3.1. Luminosity failures and coherence-restoring states

We have imperfect access to our own mental states. Assume that Ace believes a proposition p, and that he believes not-p. That might look like a blatant kind of irrationality. But let me fill in the details of the case. While Ace believes that he believes p, he doesn't believe that he believes not-p. In fact, he believes that he doesn't believe not-p (‘Why would I? That would be inconsistent with my other beliefs!’ – he thinks). So, even if Ace does his best to survey the contents of his beliefs, he can find no logically inconsistent pairs: for all he believes and knows, his beliefs are perfectly consistent.

Along with many other epistemologists, I am persuaded that for just about any kind of mental state, it is possible to be in a situation in which one is in no position to know that one is in that state: few states are perfectly luminous.Footnote ¹⁶ Assume that this is true of Ace. It's not just that he hasn't bothered to inspect the contents of his mind, but that he isn't even in a position to know that he believes not-p. Ace violates a coherence requirement prohibiting pairwise inconsistent beliefs. However, things don't look that bad from his perspective: while his beliefs are inconsistent, he cannot access the incoherence within his mind. Ace is certainly not irrational in some blatant manner, and he looks very different from the paradigmatically incoherent subjects the literature is focused on. Proponents of coherence requirements persuaded by such cases will have to weaken their preferred principles.Footnote ¹⁷

It is not surprising that prominent attempts to delineate the relevant notion of coherence make luminosity assumptions. It might be tempting to think that certain combinations of states – those that are incoherent in the relevant way – pop out in a way that is difficult to miss. According to this thought, weeding out incoherence should be a simple exercise in mental house cleaning, simpler than being rational in some more demanding sense that requires doing things like taking into account one's evidence in appropriate ways. Indeed, Kolodny (Reference Kolodny2007a: 242) argues that incoherent subjects are always in a position to know that they are incoherent in a specific way: for instance, if a subject both believes a proposition p and believes not-p, this fact is “available” to her, and she is therefore in a position to know that she either believes p or believes not-p without sufficient reason. This is what, Kolodny thinks, distinguishes coherent but mistaken agents from incoherent ones: coherent agents are not always in a position to know they are mistaken. But if we are sometimes in no position to know what we believe or intend, then it is simply false that incoherent subjects are always in a position to know such things.

Various attempts to bring out just why specific (putative) forms of incoherence are bad make implicit or explicit assumptions about access. For instance, Worsnip (Reference Worsnip2018) motivates a requirement prohibiting epistemic akrasia on the grounds that it is difficult for epistemically akratic subjects to make sense of themselves. But in pressing this point, he considers a case in which a subject recognizes that she is epistemically akratic, which of course requires her to have access to her own mental states – it requires her to know, or to at least truly believe, that she both believes p, and believes that she lacks adequate evidence to believe p. Being epistemically akratic, Worsnip argues, at least involves taking oneself to be believing against the evidence. But if I have no access to the fact that I believe p, I don't seem to be taking myself to hold a belief that goes against my evidence.

A very common strategy in the literature on epistemic akrasia has been to point to the Moore-paradoxical nature of assertions like “p, but p is highly unlikely on my evidence”.Footnote ¹⁸ But consider a subject who believes a proposition p, and believes that she doesn't believe it – in fact, she may have every reason to think that she doesn't believe p, and be in no position to know that she believes it. It would be very strange for such a subject to assert p. It is very difficult to think of cases in which we assert a proposition p, but are in no position to know that we asserted p. This is an important difference between believing p and asserting p: our beliefs are often hidden in ways that our assertions are not. This observation casts doubt on the strategy of arguing that certain mental states are defective because it would be bizarre to make the corresponding assertions.

Luminosity failures make putatively incoherent states look much better. Perhaps even more interesting are examples in which subjects hold (presumably) false beliefs about rationality, or about normative matters more generally. Di believes both a proposition p, and believes its negation not-p. However, Di is a dialetheist: she thinks that some contradictions are true, and that p and not-p is a true contradiction. Perhaps, for instance, she thinks that the proposition expressed by the Liar sentence ‘This sentence is false’ is both true and false. We can assume this to be a belief that Di has reached as a result of careful philosophical investigation. With this background story in place, Di's beliefs look less irrational, irrespective of whether her belief in dialetheism is true: she holds beliefs that, by the lights of her own responsibly formed belief, are rational. It is notable that John Broome, for instance, responds to such cases by weakening the condition against inconsistent beliefs:

The thought is that while it is incoherent to merely (1) believe p and (2) believe not-p, it is not incoherent to (1) believe p, (2) believe not-p and (3) believe that in one's current situation there is nothing irrational about believing p and believing not-p. According to this line of thought, the third meta-level belief is what I will call a coherence-restoring state.

It would be ad hoc not to allow further beliefs to thus restore coherence in the case of other coherence requirements. Further, we need to consider not only subjects who believe that they are in special circumstances involving an exception to an otherwise true requirement – what about subjects who believe that a proposed requirement is not a true requirement of rationality in the first place? It might be more difficult to imagine how more radical (and perhaps wholly misguided) beliefs about what rationality requires could be justified, or how one could have reason to hold them. But note that the revisions we are considering do not impose epistemic conditions on the coherence-restoring beliefs. And this is as it should be, for such conditions would be at odds with the very idea of coherence at issue: the incoherence (or coherence) of a set of states doesn't depend on their epistemic credentials.

Moreover, I think we can have reasons (e.g. evidence) to believe false normative theories – in fact, this is inevitably the fate of a large proportion of our profession.Footnote ¹⁹ As an example, consider Evie, who has been persuaded by Susanna Rinard that there are no real, specifically epistemic norms governing belief, and that practical reasons can constitute reasons for or against believing.Footnote ²⁰ Evie believes that p is unlikely on her evidence, but recognizes that there are ample practical reasons to believe p. So she is persuaded that she in fact ought to believe p. According to a popular characterization of epistemic akrasia, Evie is epistemically akratic – and hence, irrational. But she has principled reasons to think that despite failing to respect her evidence, she is respecting the true norms governing belief. She does not fail by her own lights in the way that paradigmatic akratic subjects do.

We are on a slippery slope. Given the concession about coherence-restoring states, one cannot formulate requirements imposing more objective conditions that demand what Broome terms strict liability.Footnote ²¹ The problem is that every possible state that a more objective requirement deems incoherent has a coherence-restorer: for each candidate incoherent state, there are further beliefs that render one's overall mental state seemingly coherent.

The claim that believing that a state is rational can make that state rational by restoring coherence is striking. Hardly anyone thinks that meta-level beliefs to the effect that one is doing morally well can restore the morality of one's actions; or that meta-level beliefs to the effect that one is doing epistemically well (for instance, beliefs that one's belief is supported by the evidence) can restore the epistemic credentials of one's beliefs. But the problem is not just that proponents of structural rationality must accept a striking conclusion. For the concession that there are coherence-restoring states undermines the possibility of formulating any requirements to begin with – and they certainly undermine the canon of requirements put forth in the literature.

As far as I can see, the only principled way of resisting the slippery slope is by resisting the very first move: no matter what Di believes about dialetheism, it is irrational for her to hold inconsistent beliefs. But I certainly feel a strong pull to say that, at the very least, Di's believing both the Liar sentence and its negation involves a different kind of irrationality from that exhibited by paradigmatically incoherent subjects. If Di is irrational, her irrationality is of a rather sophisticated kind.

One response to the above arguments is to admit that any norm can be flouted in an excusable way, and the subjects considered – Ace, who lacks access to his own beliefs, and Di, who thinks believing some contradictions is perfectly appropriate – are structurally irrational, but seem less irrational because their incoherence is excusable.Footnote ²² I agree that any real norm can be excusably violated. But I want to remind the reader of the dialectic here. What reason is there to posit an array of structural requirements? These requirements, and the kind of coherence they delineate, explain our intuitive judgments and evaluations regarding a range of cases. In particular, they explain a distinct-seeming charge of a blatant kind of irrationality. But I have argued that our judgments and intuitions about cases are messy, and very imperfectly aligned with the coherence requirements we started out with. This considerably undermines any attempt to support structural requirements by appeal to intuitions or systematic patterns of a distinct kind of normative disapproval.

I have argued that violating a structural requirement is not sufficient for eliciting a judgment of a distinct, blatant-seeming normative failure. I will now argue that neither is it necessary.

3.2. Blatant irrationality without violations of structural requirements

Consider Al, who believes that the New York Times reported that p, and that there is absolutely no reason to distrust the report. However, even after considering the question, he fails to believe p. This is somewhat baffling: isn't Al just irrational? He resembles an epistemically akratic subject: perhaps his irrationality is slightly less blatant, but it still appears distinct from the failure of a subject whose evidence is complex and difficult to decipher. It's puzzling that Al doesn't treat a reliable newspaper reported that p, and there is no reason to distrust the report as a reason to believe, or at least be confident in, p. One could try to argue that Al in fact violates some familiar coherence requirement. But it can't be closure, for NYT reported that p and there is no reason to distrust the report doesn't logically entail p. Neither can Al be faulted on the grounds that he is epistemically akratic, for he may not have formed the (further) belief that p is likely on his evidence, that rationality requires him to believe p, or that he ought to believe p. Al doesn't violate one of the familiar coherence requirements. Yet, he seems irrational in much the same way as paradigmatically incoherent (e.g. akratic) subjects.

Perhaps it was wrong to turn to structural requirements to explain the kind of charge of irrationality elicited by considering paradigmatically incoherent subjects. For instance, Setiya (Reference Setiya2004: 276) argues that there is a sense of ‘irrational’, to be found in ordinary thought, on which irrationality is a matter of legitimate expectations: “A is irrational in doing ϕ only if A could be legitimately expected not to ϕ”.Footnote ²³ But cases in which subjects fail to meet legitimate expectations are not restricted to canonical cases of incoherence. For instance, at least without some elaborate background story involving false normative theories, it would seem that Al can be legitimately expected to form a belief based on the newspaper report. Further, what can be legitimately expected of a subject can, it would seem, shift depending on what she has access to and what she believes. It's not completely clear to me whether it is legitimate to expect subjects not to hold contradictory beliefs, but let us assume that it is. Nevertheless, such expectations are undermined if a subject has done her best to inspect the contents of her mind, and found no contradictions – though she does both believe p and believe not-p, she is in no position whatsoever to know that she believes p. Or, consider a subject in the grip of a false normative theory. In so far as Di believes that some contradictions are true, including a specific one she believes, it is at least much less clear whether we can legitimately expect Di to not believe contradictions. Similarly, take Worsnip's idea that certain states are difficult to make sense of from a first-person perspective. It would seem difficult to make sense of Al's mental states, even if he doesn't violate a requirement of structural rationality.

Structural requirements offer a poor explanation of our evaluative inclinations, and of a broader class of data. In so far as structural requirements capture a distinct kind of irrationality, our evaluative practices only track it in very incomplete ways. I now turn to my positive competence-based account of the badness of incoherence. I will argue that it offers a far better explanation of our rather messy evaluative practices.

4.1. AkrasiaFootnote ³³

Here is a rather standard understanding of what akrasia involves: to be akratic is either to believe that one (overall) ought to do A, while failing to intend to do A, or to believe that one ought not to do A, while nevertheless intending to do A. It is a kind of clash between one's intentions and one's normative beliefs. Below I discuss problems with formulating a requirement of means-ends incoherence that also affects this characterization of akrasia, but for now I will bracket these problems.

My positive account of the distinct-seeming badness of paradigmatically akratic agents appeals to the distinctive-seeming badness of certain normative failures to intend to do what one ought to do. I want to abstract away from questions of what the substantive, non-structural norms on belief, intention, and action are (e.g. should we believe what is true, what is known, or what is likely on our evidence?) For this reason I will deploy the ideology of reasons. My use of this ideology should not be interpreted as indicative of a commitment to any sort of “reasons first” program: I don't assume, for instance, that all normativity can be reduced to reasons. For all I say, moreover, the relevant reasons consist of one's evidence, which in turn is the totality of propositions one knows. In the end, my account could be stated without any talk of reasons: instead of saying that it seems blatantly bad to fail to intend to do A when I ought to do A is among one's reasons, I could instead observe that it seems blatantly bad to fail to intend to do A when one knows I ought to do A.

I will make the following assumptions about reasons. First, not all reasons are mental states: at least some reasons are propositional.Footnote ³⁴ (Note that a non-mentalist view of normative reasons is the standard view in meta-ethics.Footnote ³⁵) Second, p cannot be a normative reason for belief or action if p is false: falsehoods may serve as merely apparent reasons, but not as genuine reasons.Footnote ³⁶ Third, I shall assume that there is a distinction between normative reasons and the normative reasons a subject has.Footnote ³⁷ In order for a true proposition p to be among the reasons I have, I must bear some appropriate epistemic relation to p. I will assume that the relation entails believing p. For instance, that the number 6 tram is not currently operating might be a reason to take the metro instead, but if I bear no epistemic relation to the proposition that the number 6 tram is not operating (unless I, minimally, believe it), I may have no reason to take the metro. I find it plausible that p is a possessed normative reason just in case p is part of one's evidence, and that p is part of one's evidence just in case it is known. However, nothing I say below will assume this.

There are all sorts of reasons to believe and act, but I now want to focus on a special class. Consider the propositions I ought to do A, or I ought not to do B, for some actions A and B. In so far as I have any grasp on the notion of a reason, I find it difficult to deny that the first is (when true) a reason to do A, and that the second is (when true) a reason not to do B. If anything does, the first certainly counts in favor of doing A, and the second counts against doing B!Footnote ³⁸ In fact, the force of these reasons is maximal, for they are conclusive: there couldn't be other reasons that outweigh them in the overall balance of reasons. My situation might change, and I may acquire new reasons against doing A the result of which is that I no longer ought to do A, but were this to happen, I ought to do A would no longer be true, and hence, not a genuine reason. But there is also another important respect in which these reasons are special, for they as it were wear their force on their sleeves. It is sometimes hard to figure out what I overall ought to do, even when it is clear what the relevant considerations are – for instance, there may be a multitude of considerations pulling in different directions – but it is not that hard to figure out that I ought not to do B doesn't count in favor of doing B. The force of such reasons – what they point to – is conspicuous in a way that the force of many of our reasons is not.Footnote ³⁹ At least without some elaborate background story, failing to appropriately take such reasons into account appears blatantly vicious: in the absence of additional information, we expect agents to be appropriately responsive to them.

Consider Sai, a subject in a difficult and complicated situation who must balance the welfare of various members of his family, and come to a decision about how to use limited financial resources. In fact, Sai ought to use his savings to fund his children's education. Nevertheless, he just cannot figure out what he ought to do – quite understandably, given the complex nature of his situation. Luckily Sai can consult an omniscient oracle. The oracle tells him that he ought to invest in his children's education, and Sai believes what he is told, thereby coming to know what he ought to do. If he still forms no intention to invest in his children's education, it is considerably more difficult to make sense of him: it will certainly no longer do to point to the complex and opaque nature of his situation. Sai's failure at this point looks distinct and more blatant than his initial failure, for it is no longer unclear to him what he ought to do.

My general suggestion will be that the seeming badness of akrasia can be explained in terms of the badness of failing to take into account conspicuous reasons, which, at least other things being equal, constitutes a blatant-seeming normative failure. More precisely, the badness of akrasia is a matter of manifesting dispositions that indiscriminately manifest as failure to take into account seeming conspicuous reasons, whether or not those seeming reasons are real.Footnote ⁴⁰ In this respect, the akratic subject is like the candidate lifeguard who fails to respond in a simulation of someone drowning, or the airport security worker who fails to respond to an image of a real-looking decoy bomb.

My account relies on an assumption about feasibility. The core assumption will be that it is not feasible to have dispositions that discriminate between cases in which one has a genuine reason, and cases in which one has a mere belief. As a special case, it is not feasible to have dispositions that discriminate between cases in which one possesses the real reason I ought to do A (or I ought not to do B), and cases in which one merely believes this. Beliefs serve as the inputs of certain inferences leading to other beliefs – and more generally, trigger doxastic revisions – and interact with motivational states like desires and intentions to produce action. This much is almost uncontroversial. But in basing further beliefs via inference on a belief in p, and using this belief as a basis for action, one treats p as a reason. Here is what I find to be a very appealing, albeit rough, characterization, of the functional profile of belief: believing p looks, from the inside, like possessing the genuine reason that p.Footnote ⁴¹ That is, belief plays the same local role as the epistemic state required for possessing a genuine normative reason – for instance, if possessing p as a reason requires knowing p, then belief plays the same local role as the state of knowing.Footnote ⁴²

A subject treats the contents of her own beliefs in the same way as she treats genuine reasons. It follows from this that competent subjects respond to mere beliefs in ways that are appropriate when possessing genuine reasons. If this is right, then exercising basic competence with reasons like I ought to do A will involve forming an intention to do A (if one doesn't already have it) based on a mere belief that one ought to A, just as exercising one's competence qua airport security worker will involve taking action when encountering a mere real-looking decoy bomb. Similarly, exercising a competence to respond appropriately to I ought not to do B will involve giving up one's intention to do B (if one has such an intention) in response to one's belief that one ought not to do B.

Here, then, is my positive account of why paradigmatic cases of akrasia seem bad, and of what this badness involves. Consider the akratic subject who believes that she ought to do A, but who fails to intend to do A. There are two kinds of cases to consider. First, I ought to do A may be among the subject's reasons, which entails that she ought in fact to do A. If she fails to intend to do A, then she fails to appropriately respond to a real conspicuous reason she has – which is surely a very bad kind of normative failure. More interesting and trickier are cases in which, though she believes I ought to do A, this proposition is not among her reasons. Assume that it is false and hence, cannot be a real reason.Footnote ⁴³ Since it is false that she ought to do A, she is not flouting substantive normative requirements by failing to intend to do A. But now consider the disposition she manifests by being thus unresponsive to the contents of her belief that she ought to do A. She manifests at least a local disposition to fail to intend to do something that she believes she ought to do. How good is this disposition – does it fare across counterfactual cases in which it manifests itself? Well, given the above point about feasibility, it doesn't fare well across those counterfactual cases in which the contents of the subject's beliefs in fact constitute genuine reasons: in those cases disregarding the contents of beliefs such as I ought to do A constitutes failure to take into account a real reason. And such a normative failure seems particularly bad.

It is important to note that a subject may simultaneously be manifesting some good and some bad dispositions. In fact, she may even be exercising some part of competence, while flouting others. Assume that Dr Akra is epistemically akratic: she believes that she ought to do A, but she fails to intend to do A. However, Akra's belief that she ought to do A is false, and may even be based on no good reasons or evidence. In believing that she ought to do A, Akra fails to believe as she ought. But this does not wash out the failure of being akratic. True, Akra may act and intend exactly as she ought – from this perspective, she may even be better off not taking heed of her false normative belief. But recall that the kinds of evaluations that go together with imputing some form of incoherence are local: we might criticize a subject for failing to intend to do what she believes she ought to do, while recognizing that overall this isn't such a bad thing, given how skewed her beliefs about what she should do are.Footnote ⁴⁴ On the account I have sketched, the badness of being akratic involves manifesting some problematic dispositions, dispositions that manifest across a range of cases as a certain kind of blatant-seeming normative failure, and that fly in the face of some component of normative competence. Even if, all things considered, a subject in fact does better by failing to intend to do A while believing that she ought to do A, she is still manifesting a bad disposition.Footnote ⁴⁵

Various accounts of conspicuousness are compatible with my proposed explanation. An account of conspicuousness might appeal to special epistemic features of the relevant reasons: perhaps we have a special kind of epistemic access to what these reasons are reasons for or against, where the access might be a priori, or conceptual/analytic, flowing from an understanding of the relevant concepts, etc. Or, an account of conspicuousness might appeal to our expectations as evaluators: absent background information, conspicuous reasons are reasons that we expect any minimally competent subject to be able to appropriately take into account.Footnote ⁴⁶ Relatedly, it may be that the more conspicuous a reason is, the more of an explanation we need in order to rationalize, or make sense of, a subject who fails to take into account the reason in question. Whatever conspicuousness amounts to, my account ties together the conspicuousness of a reason with our inclinations qua evaluators: other things being equal, subjects who fail to take into account conspicuous reasons fail in a blatantly bad-seeming way. For all I say, the conspicuousness of a reason just is a matter of how we tend to evaluate failures to take it into account.

Whatever the correct account of conspicuousness is, whether or not R is a conspicuous reason to φ can depend both on matters of access to our own mental states, as well as on possible normative beliefs. For instance, p is unlikely on my evidence is a conspicuous reason to not believe p.Footnote ⁴⁷ However, it may not be conspicuous to a subject in the grip of a false theory who thinks that what she ought to believe is not a matter of what her evidence supports. As an example, consider our subject who has been persuaded by Susanna Rinard that there are no real, specifically epistemic norms governing belief, and that practical reasons can constitute reasons for or against believing. She believes that p is unlikely on her evidence, but recognizes that there are ample practical reasons to believe p. So she is persuaded that she in fact ought to believe p. Even if Rinard is wrong and p is unlikely on my evidence is a reason not to believe p, the failure of such a subject is no longer blatant. For instance, we may no longer expect her to correctly take the reason into account, and it is much easier to make sense of why she doesn't believe p, despite believing that it is likely on her evidence. Even if she fails normatively, her failure is now of a more subtle, less blatant kind.

I have argued that at least paradigmatically akratic subjects manifest a problematic disposition: a disposition to be unresponsive to the contents of certain beliefs, a disposition that fails to discriminate between cases in which she has a merely apparent reason and cases in which she has a genuine conspicuous reason. The account I have proposed locates akrasia in a larger class of failures that involve manifesting problematic dispositions. The sense that akrasia is a distinct kind of failure is explained by the special nature of the kinds of reasons discussed above – in particular, their conspicuousness. The distinctness of the failure to appreciate the force of such reasons is one of degree, not kind. But this is as it should be, given the arguments in the first half of this paper.

4.2. Inconsistent beliefs and closure incoherence

The competence-based account is straightforward to apply to various other putative cases of incoherence.

Consider, for instance, a subject who both believes a proposition p, and believes its negation, not-p. p is a conclusive and conspicuous reason to not believe not-p, and not-p is a conclusive and conspicuous reason to not believe p. If I possess the reason it's raining, then I have a conspicuous reason not to believe that it is not raining; if I possess the reason it's not raining, then I have a conspicuous reason to not believe that it is raining. Failure to appropriately take into account such reasons appears blatantly bad, in a way that failure to take into account complex meteorological data supporting rain does not.

Consider also subjects who are closure incoherent. I will assume that it is not incoherent in the intended sense to fail to believe just any entailment of what one believes: first, some entailments are not at all obvious (for instance, complex mathematical theorems) and second, cluttering one's mind with all sorts of irrelevant beliefs might be a positively bad thing to do. But it certainly seems bad to fail to believe a proposition q if one believes p, the entailment between p and q is obvious enough, and the question about whether q is salient. Now, when the entailment between p and q is obvious enough, p is a conspicuous reason to believe q. Hence, the problem with closure incoherent subjects is that they fail to exercise a competence to respond to conspicuous reasons. Assuming that believing all of the obvious entailments of one's beliefs is not a desideratum, the competence in question involves engaging in such conspicuous pieces of reasoning only when a given question arises to salience – for instance, when a she cares about whether a given proposition is true.

I have argued that the seeming badness of many paradigm cases of a blatant kind of irrationality involves manifesting bad dispositions, thereby flouting some component of normative competence. A security worker who fails to take action upon observing a real-looking fake bomb manifests bad dispositions, for whatever dispositions are at play will indiscriminately manifest as failure to act across a range of cases involving real bombs. Similarly, a subject who believes that she ought to do A, but who fails to intend to do A, manifests bad dispositions, even if it is false that she ought to do A, for mere beliefs look from the inside like real reasons; it is simply not feasible to be disposed to only take into account real reasons.

Let me now turn to how the competence-based account can be extended to explain what is wrong with subjects who are means-ends incoherent.

4.3. Means-ends incoherence

Recall Ulla, who intends to run an ultramarathon in November, despite lacking good reasons to do so. Nevertheless, given that Ulla believes that it is necessary for her to train in order to run the marathon, she appears irrational if she has no intention whatsoever to integrate runs into her routine. At the very outset, however, I want to flag some problems for common characterizations of means-ends incoherence, for I think they are symptomatic. After putting forth my positive account, I will say why it is better suited to deal with these problems than ones positing requirements of structural rationality.

The following should sound familiar:

There are a range of cases, however, in which a subject is means-ends incoherent in this way, but nothing seems amiss. Some means are automatic: I know that flexing my biceps is a necessary means for picking up the glass I am holding, but even if I intend to pick up my glass, I don't need to intend to flex my biceps.Footnote ⁴⁸ In some cases, necessary means are actions I can trust myself to perform, whether or not I now so intend.Footnote ⁴⁹ Note that these problems also threaten standard formulations of enkratic principles prohibiting akrasia. For instance, I believe that I ought to flex certain muscles (since I ought to open the door). But flexing my muscles is automatic; I don't need to intend to do so.

A widely endorsed response to such worries is to opt for a different characterization of means-ends incoherence:

But consider just how restricted this principle is. I very much doubt that even intellectually sophisticated adults competent with the notion of an intention often hold beliefs about specific intentions being necessary means for doing various things. Consider a subject who intends to make it to the meeting in time, and who believes that leaving now is necessary for doing so, but who doesn't hold any such belief about the necessity of intending to leave now. At the very least, being in this combination of states is surely possible. If the subject forms no intention to leave, she nevertheless seems irrational.

In sum: while failing to intend a believed necessary often appears irrational, this is not always the case. An adequate account should be able to explain this, and to distinguish between cases in a principled way.

My general strategy has been to explain the seeming badness of various kinds of incoherence by appeal to the idea that incoherent subjects manifest bad dispositions, dispositions that are incompatible with some component of normative competence, and that, moreover, manifest in some cases as blatantly bad-seeming normative failures. The same is true, I will argue, of means-ends incoherent subjects, though the details will be a bit different. I will appeal to dispositions to track (or fail to track) an important structural feature of normative reality, the transmission of oughts to necessary means:

I will assume that what is at issue here is what one all-things-considered ought to do.Footnote ⁵² Unlike a requirement of means-ends coherence or instrumental rationality, Transmission doesn't make reference to an agent's intentions or beliefs. It says that oughts transmit to what are in fact necessary means, irrespective of the intentions or beliefs of an agent.

I cannot here offer a defense of Transmission. Along with many others, I find it intuitively compelling.Footnote ⁵³ And I find the counterexamples to the principle unconvincing.Footnote ⁵⁴ My conclusion here may be read as conditional: if Transmission is true, then we can give an account of what is wrong with means-ends incoherence. The idea will be that a core part of normative competence consists in being disposed to track this transmission of oughts, which I will argue typically involves extending one's intentions to believed necessary means.

If Transmission is true, then some facts about what one ought to do are true because doing these things is a necessary means for doing other things that one ought to do. Call these the transmitted facts, and those not true in virtue of the transmission of oughts the source facts. Normative competence involves being disposed to track both the source and transmitted facts. It is plausible that these in turn will involve different sub-competences: first, a (complex set of) dispositions allowing one to be moved to act in accordance with the true source facts, and second, a more structural kind of competence consisting of dispositions to extend one's motivations in a specific way. This more structural competence will play a key role in my account: I will argue that paradigmatically means-ends incoherent subjects fail to exercise this structural competence. Moreover, extending one's motivations in the way to be specified seems like the easy part of tracking facts about what one ought to do. Sometimes having the right ends is difficult, as is knowing what is a necessary means for what, but when one does have the right ends and knows what the necessary means to one's ends are, failing to pursue these necessary means seems like a particularly bad kind of normative failure.

Assume that I intend to do A, as I ought, and since doing B is necessary for doing A, by Transmission I also ought to do B. Consider how we paradigmatically manage to act in accordance with such transmitted ought-facts. At least very often (though not alwaysFootnote ⁵⁵) doing something requires being in a prior motivational state directed at doing that very thing – in particular, I must get from an intention to do A to an intention to do B. Moreover, this will normally require knowing (and hence, believing) that doing B is necessary for doing A. But given points made above about the nature of belief, it is not feasible to be disposed in ways that discriminate between cases depending on whether something constitutes a real reason or a mere belief. For instance, it is not feasible to be disposed in ways that discriminate between mere beliefs and knowledge: one cannot be disposed to extend one's motivations to real necessary means, but not to mere believed necessary means.

Hence, a disposition to extend one's motivations from ends to necessary means will involve a disposition to extend one's motivations to believed necessary means. Being thus disposed, I want to suggest, is precisely the structural sub-competence enabling one to track the transmission of oughts to necessary means. The transmission of oughts is a structural feature of normative reality, and mirroring it is a structural competence, a disposition for one's motivations (or dispositions to act) to transmit from ends to believed necessary means. It involves responding to certain features of one's mental states – specific belief-intention pairs – by coming to be disposed to pursue the believed necessary means to one's ends. And at least in many cases, we come to be thus disposed by forming a new intention.

Consider again Ulla, who intends to run an ultra-marathon in November, believes that some serious training is necessary for doing so, but who forms no intention to start training. Recall that Ulla in fact ought not to run an ultra-marathon. But I have argued that a competence to track the transmission of oughts must involve a structural sub-competence, one that manifests as coming to be disposed to pursue believed necessary means to one's ends. Whatever dispositions Ulla is manifesting, it is clear that they are out of line with such a sub-competence. Having the right ends is not that easy, but at least in many cases, thus extending one's motivations seems like the easy part of tracking the transmission of oughts.

It is worth returning here to the local nature of the seeming badness of structural irrationality. My claim is not that every subject would be normatively better off across relevant, somewhat normal cases if they manifested the structural competence I have outlined. The local nature of the evaluations we are after is explained, I am arguing, by looking at components of normative competence – roughly, dispositions that, together with other feasible dispositions, tend to manifest as normative success (doing, intending, believing, etc., as one ought). Consider again what it takes to be a good lifeguard. This will involve various sub-competences, manifesting as component successes: for instance, one must be able to recognize alarming situations (e.g. the signs of drowning), and to make the right interventions, which result in getting the person identified as being in distress out of the water. If someone is very bad at recognizing alarming situations, their complete incompetence at even getting to the people they have identified as needing help might in fact prevent them from making mistakes like dragging someone happily swimming out of the ocean. Despite this, they manifest incompetence, and are criticizable for doing so.

Similarly, consider the success of acting as one ought. Someone who systematically fails to track the source ought-facts, being motivated to do things they ought not to do, might overall be better off if she also flouted the structural sub-competence, for her systematic inability to pursue necessary means to her ends would then fend off normative failure. Still, if I am right that a competence to track transmitted ought-facts requires a disposition for one's motivations to transmit to believed necessary means, she is still manifesting normative incompetence. Moreover this part of normative competence seems less demanding than, for instance, that involved in having ends corresponding to the true source-facts. Other things being equal, for instance, we would expect subjects to be able to extend their motivations to believed necessary means. Indeed, when one's normative failure is due to such failure to extend one's motivations, it seems particularly blatant: if I intend to do A, as I ought, and I know that doing B is necessary for doing A, then other things being equal, it seems particularly bad for me to fail to intend to do B.

An account of what is wrong with means-ends incoherent subjects along these lines has several virtues. Unlike most other accounts that don't simply invoke a requirement of means-ends coherence, my account doesn't rely on any particular view of intention. Neither does it rely on an implausible luminosity assumption to the effect that means-ends incoherent subjects are always in a position to know that they are means-ends incoherent.Footnote ⁵⁶ Further, recall the kinds of problems encountered when trying to formulate a requirement of means-ends coherence: a familiar first pass requirement has counterexamples, but a popular fix only applies in a narrow range of cases. A merit of the account proposed is that it can explain why failing to intend believed necessary means seems blatantly irrational in some cases, but not in others. The structural aspect of a competence to track facts about transmission requires that one's dispositions to act spill over from ends to believed necessary means. In paradigm cases, this involves the spilling over of intentions from ends to believed necessary means. But not always: I don't need to intend to flex my muscles in order to do so. I become disposed to flex my muscles (in order to get up) just by forming an intention to open the door. Similarly, in cases in which I rightly trust myself to do certain things, I am disposed to do those things even if I don't now intend to do them.

Before concluding my discussion of means-ends incoherence, let me address an objection to the account I have sketched.Footnote ⁵⁷ Consider so-called “permissive cases”: cases in which a subject is permitted, but not required, to do A. Assume that a subject in such a case intends to do A, and believes that doing B is a necessary means for doing A, but nevertheless fails to intend to do A. The worry is that my account cannot explain why such a subject is criticizable. In particular, we could imagine a subject who is disposed to track the transmissions of oughts, and whose means-ends incoherence tends to be restricted to such permissive cases. Even if permissions transmitted to necessary means, not doing (or intending to do) something one is merely permitted to do is not a normative failure.

Given widely accepted views about reasons, such “permissive cases” will have the following structure.Footnote ⁵⁸ First, in order for one to be permitted to do A, there must be an undefeated reason to do A. However, one is merely permitted (and not required) to do A because one also has an undefeated reason to do B, and one cannot do both. (For instance, perhaps I have equally good reasons to cook lentil soup and pasta for dinner, but because of time constraints, I cannot do both.) In such a case, one is permitted to intend to do A, and one is permitted to intend to do B. However, I think it is plausible that in such a case resolving the standoff by forming an intention makes a normative difference: if I form the intention to cook lentil soup instead of pasta for dinner, my overall balance of reasons now weights in favor of lentil soup. Here is one explanation why.Footnote ⁵⁹ I ought, after all, do either A or B (assuming those are the only options). And in order for me to do either A or B and not end up like Buridan's ass, I must commit to one and then, having made a choice, carry through with it. It is necessary for me to either choose to do A or choose to do B, and to carry through with whichever choice I make. The fact that I ought to do one of the things I am permitted to do, then, together with an application of Oughts Transmission, explains why, in such permissive cases, forming an intention one way or another makes a normative difference.Footnote ⁶⁰ If this is right, then if I am in the sort of permissive case described, and I permissibly intend to do A, then I in fact ought to do A: my forming an intention one way rather than the other breaks the balance of reasons. If this is right, and I intend something I am permitted to do, then we can once again apply Oughts Transmission: I ought to pursue the necessary means to cooking lentil soup.

Before concluding, let me make some final remarks. The kind of evaluative perspective I have deployed to explain the seeming badness of paradigm cases of incoherence is focused on the dispositions manifested by incoherent agents: these dispositions, I have argued, are at odds with normative competence, and manifest in other cases that one cannot feasibly discriminate from one's actual case as blatant-seeming normative failures. Various authors have tried to explain the seeming normativity of structural requirements of rationality by using the language of dispositions. A common theme has been the idea that being structurally rational is a means of doing what one has reason to do.Footnote ⁶¹ However, at least given a natural way of construing talk of instrumental means, on my view a disposition to conform to structural requirements is not merely an instrumental, contingent means to believing and doing what one has reason to believe and do. What makes a disposition good or bad in the relevant sense is not whether its manifestations are means to success, but whether its manifestations tend to constitute successes. Getting enough sleep might be an instrumentally good disposition for an airport security worker to have, for being alert is instrumental to successfully recognizing bombs. By contrast, dispositions constituting recognitional capacities and competences are related to success in a more direct way: manifesting them isn't merely an instrumental means to success; in good cases, manifesting them constitutes successes like recognizing real bombs.

One objection often raised for views that regard rationality as a kind of virtue is that they cannot account for the normative (as opposed to merely evaluative) dimension of the distinct charge of irrationality. Kolodny (Reference Kolodny2005: 554–5) argues that this normative dimension is manifest both ‘from the outside’, in the form of advice about what one ought to do or believe, and ‘from the inside’, as pressure to be in states that rationality requires of us. But I claim that we often do make normative-sounding evaluations from a perspective oriented around good dispositions. Recall Simu, the candidate lifeguard who fails to act despite witnessing a simulation of a classic drowning response. It would feel perfectly appropriate for bystanders aware of the virtual-reality test to say, “He should jump in!”. It is this same evaluative perspective centered around dispositions and competences that explains our normative indignation at paradigmatically incoherent subjects.