1 Introduction

On many views of rationality, rationality is at least partly about avoiding incoherence. The basic idea is that there are incoherent combinations of attitudes which it is intuitively irrational for one to have, and for each of the incoherent combinations there is a corresponding requirement of rationality which prohibits it. It has been widely accepted that such structural (or coherence) requirements form a fundamental part of rationality.

Recently, however, Benjamin Kiesewetter (2017) and Errol Lord (2018) have forcefully argued against the idea that rationality primarily concerns coherence. On their views, rationality is fundamentally about responding correctly to reasons you have for individual attitudes, which goes under the name of substantive rationality.Footnote 1 The reason why rationality appears to be about coherence is that an incoherent combination of attitudes has an interesting property: when you have an incoherent combination of attitudes, it is guaranteed that you fail to do what you have decisive reason to do. For example, if you believe that p and believe that not-p, either your reasons decisively count against the former or they decisively count against the latter, and so at least one of your beliefs fails to be responsive to your reasons, which explains your irrationality. Call this the reasons-based (‘RB’, henceforth) account of incoherence.Footnote 2 If the RB-account were right, there would be no structural requirements that are independent of the demands of reasons, and coherence would only be of derivative significance from the standpoint of rationality.

The aim of this paper is to offer a novel argument against the RB-account of structural requirements, and to explain what it is about (in)coherence that makes it resistant to any account of it in terms of responsiveness to reasons. I will argue that a central assumption behind the RB-account is false: the assumption that being guaranteed to fail to respond to one’s reasons implies irrationality. It is possible for you to have, without being irrational, a combination of attitudes that guarantees the failure to respond to your reasons. This means that even if the RB-account succeeds in showing that every case of incoherence involves the guaranteed failure to respond to reasons, it falls short of providing an adequate account of the irrationality of incoherence.

I proceed as follows. Section 1 explicates the RB-account of incoherence. Section 2 argues against the RB-account by contrasting two cases, one involving an intention and the other involving a desire. The problem is that the RB-account predicts that the agents are irrational in both cases, when incoherence (or irrationality) is present only in the former. Section 3 extends the same point to the contrast between belief and high credence. Section 4 forestalls possible responses. Section 5 draws a general lesson from the failure of the RB-account: any adequate account of structural rationality of belief and intention should take into account the distinctive kind of commitments constitutive of each type of attitude.

2 The reason-based account of incoherence

Consider the following, intuitively irrational combinations of attitudes:

Akrasia: It is irrational for you to: believe that you ought to A; but fail to intend to A.

Means-End Incoherence: It is irrational for you to: intend an end E; believe that M is a necessary means to E; but fail to intend M.Footnote 3

A widely held view as to why it is irrational for you to have such a combination of attitudes is that, for each such combination, there is a corresponding structural requirement of rationality that you violate in having it. Such a requirement demands that your attitudes avoid incoherence or that they stand in the right structural relations with each other: what goes wrong when you violate such a requirement is that your attitudes don’t fit together, which is supposed to be distinct from having an individual attitude that is not supported by your reasons. Copious ink has been spilled on the logical form of the structural requirements, but the entire discussion has been premised on the assumption that avoiding incoherence is a fundamental part of rationality.Footnote 4

The RB-account of structural rationality rejects this assumption. On this account, rationality is fundamentally a matter of responding correctly to one’s reasons, and there are no independent requirements calling for coherence between one’s attitudes. The key idea in the RB-account is that the irrationality of having an incoherent combination of attitudes, such as Akrasia and Means-End Incoherence, is to be explained in terms of the failure to respond correctly to reasons, not in terms of the failure to conform to a structural requirement.Footnote 5 In particular, the RB-account explains why having an incoherent combination of attitudes is irrational, on the basis of the following:

The Guarantee Hypothesis: If you have an incoherent combination of attitudes, it is guaranteed that at least one of your attitudes fails to respond correctly to your reasons.

Guarantee-Irrationality Link: If you are in a state of mind that is guaranteed to fail to respond to your reasons, you are irrational. (Kiesewetter, 2017: p. 236)

I shall standardly assume that a reason comes with a weight, or the degree to which it supports a response, and that sometimes a set of reasons is weighty enough to decisively support a response. I will say that your reasons require the response in such a case. If you fail to have a response your reasons require, you fail to respond correctly to your reasons. Sometimes a set of reasons is sufficient (without necessarily being decisive) in strength to support a response. I will say that your reasons permit the response in such a case.

To see how the RB-account covers particular cases, take the explanation of Akrasia, which begins with the following principle about reasons for doxastic attitudes:

Permission-Entails-Requirement (‘PER’): If your reasons permit having a doxastic attitude D towards a proposition p in a situation S, your reasons also require you to have D towards p in S. (cf. Kiesewetter, 2017: p. 185; Lord, 2018: p. 48)

PER is plausible given evidentialism, the view that only evidence for the target proposition counts as a reason for a doxastic attitude. The best available argument for PER invokes the idea that one’s evidence isn’t permissive: thinking that one is permitted by one’s evidence either to have a doxastic attitude (like belief) or not to have it makes the formation of the doxastic attitude objectionably arbitrary from the standpoint of truth.Footnote 6 While not uncontroversial, I shall simply grant PER to defenders of the RB-account, since it is a crucial principle which lets their account of the incoherence involving doxastic attitudes go through.

Assuming PER, there is a straightforward explanation of Akrasia: either your reasons require you to believe that you ought to A or they require you not to believe that you ought to A. If the former, it is plausible that the same reasons also require you to intend to A, and you fail to respond correctly to your reasons in lacking the intention. If the latter, however, your reasons require you not to believe that you ought to A, in which case you fail to respond correctly to your reasons in having the belief. Either way, you fail to do what your reasons require, which is irrational, given Guarantee-Irrationality Link.

The explanation of Means-End Incoherence divides into two kinds of cases. The first is where your reasons either require you to intend E or require you not to intend E. If the former and you also have decisive reason to believe that M is a necessary means, the decisive reason you have for the end-intention and the means-end belief plausibly transmits to intending M and requires you to intend M. If otherwise, either your end-intention or means-end belief fails to respond correctly to reasons. However, there are cases of the second kind, or cases of mere permissibility, in which your reasons neither require you to intend E nor require you not to intend E (Way, 2013, 2018). The problem for the RB-account is that Means-End Incoherence holds even in the cases of mere permissibility, but it is difficult to explain the irrationality in question as the failure to respond correctly to reasons. For in such cases the reasons to be transmitted from E to M are not strong enough to require you to intend M. In response, Lord (2018: Ch.2) argues that even when E itself is merely permissible, once you intend E, the fact that you intend E intensifies the weight of your reasons favoring M, making it the case that you are required to intend M. Kiesewetter’s (2017: Ch.10) alternative is to appeal to the costs of intending an end while lacking the intention to take what you believe to be a necessary means to it, which are generally higher than the costs of either giving up the end-intention or forming the means-intention.

The RB-account has several attractions. First, it offers a unifying account of rationality. Plausibly, there are normative reasons for attitudes like belief and intention, the responsiveness to which is one dimension along which the attitudes can be rationally assessed. If one thinks that there is a further dimension of rationality that distinctively concerns coherence, independently of the demands of reasons, one is committed to thinking that there are at least two distinct kinds of rational demands. By contrast, the RB-account, if adequate, would elegantly subsume structural rationality under substantive rationality.Footnote 7 Second, unlike Kolodny’s view, on which rationality understood as formal coherence as such is a simply a myth, the RB-account explains why it is irrational to be incoherent. Moreover, it does so in a way that makes incoherence a distinctive form of rational failure: having an incoherent combination of attitudes is a matter of being guaranteed to fail to be responsive to reasons, unlike having an attitude that is unsupported by reasons in a particular situation. Finally, it does so without positing wide-scope or narrow-scope structural requirements, each of which faces formidable objections.Footnote 8

Among the two fundamental principles of the RB-account, Guarantee Hypothesis has already been challenged. Jonathan Way (2013) argues, on the basis of mere permissibility cases, that Means-End Incoherence (and having inconsistent intentions) need not involve any failure to respond to reasons. Alex Worsnip (2018a) even argues that sometimes you can only meet the demands of reasons by being incoherent. Such objections have been responded to in detail, however, and I won’t dispute Guarantee Hypothesis here.Footnote 9 I shall assume Guarantee Hypothesis, along with the assumptions needed to make it plausible, and argue instead that Guarantee-Irrationality Link, given these assumptions, is highly implausible. If so, the two principles cannot both be plausible, which casts doubt on the viability of the RB-account.

3 Guaranteed failure without irrationality (I): intention and desire

This section argues that the RB-account of Means-End Incoherence overgeneralizes. Unlike being means-end incoherent, it seems that you can remain coherent when you predominantly desire an end E, believe that M is a necessary means to E, but fail to intend M. However, the RB-account implausibly entails that whenever you have such a combination of attitudes, you are irrational in the same way as means-end incoherent agents.

3.1 Equivalence and the problem for the RB-account

Suppose that Al learns that there will be an amazing concert tonight, to which he strongly wants to go. He considers going, sees that he has no conflicting schedule, and decides to go: he thereby comes to intend to go to the concert. He believes that getting on the coming bus is the only available way to do so but fails to intend to get on the bus. Now, consider another agent, Beth, who also learns about the concert and strongly wants to go there. She also believes that getting on the bus is the only available way to do so. But she doesn’t form the intention to do so: unlike Al, she hasn’t yet finished considering whether to go to the concert.

Intuitively, there is an important contrast between their states of mind. It is extremely difficult, if not impossible, to see how Al’s attitudes could make sense, whereas Beth’s state of mind seems to make much more sense. This intuitive asymmetry between their mental states suggests that, unlike being means-end incoherent, it need not be incoherent for you to desire E, believe that M is a necessary means, and yet not intend M (cf. Lee, 2020; Worsnip, 2018b).

This asymmetry plausibly has to do with the nature of intention. As Michael Bratman (1987) and Alfred Mele (1992, 2003) argue, intention is distinguished from mere desire by the fact that it involves being settled on its object: intending to A rules out further deliberation about whether to A, whereas even predominantly desiring to A is compatible with deliberating further about whether to A. But when you aren’t so settled on A-ing and still deliberate about whether to A, it doesn’t seem incoherent for you not to intend what you believe to be a necessary means to A-ing. Hence, an account of rationality should not predict that not intending what you believe to be a necessary means to an end you desire is always irrational.

But the RB-account predicts this. For the following is true given the assumption that all reasons for attitudes are object-given, which the RB-account cannot plausibly do without:

Equivalence: Your reasons require you (not) to intend to A if and only if your reasons require you (not) to predominantly desire to A.

Let me explain. First, by ‘desire’ I shall refer to what Mele calls action-desires, which are “desires to A, where ‘A’ is a variable for prospective course of action, and if they are satisfiable, they are satisfiable by, and only by, actions the agent performs.” (Mele, 2003: p. 18) This rules out state-desires, or desires that some states of affairs obtain, such as my desire that aliens exist or my desire that the team I root for wins.Footnote 10 Second, I shall understand an action-desire’s predominance in terms of its motivational strength, that is, how strongly you are motivated to bring about its content: you predominantly desire an object o (at t) just in case there is no alternative object at t that you are more strongly motivated to bring about. In what follows, I shall use the expression ‘desire*’ to make my reference to predominant, action-desire explicit. That said, Equivalence doesn’t concern outweighed desires. For example, it is consistent with your having decisive reason not to intend to smoke without having decisive reason not to desire to smoke, when the desire is outweighed.

Here, then, is the argument that Equivalence is true, assuming that your reasons for intention and desire* are entirely object-given. Object-given reasons for intention or desire* are reasons that bear on features of its object: reasons to intend (or desire*) to A that are provided by normatively relevant features of A-ing.Footnote 11 They contrast with state-given reasons for an attitude: reasons provided by facts about having the attitude itself, such as the fact that you stand to gain a huge monetary reward for having the attitude.

Given this distinction, Equivalence is highly plausible. Both an intention to A and a desire* to A have A-ing as their object, and are action-prompting states in the following sense: you intend or desire* to A only if you are disposed to (try to) A, were you to be given an opportunity to A, where this includes having relevant true beliefs about how to A and having the ability to A. This doesn’t hold for other attitudes, such as wishing, hoping, being inclined, to A. Given that both intention and desire* are action-prompting in this sense, it is plausible that you have decisive reason (not) to intend or (not) to desire* to A just in case you have decisive reason (not) to A (or you all-things-considered ought to A).Footnote 12 Hence, when it comes to object-given reasons for intention and desire, you have decisive reason to intend to A and decisive reason to desire* to A under the very same conditions.

Moreover, Equivalence is partly supported by the fact that there is an important sense in which intention constitutively involves predominant motivation to act, assuming plausibly that you have decisive reason to do something only if you also have decisive reason to do what it constitutively involves. Since an intention is essentially an action-prompting state, your intention to A disposes you to intentionally A (or try to A), when given an opportunity to choose between A-ing and other available alternatives. But it is unclear how your intention could so dispose you without incorporating predominant motivation. Suppose, for example, that you are, at t, more strongly motivated to B than to A, which entails that you are disposed to B rather than A, when you are given an opportunity to choose between A-ing and B-ing. Then the idea that you are, at the same time, disposed to A would be conceptually puzzling.Footnote 13

A possible worry about Equivalence is that you might rationally intend do to something without desiring* to do so: you might intend to finish grading while more strongly wanting to go to the club instead. However, this point can be accommodated consistently with the truth of Equivalence, since there are other, equally natural senses of ‘more strongly’ that could be invoked: your desire to go to the club might be stronger than your intention to finish grading in terms of its felt intensity, or the frequency with which the prospects of its satisfaction occupy your attention. More generally, there are several aspects of a desire in terms of which its strength could be measured: the hedonic aspect, which disposes the agent to experience pleasure (or displeasure) as the subjective probability of its satisfaction (or frustration) increases; the attention aspect, which disposes the agent to attend to things they associate with its object (cf. Sinhababu, 2017: Ch.2). The idea that intention incorporates predominant motivation to prompt action is consistent with its not being predominant in these other aspects. In the same way, Equivalence is consistent with the view that you might have decisive reason to intend what you don’t have decisive reason to predominantly desire in these other aspects. All Equivalence (plausibly) denies is that you can be required to intend to A without also being required to be in the kind of motivational state that would prompt A-ing under the appropriate conditions.

Another possible worry is that you might rationally desire* to A without intending to A. Suppose, for example, your friend is drowning in a river but you know that you cannot swim and so you are unable to save him. It might be thought that you lack decisive reason to intend to save him, but you still have decisive reason to desire* to save him: something would be amiss if you did not most strongly want to save him. But again, the point can be accommodated consistently with the truth of Equivalence. First, it might be that you should have a desire towards saving your friend that is strong in the hedonic or attentional aspects: you should feel extremely painful about the foreseen failure to save him; or you should keep dwelling on the possibilities of saving him. Second, it might be that you should have a predominant state-desire that you save him, even when you know that you cannot save him. What Equivalence denies is only that you should have a motivationally predominant action-desire to save him in such a case. This is again plausible: given that your attempt to save him would be futile and would only risk your own life, you lack decisive reason to be in the kind of motivational state that would prompt you to (try to) save him.

So much for the initial defense of Equivalence. Given Equivalence, however, the RB-account has the implausible consequence. To see this, consider Al and Beth again. Suppose first that they are in a non-permissive case: either they have decisive reason to both intend to go and have the means-end belief, or they have decisive reason not to do both. The RB-account entails that Al is irrational, as in the above (Sect. 1). Given Equivalence, however, it follows that Beth is irrational in just the same way: either Beth has decisive reason to desire* to go or decisive reason not to desire* to go; if the former and she also has decisive reason to have the means-end belief, by Equivalence*, she also has decisive reason to intend the means, which she fails to do; if otherwise, either her end-desire* or means-end belief fails to respond correctly to her reasons.

What if they are in a permissive case? At this point, one might doubt that desire* can do the work intention does in Lord’s and Kiesewetter’s accounts. The idea central to Lord’s account is that your intention for the (merely permissible) end intensifies the weight of reason you have for intending the means, making it the case that you have decisive reason to do so. And one might think that only intention can play such a role in virtue of its distinctive functional role.

Lord’s proposal faces a serious problem. If your end-intention only makes it the case that you are required to intend the means without being required to intend the end itself, so that you can still permissibly give up your end, then you can be permitted to give up the end while being required to intend the means, which seems incoherent (Kiesewetter, 2017: p. 274; Brunero, 2020: pp. 107–108). On the other hand, if your end-intention also makes it the case that you are required to intend the end, then it looks like an implausible form of bootstrapping.Footnote 14 Moreover, even if Lord’s view works, it is unclear why desires* should not be allowed to play the same, intensifying role. Suppose that each of us is permitted to E and also permitted not to E, but I desire* to E while you are simply indifferent to E-ing. It seems that I am, but you are not, under some normative pressure to take steps towards E-ing, including seeking necessary means to E-ing, and Lord seems to have no principled reason not to explain this difference in the same way.

Kiesewetter’s account appeals to the risks of engaging in pointless activities, which a means-end incoherent combination of attitudes is very likely to lead to. But if such an economic consideration is what accounts for the irrationality in permissive cases, it is unclear why the same point cannot apply to desire*. For desiring* an end, just like intending it, involves various motivational, affective, attentional dispositions, which makes it a significant mental investment, and you are likely to engage in pointless activities if you do not intend what you believe to be a means implied by what you desire*. Thus, whichever way one chooses to account for Means-End Incoherence, it follows from Guarantee-Irrationality Link that Beth is irrational just like Al, which, as we have seen, is problematic.

3.2 Response (1): state-given reasons to the rescue?

One way to avoid the problematic consequence is to deny that all reasons for attitudes are object-given. But this isn’t a viable option for defenders of the RB-account.Footnote 15 For admitting state-given reasons of all kinds into the picture runs the risk of undermining the RB-account itself. Suppose that Al has decisive object-given reason to intend to go to the concert and to believe that getting on the bus is a necessary means but is given an enormously weighty state-given reason not to intend to get on the bus: an evil demon threatens to destroy the world if he intends to do so. If the state-given reason were to count, correctly responding to all of his reasons would require being means-end incoherent, falsifying Guarantee Hypothesis.

Still, one might think that the RB-account is consistent with the existence of a restricted class of state-given reasons. Mark Schroeder (2012) has argued that there are state-given reasons against intention which turn on the distinctive role of intention: settling on a course of action and precluding further deliberation. For example, suppose that you are trying to decide whether to drive to LA tomorrow, which would be worth the traffic if, and only if, your brother happens to be in LA tomorrow and you could meet him there. While you find it likely that he won’t be there, you know that your brother will give you a call later today to let you know whether he will be in LA. In this case, the fact that your brother will give you a call later seems to be a reason not to make up your mind now about whether to drive to LA, and hence a reason not to intend to drive to LA, which is not an object-given not to intend to do so. And since this reason depends on the distinctive role of intention, it need not be a reason not to desire* to drive to LA. On this basis, one might argue that Equivalence is false: you might have decisive reason not to intend to A without having decisive reason not to desire* to A.

One worry about this move is that it is unclear whether such a reason really is a state-given reason not to intend, as opposed to an object-given reason to forbear engaging in practical reasoning (Shah & Silverstein, 2013). A more serious worry is that even Schroeder-style state-given reasons undermine the RB-account itself. To see why, notice first that if the RB-account embraces such reasons against intention, then it cannot plausibly avoid recognizing the same kind of reasons against belief. For one thing, Schroeder himself (2012: p. 471) argues that a similar consideration, such as the fact that further evidence bearing on p is forthcoming, is a state-given reason not to believe p. For another, the RB-account is committed to the view that if you have decisive reason to believe that you ought to A, you have decisive reason to intend to A. However, it is unclear how this could be true if there were state-given reasons against intention but not against belief. For it would then be entirely possible for you have decisive evidence, or decisive object-given reason to believe, that you ought to A and yet lack decisive reason to intend to A, due to a weighty state-given reason not to intend to A.

Now, the problem is that if there are state-given reasons for belief, then it is extremely plausible that there are situations in which you are permitted to believe a proposition but also permitted not to believe it. To see why, consider a case where you expect further evidence bearing on a proposition P to be forthcoming and so have a state-given reason not to believe P, but it is not so weighty because little hangs on whether P is true. Suppose that you have accepted a bet where you will win $1 if you are right about whether P is true and lose $1 if you are wrong. You happen to be sitting next to a friend who is an expert on the subject, and she tells you that it is true, so your evidence already makes it very likely that P is true. But your friend is not completely certain, and has nothing better to do, and so offers to double-check whether she is right and is now gone for ten minutes to do so. In this sort of case, you have a Schroeder-style state-given reason not (yet) to believe P, which plausibly permits you to wait until your friend comes back. But intuitively, this reason is not weighty enough to require you to wait, given the low costs of being wrong about P: given that you already have strong evidence, it makes sense for you to go ahead and believe P, and do something other than waiting for your friend. But if you are both permitted to believe P and permitted not to believe P, PER is false, and the RB-account cannot even get off the ground. Thus, even Schroeder-style state-given reasons are unlikely to help the RB-account.

3.3 Response (2): further reason for intention

Alternatively, defenders of the RB-account might attack Equivalence more directly, on the basis of the fact that intention is, in some sense, a stronger attitude than desire*: intention is a settling attitude, which disposes one not to deliberate further. They might argue that to have decisive reason to intend to A, you must have further object-given reason that is absent when you merely have decisive reason to desire* to A. Call this view Further Reason.

One version of Further Reason concerns practical reasons for intention and desire. It states that even if there are only object-given reasons for desire* and intention, the set of object-given reasons which is barely sufficient for your having decisive reason to desire* an object is never weighty enough to make it the case that you have decisive reason to intend the same object. This is extremely odd, however. To say that you have decisive object-given reason to desire* something is, given the definition of predominant desire, to say that you have decisive reason to be more strongly motivated to bring it about than to bring about any other alternatives open to you, which is plausibly the case only if you ought to bring it about. But if this is true, it is unclear what further object-given reason is required to make it the case that you have decisive reason to intend it. To illustrate, suppose that you are given a choice between two vegan sandwiches, Avocado and Black Bean, and all that matters to you in this situation is how they taste. Suppose the fact that Avocado is twice as tasty as Black Bean is barely sufficient to give you decisive reason to desire* to have it. Further Reason entails that Avocado has to be even more tasty than Black Bean for there to be decisive reason for you to intend to have it, which is implausible.

The idea might be that you have decisive object-given reason to intend (as opposed to desire*) to A only if your object-given reasons to A also require you not to deliberate further about whether to A. But this is also implausible. Suppose that health-related considerations give me decisive reason to quit smoking, but I enjoy deliberating about whether to quit, and I have plenty of time to do so, and deliberating about it would not tempt me to smoke again. Plausibly, I have sufficient reason to deliberate even after deciding to quit. This view implies that I cannot have decisive reason to intend to quit in this situation, which is problematic.

Alternatively, one might turn to a version of Further Reason that encompasses theoretical as well as practical reasons. The basic idea would be that unlike desire*, intention is subject to some doxastic constraint. According to cognitivism about intention, intending to A is, or entails, believing that one will A (cf. Harman, 1976; Marušić & Schwenkler, 2018; Setiya, 2007; Velleman, 1989). According to a weaker version of cognitivism, intending to A only entails believing that it is possible for one to A (Wallace, 2001). Even opponents of cognitivism recognize that there are negative doxastic constraints on intending, such that one intends to A only if one does not believe that one certainly will not A, or that it is impossible for one to A (Brunero, 2014; Roughley, 2016). If any of these doxastic constraints is distinctive to intention, reasons for intention might comprise reasons for belief as well as practical reasons, and so reasons for intention and reasons for desire* might systematically differ.

One might, on the basis of strong cognitivism, argue that you have decisive reason to intend to A only if you have decisive reason to believe that you will A, since intending to A entails believing that one will A. But this is implausible, given evidentialism about reasons for belief. Suppose that you have been a heavy smoker for ten years. Today, you are warned by an extremely reliable doctor that you will get a cancer if you don’t quit right away. Given your available evidence, it is 60% probable that you will succeed, so you lack decisive reason to believe that you will quit. The suggestion at hand implies that the doctor’s warning cannot give you decisive reason to intend to quit. This is counterintuitive: it is intuitively irrational for you to fail to intend to quit, and the fact that you lack decisive evidence that you will succeed doesn’t get you off the hook. The point is not that strong cognitivism is false, but that it doesn’t provide an adequate basis for resisting Equivalence. It might be plausible that once you form the intention to A, you believe that you will A. Even so, it is implausible that the formation of an intention rationally ought to be responsive to evidence just like forming an ordinary belief about the future. Indeed, strong cognitivists themselves have argued that intention is a kind of self-fulfilling belief which is insensitive to evidence one has prior to forming it (Velleman, 1989), or a belief formed in response to non-epistemic, practical reasons. (Marušić & Schwenkler, 2018).

One might turn to some weaker doxastic constraint and argue that having decisive reason to intend to A requires having decisive reason not to believe that you certainly won’t A (or that you can’t A). But this is indeed too weak to serve as a basis for rejecting Equivalence. For having decisive reason to desire* to A also seems to require having decisive reason not to believe that you certainly won’t A. For one thing, it is plausible that, when it comes to action-desires, you desire to A only if the possibility that you A is doxastically open, as Mele (2003) argues. If so, desiring to A precludes being certain that you won’t A or believing that you cannot A, just as intending does. For another, it is extremely difficult to imagine a concrete case in which you still have decisive reason to be predominantly motivated to A, even when you have decisive evidence that you certainly won’t A or that you cannot A. As we have seen (2.1), it seems that the very facts that give you reason to be certain that you will fail to A always disable your reasons for A-ing itself, so that the balance of your reasons doesn’t decisively support desiring* to A. I conclude, then, that there is no plausible way, within the framework of the RB-account, to resist Equivalence and avoid the implausible consequence about Al and Beth.

This is not to deny that the functional difference between desire* and intention should be reflected in their normative profiles. What the above arguments show is that the difference need not consist in the difference between (object-given) reasons for them. As we have seen, it is possible for there to be Schroeder-style state-given reasons for intention. Moreover, there can be diachronic requirements of rationality which tell you not to reconsider (or give up) your intention unless new, relevant information comes in (Bratman, 1987; Broome, 2013). Such requirements would be explained by intention’s role of resisting further deliberation and would be distinctive to intention but would not be explained in terms of your object-given reasons for intention.

4 Guaranteed failure without irrationality (II): belief and credence

This section extends this overgeneralization problem, considering the RB-account of Akrasia. It need not be incoherent for you to have a high credence (which is arbitrarily close to but lower than one) that you ought to A but not intend to A. Assuming a highly natural view of the rationality of belief and credence, however, the RB-account implausibly entails that whenever you have such a combination of attitudes, you are irrational in the same way as akratic agents.

4.1 Equivalence* and the problem for the RB-account

Let us again begin with a relevant contrast case. Carla learns that there will be an amazing concert tonight. She deliberates about whether to go, considers pros and cons, and comes to fully believe that she ought to go to the concert. But she doesn’t intend to do so: she is akratic. By contrast, Dan has, given how amazing the concert is expected to be, a 0.95 credence that he ought to go to the concert, but doesn’t yet believe it: he is still considering evidence as to whether he should go and doesn’t yet intend to go. Despite his high credence, the lack of the intention to go doesn’t seem akratic or incoherent, so long as he is so considering.

Here again we find an intuitive difference between the rationality of the agents in this pair of cases. While it is difficult, if not impossible, to see how Carla's attitudes could make sense, Dan's mental state seems to make much more sense.Footnote 16 This intuitive asymmetry between believing and having a high credence suggests that, unlike being akratic, it need not be incoherent for you to have a high credence (which falls short of being certain) that you ought to A without intending to A.

Moreover, this asymmetry plausibly has to do with the nature of belief. Like intention, belief is plausibly a settling attitude: if you believe a proposition p, then you are settled on p’s truth: you treat the question of whether it is true as closed; take the truth of p for granted in further reasoning.Footnote 17 (cf. Wedgwood, 2012) It is plausibly because believing p involves being so settled that your belief in p counts as incorrect if p turns out to be false. This doesn’t apply to high credence: even when you have a very high credence in p, you might still leave the question of its truth open and refrain from using it in your reasoning. Similarly, your high credence in p doesn’t count as incorrect even if p turns out false. As Fantl and McGrath (2009: p. 141) observe, you might intelligibly reply, in such a case, as follows: ‘Look, I took no stand on whether p is true or not; I just assigned it a high probability; I assigned its negation a probability too’. And the reason why Dan is not incoherent seems to be that, unlike Carla, he is not settled on the question of whether he ought to go to the concert, despite his high credence that he ought to do so. If so, an account of rationality should not predict that the combination of attitudes held by Dan is irrational in the way that Carla’s attitudes are.

This is exactly what the RB-account predicts, however, if it is coupled with the following:

Equivalence*: Your reasons require you (not) to believe p in a context c if and only if they require you (not) to have a high credence in p in c.Footnote 18

By a high credence I mean a credence above a certain threshold T, where the value of T can either remain fixed across context or vary with context or stakes (though it plausibly has to be over 0.5). That said, Equivalence* follows from a natural idea about how confident you rationally ought to be in a proposition: the degree of credence you have in a proposition rationally ought to be proportional to the strength with which your evidence supports it. And an equally natural idea about the strength of a body of evidence is that it is a matter of how probable p is, given your evidence. If so, having sufficient evidence to believe p in a given context is a matter of p’s being sufficiently probable given the evidence you have, that is, the evidential probability of p’s being above some threshold value T. This means that whenever it is rational for you to have a credence above T in p in a context, it is rational for you to believe p in the same context. Moreover, if you have sufficient evidence to believe p, then p is sufficiently probable given your evidence, and so you have sufficient evidence to have a credence as high as the evidential probability of p. Assuming PER, Equivalence* follows.

If Equivalence* is true, however, the RB-account entails that having a high credence that you ought to A but lacking an intention to A is irrational, in the same way as being akratic is. Either your reasons require you to believe that you ought to A or they require you not to do so (PER). From this and Equivalence* it follows that either your reasons require you to have some credence above T that you ought to A, or they require you not to have such a credence. If the former, your reasons require you to believe that you ought to A, given Equivalence*, which, by the lights of the RB-account, entails that your reasons require you to intend to A, which you fail to do. If the latter, your reasons require you not to have a credence as high as T that you ought to A, which you have. Hence, you are guaranteed to fail to respond to reasons. This means that Dan’s state of mind turns out to be irrational in just the same way as Carla’s, assuming that the evidential probability that he ought to go to the concert is 0.95 and that it passes the threshold for rational belief in his context.

4.2 Response: denying equivalence*

An obvious option here is to deny Equivalence*. While it spells out a highly intuitive idea about the rationality of belief and evidential support and is indeed popular among epistemologists, it also faces standard objections. The preface paradox (Makinson, 1965) famously challenges the left-to-right direction of Equivalence*. If you are rationally required to believe each claim in the body of the book that you have written, you are rationally required to believe the conjunction of every individual claim, assuming that rational belief is closed under conjunction. But the conjunction is extremely improbable, which makes it rational for you to assign an extremely low credence (which can get arbitrarily close to zero as the number of the claims increases) to it. If so, you are required to believe a proposition without being required to have a high credence in it.

Rejecting Equivalence* in view of the preface paradox has two problems, however. First, it would simply beg the question against those who reject the assumption of conjunctive closure, rather than Equivalence* (e.g. Christensen, 2004; Foley, 1993). Second, and more importantly, it would go against the spirit of the reasons-responsiveness view of rationality. Consider, for example, your reasons for belief in the preface case. Your evidence clearly requires you not to believe the conjunction of all of the claims in the book, but at the same time requires you to believe each individual claim of the book. By contrast, the principle of conjunctive closure is arguably a structural requirement on belief which, at least on its face, isn’t about responding to reasons. By the lights of the RB-account, it is conjunctive closure rather than Equivalence* that has to go in a case where they come into conflict.Footnote 19

Let us then turn to the right-to-left direction of Equivalence*, which confronts the lottery paradox (Kyburg, 1961): in a fair lottery with one million tickets and exactly one winner, you are rationally required to have, for each ticket, an extremely high credence that it will lose, but it might seem that you rationally ought not believe that it will lose. One version of the paradox obviously relies on the assumption of conjunctive closure, which, as we have seen, is problematic in this context. But the lottery problem can be seen as an instance of a more general problem, the problem of merely probabilistic or statistical evidence. Lara Buchak (2014) argues, for example, there are cases where you cannot rationally believe that someone stole your iPhone but you can rationally have a high credence on purely statistical grounds.

However, even if naked statistical evidence poses a genuine threat to Equivalence* unrestricted in its scope, this doesn’t give a lifeline to the RB-account. For some restricted version of Equivalence*, which applies only to cases where your evidence includes both statistical and non-statistical evidence, is still plausible. If such a restricted version of Equivalence* holds, the RB-account still overgeneralizes. For it is plausible that you can neither rationally believe nor rationally have a high credence in a normative proposition to the effect that you ought (all-things-considered) to A on the basis of merely statistical evidence. For example, the mere fact that 95% of an audience enjoyed a concert, which might count as purely statistical evidence that you ought to go to the concert, rationally licenses neither believing nor having a high credence that you ought to go to the concert tonight: it seems that you need further, non-statistical evidence bearing on the relevant normative facts (e.g. the fact that it is enjoyable for a general audience is a reason for you to attend it) and other normatively relevant features of the situation (e.g. you have no conflicting commitments). If this is right, Equivalence* still applies to Carla and Dan, and both Carla and Dan turn out to have an attitude that is guaranteed to fail to respond correctly to their reasons, which, together with Guarantee-Irrationality Link, entails that they are irrational.

Finally, let me consider a way in which Lord himself might try to resist Equivalence*. Lord (2018: pp. 58–59) denies that having sufficient reasons for belief just depends on how probable the content is given the evidence, on the grounds that having sufficient reasons to believe that p also depends on your reasons to withhold belief about whether p, which, on his view, is a matter of having what Friedman (2013, 2019) calls an interrogative attitude towards whether p, such as wondering or being curious about whether p. Crucially, on Lord’s view, reasons to withhold about whether p comprise not just first-order evidence as to whether p, but also higher-order evidence (Lord, 2018: p. 59), and even non-evidential pragmatic factors, such as high stakes (Lord, 2020). But Lord might argue that having sufficient reasons for high credence depends only on the probability of the content given the evidence, and so need not compete with reasons to withhold in the way reasons for belief do. If so, Equivalence* is false.

There are two problems with this move. First, it seems to trade on intuitions about the incoherence between belief and interrogative attitudes. One might naturally ask why reasons for the former need to compete with reasons for the latter: why is it that, when one has strong evidence that p as well as strong practical reason to wonder whether p, one cannot rationally both believe that p and wonder whether p, especially if one can rationally have high credence and yet wonder? Any answer to this question is likely to appeal to the idea that to believe that p and wonder whether p is to be in a conflicted state of mind (cf. Friedman, 2019), unlike having a high credence that p and wondering whether p. But it seems that to appeal to such a conflict between attitudes just is to appeal to the notion of structural irrationality, which the RB-account should do away with.

Second, as I have argued in 2.2, the existence of non-evidential, state-given reasons for belief threatens PER. For example, if stakes are allowed to serve as reasons to withhold belief (or to have an interrogative attitude), then we can easily expect there to be cases in which (i) you have evidence that is strong enough to permit believing that p under normal circumstances and (ii) the stakes are fairly low so that the costs of being wrong about whether p is offset by the costs of inquiring further into whether p. It is highly plausible that, in such a case, you have sufficient reasons to have an interrogative attitude towards whether p (and not believe that p) and also sufficient reasons not to do so (and believe that p). If so, PER is false, and the RB-account fails.

To sum up, to reject Equivalence* is already to incur a significant dialectical burden, given its prima facie plausibility and wide acceptance. Moreover, some prominent objections to Equivalence* either sit ill with the RB-account itself (in the sense of presupposing the existence of structural requirements) or fail to help the RB-account when it comes to explaining Akrasia. This shows that the RB-account of Akrasia is either implausible or incomplete at best.Footnote 20

5 Still irrational?

The guiding assumption of my argument has been that Beth and Dan are not irrational, unlike their structurally irrational counterparts Al and Carla. But one might, in defense of the RB-account, challenge this very assumption. Each of the agents, after all, has a combination of attitudes that is guaranteed to fail to respond correctly to reasons and, in this regard, fail to be fully rational. But to say that the agents aren’t fully rational just is to say that the agents are irrational. Thus, one might think that the RB-account doesn’t make any bad predictions.

One problem with this move is that there is a distinction that has been drawn between irrationality and being (merely) less than fully rational: the former refers to a severe form of rational failure that is akin to being senseless, stupid, or crazy, whereas the latter need not.Footnote 21 It seems that not all cases of failure to respond to your reasons constitute irrationality. Suppose the evidence you have requires being certain in a proposition, but your evidence is difficult to interpret and grasping its import requires some sophisticated inference, so you end up with a 0.95 credence in it. It doesn’t seem irrational for you to do so, though it isn’t fully rational. Nor does a guaranteed failure to respond correctly to reasons seem to be sufficient for irrationality. Suppose that the proposition in question is a fairly complex mathematical truth, which is necessarily true and knowable a priori, but only through some sophisticated reasoning. You would be guaranteed to fail to be fully rational in failing to assign credence 1 to it, but arguably without irrationality. On this basis, one might argue that while none of the above agents are fully rational, only Al and Carla can properly be called irrational.

Another problem is that this defense of the RB-account is incomplete at best, even when being guaranteed to fail to respond to reasons is simply stipulated to constitute irrationality. For the fact remains that there is an intuitive rational difference between Al and Carla, on the one hand, and Beth and Dan, on the other. This cries out for an explanation. A dualist about rationality, who recognizes both structural requirements and demands of substantive rationality has an easy explanation: the agents in the first group violate both relevant structural requirements and the requirements of reasons, whereas the agents in the second group violate only the latter. On the RB-account, however, all of the agents are irrational in failing to violate the requirement of reasons and there seems to be nothing distinctively bad about the first group.

In response, defenders of the RB-account might try to explain the asymmetric judgments about irrationality by appeal to the idea that (ir)rationality is a gradable property, like tallness: to judge that someone is (ir)rational simpliciter is to judge that the agent is (ir)rational to some contextually relevant degree. On this basis, it might be argued that one’s rationality is proportional to how well one responds to one’s reasons and if so, the difference in the irrationality of the agents in the above pairs could be explained as follows: the agent who intends (rather than desires*) and the agents who believe (rather than just having a high credence) are irrational to the contextually relevant degree, whereas their less irrational counterparts are not.

The trouble is that this line of response is unavailable to defenders of the RB-account. If Al and Carla are more irrational than Beth and Dan, respectively, there has to be a further rational defect the members of the first group have that the members of the second group lack.Footnote 22 But it is unclear what this further rational defect could consist in, since the RB-account does away with fundamentally structural requirements. The agents in each pair make (if at all) the same rational error: each of them has a combination of attitudes that is guaranteed to fail to respond to his or her reasons. And the RB-account is incapable of offering a further story. For example, the only difference between Al and Beth, recall, is that Al intends (and also desires*) to go to the concert, whereas Beth only desires* to do so. But this cannot explain why Al is more irrational: if they are in a case where their reasons require them to intend to go, for example, it must be Beth who is more irrational, since she fails to respond to the decisive reason she has to intend to go and Al doesn’t! Moreover, given Equivalence, the RB-account cannot plausibly posit a further reason that Al fails to respond to that Beth doesn’t fail to respond to. The same goes, mutatis mutandis, for the other pair of agents.

6 (In)coherence and commitments

I conclude that the RB-account has no satisfactory response to my argument. In this final section, I draw a more general lesson from the failure of RB-account to subsume structural rationality under substantive rationality: any adequate account of structural requirements on belief and intention has to recognize the distinctive kinds of commitments constitutive of belief and intention that aren’t reducible to responsiveness to reasons. To identify such constitutive commitments is to articulate the suggestive idea that both belief and intention are settling attitudes, which has been briefly touched upon in Sects. 2 and 3.

Take, for example, Bratman’s influential account of intention, according to which intention is distinguished from mere desire by the fact that it is commitment-laden. (Bratman, 1987: Ch. 7) First, intention constitutively involves volitional commitment, having to do with its role as an action-controller: if you intend to A, you are committed to A-ing when the time comes. Second, intention also involves reasoning-centeredness, having to do with its role as an input to further practical reasoning: if you intend to A, then you are committed to forming further intentions directed at A-ing and constraining the formation other intentions that are incompatible with A-ing. On Bratman’s view, such commitments have both descriptive and normative dimensions. On the descriptive side, people who intend to A are disposed to A when they think that the time has come; and they are also disposed to treat A-ing as a fixed point in practical reasoning. On the normative side, the commitments in question explain why intentions are subject to the relevant structural requirements: it is precisely because intention is a commitment-laden attitude that your intentions, in contrast with your desires, are required to be consistent relative to your beliefs and that you are required to be means-end coherent. Satisfying these requirements is a matter of fulfilling the commitments involved in intention. By contrast, desires aren’t subject to the same requirements in lacking such commitments, which in turn explains why Al violates a further requirement Beth isn’t subject to.

The same idea can be naturally extended to cover the structural requirements on belief. For belief is plausibly a commitment-laden attitude, in constitutively involving commitments that are analogous to those involved in intention. If you believe P, for example, you are committed to relying on the truth of P in your action, or to acting as if P is true. Moreover, your belief in P will function as an input to further reasoning: you are committed to inferring what follows from the truth of P and refraining from believing what is logically incompatible with it. Such commitments to the truth of the target proposition, which might be lacking even in a high credence that is arbitrarily close to one, is plausibly what explains why someone who believes P is subject to structural requirements that someone who merely has a high credence is not, which in turn explains why Carla violates a further requirement that Dan isn’t subject to.

This obviously falls short of offering a complete account of structural rationality, which has to be left for another occasion. There are unsettled attitudes, like credences, which arguably lack the kind of commitments involved in belief but are still subject to the constraints of structural rationality. The nature of rational constraints on such attitudes would not plausibly be explained in terms of commitments. Two points remain, however: (1) that the exact kind of requirements or constraints they are under plausibly depends on the kind of states they are; (2) that when it comes to settled attitudes like belief and intention, the constitutive commitments play an ineliminable role in determining the structural requirements they are under. The failure of the RB-account shows that this distinctively attitudinal dimension of structural rationality cannot be understood in terms of substantive rationality. What explains the irrationality of means-end incoherence, for example, are the facts about your commitment to the end, not the facts about your reasons for pursuing the end. If so, (in)coherence must be taken more seriously than what the RB-account makes it out to be.