing p and thus holding that p is true – and that this is why we have difficulty responding to pragmatic reasons in belief-formation.Footnote ³⁵ At least if one finds the style of explanation in terms of Evidentialism-Ought satisfying, I don't see why one shouldn't find the explanation in term of Inter-Level Coherence equally satisfying (pending some worries to be considered in the next section).

If that's so, the argument for Evidentialism-Ought by inference to the best explanation is unsound. For an argument to a conclusion by inference to the best explanation to work, the conclusion has to be the single best explanation of the relevant phenomena or data. If there are other, equally good, explanations available, then it seems like the argument at best provides an argument for the disjunction of the good explanations. So, given that there is a rival explanation of our inability to respond to pragmatic reasons, involving Inter-Level Coherence, that seems equally good to the one involving Evidentialism-Ought, what we have is at best an argument for the disjunction of Evidentialism-Ought and Inter-Level Coherence.

This also raises the intriguing possibility that it is Inter-Level Coherence, rather than Evidentialism-Ought, that is a constitutive norm for belief. Indeed, if the mark of a constitutive norm on belief is that any believer (or, perhaps, any believer with the concept of belief) must be disposed to comply with it, then it seems that the case for the constitutivity of Inter-Level Coherence is actually better than the case for the constitutivity of Evidentialism-Ought. If we are disposed to believe what we take our evidence to support, this more directly and obviously constitutes a disposition to comply with Inter-Level Coherence than a disposition to comply with Evidentialism-Ought. It's a disposition to do what we think will bring us into compliance with Evidentialism-Ought, but given the relationship between Evidentialism-Ought and Inter-Level Coherence, that is very close to just restating the disposition to comply with Inter-Level Coherence. There seems to be a more direct sense, then, in which we're disposed to comply with Inter-Level Coherence than in which we're disposed to comply with Evidentialism-Ought. This bolsters the case for thinking of the former, rather than the latter, as constitutive.Footnote ³⁶

5. Doubts about an ILC-based explanation, considered

The primary purpose of this paper isn't to develop an explanation of our inability to believe for non-evidential reasons in terms of Inter-Level Coherence; it is rather to put pressure on various arguments for evidentialism from our inability to believe for non-evidential reasons. However, for the response that I gave to the argument by inference to the best explanation in the last section to work, it does need to be the case that the ILC-based explanation is at least as good as the evidentialist explanation (and that the two are independent of one another). This section considers three different doubts that one might have about this.