Summary: | I. Reasons and Requirements of Rationality. II. Rational permissions and correct reasoning. III. References. |
At the heart of John Broome’s research program in the philosophy of normativity, culminating in his magnificent book Rationality Through Reasoning, is a distinction between reasons, on one hand, and requirements of rationality, on the other.1 In my view, Broome’s insistence on pulling these notions apart, and resistance to analyzing either one in terms of the other, is fundamentally correct, and represents a major advance in the field.2 At that level of generality (and in numerous other respects, too), Broome and I are on the same team.
This paper is divided into two sections. In the first part of the paper, I’ll examine Broome’s distinction between reasons and requirements of rationality. Though the remarks come from a position of sympathy with that general distinction, I’ll discuss three worries about it, which I’ll call the threat of merely terminological debate, the threat of superfluousness, and the threat of disappearing subject-matter. In discussing how these threats can be headed off, I’ll suggest some friendly amendments and modifications to Broome’s way of presenting and developing the distinction.
The second part of the paper will be more substantively critical.3 In addition to the notion of a reason and that of a requirement of rationality, there is a third normative notion that Broome is interested in: that of (rules of) correct reasoning. Broome’s suggestion is that this third notion, correct reasoning, is closely tied to that of requirements of rationality (or at least to the broader genus of which requirements of rationality are a part). On Broome’s view, every rule of correct reasoning corresponds to a “basing permission”, which states that it’s rationally permissible to base one attitude on one or more other attitudes. I’ll argue that this proposal can’t be made to work. If I’m right, this suggests that Broome should admit rules of correct reasoning as a third sui generis normative entity, not reducible either to reasons or to requirements (or permissions) of rationality. That is: the same kind of pulling-apart that Broome has effected so persuasively with respect to reasons and requirements of rationality needs to be effected again to separate rules of correct reasoning from both of those other categories.
Parts of this paper will be of necessity programmatic, and in several places I’ll have to engage in the annoying habit of referring the reader to other work of mine. I apologize for this in advance. Though this practice may seem self-centered, there is a way in which what it really shows most is how indebted much of my own research program is to Broome’s. My own work is often relevant to my discussion of Broome here only because so much of my work addresses philosophical debates that Broome has had such a central role in framing.
I. Reasons and requirements of rationality
Roughly speaking, the distinction between reasons and requirements of rationality is this. Reasons are considerations that speak in favor of attitudes and acts. For example:
- That one hears the sound of rain is (typically) a reason to believe that it’s raining;
- That it’s raining is (typically) a reason to carry an umbrella.
Requirements of rationality, by contrast, generally permit or forbid certain combinations of attitudes (and absences of attitudes).4 For example:
- It’s rationally required that one not simultaneously believe that it’s raining and believing that it’s not raining.
- It’s rationally required that one not simultaneously intend to carry an umbrella if it’s raining, believe that it’s raining, and yet fail to intend to carry an umbrella.
One way of further strengthening one’s grip on the distinction here is to note that we seem to be able to say what the relevant requirements of rationality on an agent are without knowing anything about the situation the agent finds herself in, whereas this is at least often not so for reasons. No matter what the subject is hearing, seeing, and so on, the subject is always rationally required not to both believe that it’s raining and believe that it’s not raining. By contrast, whether the subject has a reason to believe that it’s raining depends very much on the subject’s circumstances, that is, on what the subject is hearing or seeing (and conceivably, on some views, whether it’s actually raining or not).5
There are some people who are sympathetic to a distinction broadly in the neighborhood of the one being drawn here, but who think that the generally Broomean gloss on it just given already goes wrong in important ways.6 That won’t be the tack I take here. I’m happy to accept the general gloss on the distinction just given (with the caveat that I’ll propose a terminological revision in the next subsection).7 Still, I want to raise three issues that threaten, or may appear to threaten, the depth and significance of the distinction, and discuss how they might be headed off.
1. The threat of merely terminological debate
As the foregoing makes clear, the requirements that Broome calls ‘requirements of rationality’ pertain to the rational (im)permissibility of certain patterns or combinations of attitudes, independently of the circumstances of the subject. However, many philosophers believe that we are, in some sense, rationally required to respond appropriately to our reasons: that is, to perform the acts that our reasons (on balance) support performing, and to have the attitudes that our reasons (on balance) support having - or some attenuated version of this claim.8 In fact, for many philosophers this claim is treated as a truism or axiom.9 Such philosophers can still, in principle, accept that there is an important distinction between the requirements of rationality that pertain to responding to reasons and the requirements of rationality that pertain to avoiding certain impermissible combinations of attitudes. Still, they hold that both of these are bona fide requirements of rationality.10
On one view, the dispute between Broome and these philosophers is a merely terminological one. Several philosophers who use the term ‘requirement of rationality’ to pick out roughly what Broome does present this usage as a matter of terminological stipulation.11 They’re happy to acknowledge that there’s a potentially legitimate usage of ‘requirement of rationality’ whereby there are requirements of rationality to respond appropriately to reasons; still, they don’t want to lose sight of the fundamental distinction Broome is after, and so they use ‘requirement of rationality’ in a narrower way to refer only to requirements pertaining to rationally (im)permissible combinations of attitudes.
But this isn’t how Broome himself thinks of things, at least not in his more recent work. Instead, Broome advances the claim that rationality does not require appropriately responding to reasons as a substantive claim, rather than as a matter of terminological stipulation about how to use the term ‘rationality’.12 This isn’t to say that Broome thinks reasons are somehow normatively irrelevant; on the contrary, he thinks that your reasons determine what you ought to do.13 His claim is that it isn’t irrational to fail to respond to your reasons (as such).
As I understand him, Broome’s reasons for thinking of this claim as substantive rather than terminological are as follows. According to Broome, it’s a conceptual truth that rationality supervenes on the mind.14 But, Broome thinks that the reasons one has don’t supervene on one’s mind, and consequently that what is an appropriate response to one’s reasons doesn’t supervene on the mind. Therefore, rationality is not a matter of responding to one’s reasons. A defender of the view that rationality is a matter of responding to one’s reasons has only two options to resist this argument. The first is to claim that rationality does not supervene on the mind. This is, in Broome’s view, a substantive (though conceptual) falsehood about rationality. The second is to argue for an account of reasons such that reasons (or, at least, the reasons that rationality requires one to respond to) do supervene on the mind. This is, in Broome’s view, a substantive falsehood about reasons. So either way, the philosopher who claims that rationality is a matter of responding to one’s reasons makes a substantive mistake.
At least for the sake of argument, I’ll grant Broome’s premise that rationality supervenes on the mind. I’m also happy to grant that there is a reading of ‘reason’ such that one’s reasons do not supervene on the mind.15 On this reading of ‘reason’ -sometimes unhelpfully called the ‘objective’ reading of ‘reason’- facts that are totally outside of one’s ken can be among one’s reasons. For example, to use Williams’ classic example, perhaps the fact that the glass contains petrol is a reason not to drink it, even if one is not aware that the glass contains petrol and (justifiably) believes that it contains gin and tonic. Plausibly, this notion of reasons doesn’t supervene on the mind: two subjects could have identical mental states, but if the first is in a case where the liquid in the glass is petrol, and the other is in a case where the liquid is gin and tonic, then the former has a reason (in this sense) not to drink that the latter lacks.16 And -as we would expect, if we agree with Broome that it’s a conceptual truth that rationality supervenes on the mind- it does seem that there’s no recognizable sense of the term ‘rationality’ on which the former subject is irrational if she drinks (or intends to do so), given that the reason not to drink is one that she is totally unaware of. Thus, Broome is right that rationality is not a matter of responding to (all of) one’s reasons, on this understanding of ‘reason’. And this is not just a matter of terminological stipulation: it seems to follow from conceptual truths about rationality.
But sophisticated advocates of the view that rationality is a matter of responding to reasons have not claimed otherwise. Instead, they have typically made one of two claims. The first is that rationality requires you to respond to some privileged subset of your reasons: for example, the reasons that are epistemically accessible to you.17 On this proposal, we still understand the notion of a reason in a way such that facts outside of one’s ken are reasons, but we claim that it is only the reasons that are within one’s ken that rationality requires one to respond to. The second is that rationality requires one to respond to one’s respond to all of one’s reasons, but on a reading of ‘reasons’ where facts outside of one’s ken are not reasons at all.18
Note that neither view need collapse into the view that rationality is a matter of responding to one’s beliefs about the (objective) reasons one has. It might suffice for R’s being a reason to Ф in the relevant sense that one is aware that R -one need not believe that R is a reason to Ф. Or it might suffice for R’s being a reason to Ф in the relevant sense that one’s evidence supports believing that R is an (objective) reason- whether or not one believes that R is a reason to Ф (and perhaps even whether or not one believes that R).19 Broadly speaking, these sorts of accounts says that the reasons you’re rationally required to respond to are your evidence-relative reasons. They thus lie between the two extreme versions of the reasons-responsiveness view that Broome considers in chapters 5 and 6 respectively -on one hand, one that operates on a conception of a reason that is fact-relative and unconstrained by evidence, and on the other hand, one that operates on a conception of a reason that is belief-relative (or that talks of responding to beliefs about reasons). More generally, many ethicists have been stuck in a false dichotomy between so-called “objective” reasons, which are relative to all the facts, and “subjective” reasons, which are relative to one’s beliefs. This misses the intermediate notion of an evidence-relative reason. That evidence-relative reasons have been obscured from view in debates in ethics and practical rationality is odd, since they form the basis for the default view in accounts of rationality in epistemology.
On these versions of the reasons-responsiveness view, it is quite plausible that whether one has rationally responded to the relevant reasons does supervene on the mind. Broome takes this possibility to be a threat.20 Because both his own view and these views might be consistent with the claim that rationality supervenes on the mind, he takes them to be rival views, neither of which is ruled out by the central conceptual truth about rationality that ruled out less sophisticated versions of the view that rationality consists in responding to reasons. But here I think Broome is mistaken. Plausibly, Broome’s account of rationality and sophisticated accounts of rationality as responding to reasons are not accounts of the same single thing, rationality simpliciter. Broome seems to be assuming that because both accounts accept the constraint that the notion they’re theorizing supervenes on the mind, they must be rival accounts of the same thing. But this patently does not follow. There can be two distinct normative notions, both of which supervene on the mind, and which don’t compete as rival accounts of the same thing.
A concrete example may help to illustrate the point. Suppose you know that 99% of scientific experts agree that human activities have contributed to climate change. Call that proposition (or the fact that it corresponds to, if you prefer) E. Assume, as is plausible, that this E is a decisive reason to believe H -the proposition that human activities have contributed to climate change. But suppose that you deny that E is a decisive reason to believe H. For you hold that scientific experts are all involved in a giant global conspiracy to deceive people into false beliefs about climate change. Moreover, suppose that there’s no interfering higher-order evidence -your belief that scientific experts are involved in this giant conspiracy is itself unjustified (though we may nevertheless suppose that you believe it to be justified). Finally, suppose that you fail to believe H.
Now, the following seems clear. Your failure to believe that H is unjustified, in the ordinary epistemological sense of ‘unjustified’: it is a failure to correctly respond to your evidence.21 It is thereby a failure to correctly respond to your reasons, (even) in the sense (or of the kind) that the sophisticated reasons-responsiveness accounts of rationality fix on.22 The relevant reason that you’re failing to respond to - E - is not something outside of your ken; nor does your evidence suggest that E is not a reason to believe H. At the same time, you need not violate a requirement of rationality of the sort that Broome’s theory focuses on. That is to say, you are not (or need not be) structurally irrational; there need be no internal incoherence in your belief-set.23 Moreover, both of these statuses -that you have an unjustified (or reasons-unresponsive) belief, and that your belief-set is internally coherent- are ones that, for all that has been said, supervene on the mind.24 But that need not mean that, in fact, they are rival accounts of a single notion, rationality. Rather, they seem simply to be distinct normative statuses.
Similarly, consider a structurally analogous moral case. Suppose you’re a CEO who knows that some action (say, opening a particular new factory) will gravely damage the environment. And suppose the fact that it will gravely damage the environment is a decisive reason not to open it (and that you lack interfering higher-order evidence that it’s not a decisive reason not to open it). But suppose you simply don’t care that it will gravely damage the environment, and don’t believe that this fact is any reason at all not to open it. Again, you’re failing to respond to a reason of the kind that the sophisticated reasons-responsiveness accounts of rationality fixes on. But there’s nothing incoherent about your states. Again, both of these statuses may supervene on the mind, yet they seem distinct and not rival account of one single thing, rationality.
That’s not to say that there won’t be those who insist that the term ‘rationality’ be reserved for either one or the other of these statuses. But that really is a matter of terminology (and neither usage obviously violates the conceptual truth that rationality supervenes on the mind). The important thing is that we mark the distinction between reasons-responsiveness and coherence somehow.
Interestingly, it seems that when focusing on the epistemic case, it seems to be the usage of ‘rationality’ to refer to coherence that more often gets jettisoned, whereas when focusing on the moral case, it seems to be the usage of ‘rationality’ to refer to reasons-responsiveness that more often gets jettisoned. Sometimes, the very same philosopher seems to use the term ‘rationality’ to refer to (or at least include) reasons-responsiveness in the epistemic case, but to refer to coherence only in the moral (or more generally practical) case.25
As I’ve argued elsewhere,26 there is no justification for this asymmetry. Only confusion will result from using ‘rationality’ to refer to (or include) reasons-responsiveness in epistemology, but to refer to coherence (only) in ethics.
Indeed, Broome himself may be guilty of this. In a brief but telling passage, he writes:
No doubt one or more requirements connect a belief that you ought to F with your evidence that you ought to F; I do not try to specify those requirements in this book. So there are a number of states you might be in -call them ‘grounding states’- that ensure you are not rational unless you believe you ought to F (140).27
Here, Broome seems to allow that there are rational requirements that require one to have beliefs that are appropriately responsive to one’s evidence. But appropriate responsiveness to one’s evidence is just appropriate responsiveness to a particular kind of reason -evidential reasons. So, if there are such rational requirements, they are rational requirements to respond to one’s reasons. This is striking, since as we’ve seen, Broome denies -with primary attention to the practical, rather than the epistemic case- that it’s rationally required to respond to one’s reasons. Of course, if evidence supervenes on the mind, Broome’s concession here doesn’t violate his guiding principle that rationality supervenes on the mind. But that just reinforces my point that there is a notion of reasons-responsiveness that supervenes on the mind. It’s most familiar from epistemology,28 but it can be extended to the practical case also. Indeed, anyone who denies that there’s at least an evidence-relative sense of the term ‘reason’ will have great difficulty accounting for the epistemic case.
Still, any requirements to respond to one’s evidence (or reasons more generally) are of a fundamentally different kind from the requirements of coherence that Broome is primarily engaged in theorizing, and the two should be kept apart. So, although I’m not taking a stand on whether ‘rationality’ should be used to refer to reasonsresponsiveness or coherence, I do think we should strive to make our usage consistent across domains. One disadvantage of working simply with the term ‘rationality’ is that it obscures possible inconsistencies of this kind.
The result of all of this is that, once we have the best reasons-responsiveness view on the table, the debate about whether rationality is about reasons-responsiveness or about coherence does seem to be close to being terminological. This might seem like a disappointment for Broome’s project, since two whole chapters of Rationality Through Reasoning are devoted to arguing that rationality doesn’t require responding to reasons. But in fact, I think that it doesn’t really compromise the ultimate project at all (indeed, I think it strengthens it). My proposal is that we use the term ‘coherence requirements’, or ‘structural requirements’, to refer to the sorts of requirements that Broome is engaged in theorizing, and ‘coherence’ or ‘structural rationality’ to refer to the phenomenon that they’re a part of.29 It would be a change of vocabulary for Broome to talk this way, instead of just talking of ‘rational requirements’ and ‘rationality’, simpliciter. But the crucial point is that it’s compatible with this change of vocabulary to continue to think that the distinction between reasons and structural requirements is deep and important -in contrast to the many philosophers who have ignored it, equivocated on it, and so on. The change in vocabulary also preserves the possibility of asking many questions about the relationship (or lack of ) between reasons and structural rationality, and to translating Broome’s claims about the relationship of reasons and rationality simpliciter into those terms.
Finally, suppose I’m wrong that the issue between the reasonsresponsiveness view and the coherence view is terminological. Though the constraint that rationality supervenes on the mind won’t rule out the (best version of the) former view, perhaps there’s some other conceptual constraint that does. Note that the change in vocabulary I’m proposing wouldn’t preclude recognizing this. We would simply present this claim as the further claim that structural requirements exhaust the requirements of rationality as a whole. Indeed, I think it’s more perspicacious to present this claim as just what it is: a further claim -rather than using ‘requirements of rationality’, in effect, to refer to structural requirements by stipulation. So I think that however things shake out, we should adopt my proposed change in vocabulary, and this is what I’ll do in what follows. To have a single set of terminology, I’ll use the vocabulary of structural requirements and of structural rationality. But that of coherence requirements and of coherence would do just as well.
2. The threat of superfluousness
A different threat to the significance of the distinction between reasons and structural requirements comes from the thought that structural requirements might be superfluous. The idea here is that, as long as one responds appropriately to one’s reasons, this will guarantee that one obeys all the structural requirements.30 In that sense, structural requirements are superfluous, because they don’t introduce any further demands on one’s attitudes that one’s reasons don’t already impose. An example will help here. One might think that, whatever one’s evidence is, it could never both permit believing p and permit believing not-p. If that’s so, correctly responding to one ́s evidential reasons guarantees that one satisfies the (structural) requirement to avoid contradictory beliefs. The superfluousness charge aims to extend this general sort of challenge to other structural requirements.
Of the three threats I am exploring here, this is the only one that Broome directly discusses.31 He discusses it in the context of his aforementioned contention that rationality is not a matter of responding appropriately to reasons. Part of what Broome is claiming is that even if rationality did require responding to reasons in some sense, responding to reasons is not the only aspect of rationality.32 But if structural requirements are superfluous, there’s a sense in which that is not so. On such a view, once we have said what our reasons demand of us, we are done saying what rationality requires. Or, if rationality consists only in structural rationality, and does not require responsiveness to our reasons, then rationality as a whole is superfluous: we can capture all the demands on our attitudes by talking only about our reasons.
There’s a stronger and a weaker way of construing the threat of superfluousness. On the stronger version of the threat, showing that putative structural requirements are superfluous shows (or at least provides strong reason to think) that there in fact are no such structural requirements. Once these requirements are superfluous, we can simply eliminate them from our ontology.33 So, for example, it will be true that it’s never permissible to both believe p and believe not-p, but not because there’s a requirement banning this combination as such, but rather because, for any proposition and any evidential state, one’s evidence will forbid (at least) one of these two beliefs. On the weaker version of the threat, structural requirements might be real in spite of their superfluousness, but they are nevertheless uninteresting, since they don’t tell us anything to do anything that our reasons don’t already tell us to do (and fail to tell us to do some of the things that our reasons do tell us to do). They can, on this view, be fairly ignored by those interested in normativity.
I’m not sure I agree that, were structural requirements superfluous, they would be completely uninteresting. It might be that failures to conform with structural requirements represent a deeper kind of irrationality than failures to respond appropriately to reasons, or that there is a distinctive mode of criticism involved in charging someone with structural irrationality.34 Moreover, structural requirements may play a constitutive role in the attribution of mental states, or in the explanation and prediction of behavior, that reasons (or requirements to respond appropriately to reasons) do not.35 Still, it’s clearly at least something of a disappointment for those who want to draw attention to structural requirements if they turn out to be superfluous. I’ll mention two ways in which the superfluousness result might be blocked.
The first is offered by Broome himself, and attempts to exploit cases where one’s reasons are permissive.36 The trick is to find cases where one’s reasons permit one to have some attitude A, and permit one to have some attitude B, but having both attitude A and attitude B is structurally irrational.37 For example, it might be that it’s permitted by one’s evidence to have a credence for p anywhere in the range between 0.32 and 0.33, inclusive. It plausibly follows from this that it’s permitted by one’s evidence to have a credence for not-p anywhere in the range between 0.67 and 0.68, inclusive. Nevertheless, it would be structurally irrational to have credence 0.33 in p while also having credence 0.68 in not-p. Or, to take an example from the practical domain: it might be permissible, so far as one’s reasons go, either to intend to go the office or to intend to work from home, and thus also be permissible either to intend what you believe to be necessary means toward going to the office (say, asking to borrow the car), or to not intend those believed means. But it’s structurally irrational to intend to go to the office but fail to intend to ask to borrow the car (while believing that you need to ask to borrow the car in order to go the office).
There are, however, some complications here. In general, it doesn’t follow from its being permissible to Ф and its being permissible to Ψ that it’s permissible, even by the same set or kind of norms, to (Ф and Ψ). Thus, one’s reasons might permit two things individually, without permitting them jointly. For example, it might be that your reasons permit you to propose marriage to your partner tonight, and that your reasons also permit you to watch a gory horror movie with your partner tonight. But for all that, your reasons might not permit doing both: it might be that although each plan is an individually reasonable course of action, the two together are a very unreasonable course of action. This isn’t, it bears stressing, a matter of structural irrationality.38 On its own, there’s nothing structurally irrational about intending to propose and simultaneously intending to watch the gory horror movie. You might be structurally irrational if you also have certain further attitudes: say, believing that conditional on watching the horror movie, you shouldn’t propose. But it’s possible for there to be no structural irrationality here (you might wrongly think that it’s a great idea to propose while watching a gory horror movie), compatibly with its still being the case that it is in fact unreasonable to (intend to) do both. Your reasons forbid doing both jointly not because of the structural irrationality or incoherence of having both intentions, but rather because the reasons that speak in favor of each individual plan are undermined by pursuing the other one: the reason-generating benefits of proposing are unrealized if you do it while watching a gory horror movie, and the reason-generating benefits of watching a gory horror movie are unrealized if it is interrupted by a proposal.39
Having clarified that one’s reasons can forbid doing two things jointly even while permitting each individually, it’s now open to someone defending the superfluousness of structural requirements to claim that in the permissive cases where it is structurally irrational to have both attitudes, one’s reasons also forbid having both attitudes (though they permit each individually) -so that the structural requirement is in fact superfluous after all. For example, perhaps it’s unreasonable (and not just structurally irrational) to intend to go to the office, while not intending to ask to borrow the car. After all, one might say, the primary point of intentions is to help us get things done. The point of the intention to go to the office is (typically) in the actually going to the office; but if one doesn’t intend to ask to borrow the car, one won’t actually go the office, and so the reason-generating benefit of intending to go is unrealized. So perhaps, though taken individually, both intending to go to the office and not intending to ask to borrow the car are reasonable, jointly they are unreasonable (and not just structurally irrational). Admittedly, it’s a bit harder to see how to generalize this to doxastic permissive cases. But perhaps it can be done, and besides, doxastic permissive cases are more controversial than practical permissive cases. So it’s uncertain, I think, whether permissive cases really do save us from superfluousness. By this, I really do mean that it’s uncertain. Permissive cases are a good way to put some pressure on the claim that structural requirements are superfluous, but more work remains to be done to show that they defeat it.
Elsewhere, I’ve pursued a different way to undermine the claim that structural requirements are superfluous, namely, to argue that there are cases where reasons and structural requirements conflict with each other, in the sense that it’s impossible both to take the attitudes your reasons support and to satisfy the structural requirements.40 In such cases, taking the attitudes one’s reasons support would put one into violation of the structural requirements; a fortiori, taking the attitudes one’s reasons support does not suffice for satisfying the structural requirements. Thus, it directly follows from the possibility of such conflicts that the structural requirements are not superfluous. My particular strategy is to appeal to cases of misleading higher-order evidence, where one has misleading evidence either about what one’s evidence is, or about what it supports. I argue that in some such cases, one’s total evidence can support a doxastic attitude A, yet one’s total evidence also supports the belief that one’s total evidence does not support A. In such cases, the only way to have the attitudes one’s evidence (and thus, assuming no interfering non-evidential reasons, reasons) supports is to have attitude A, but also believe that one’s total evidence does not support A. And that, I claim, is structurally irrational. So, we have a conflict between reasons and structural requirements. I won’t rehearse my defense of this line of argument here, and I am not pretending that it is uncontroversial. I do think that, if it succeeds, it can be extended to the practical domain, giving us cases where one ought to (intend to) Ф, but also decisive reasons to believe that one ought not Ф - again, a combination that would make one structurally irrational.41
It’s worth mentioning a two other, perhaps simpler, ways in which we might hope to get conflicts between reasons and structural requirements. First, there are cases where having some structurally irrational combination of attitudes is beneficial. For example, some eccentric madman might offer you millions of dollars to violate a structural requirement (or threatens to do something awful to your family if you don’t). If the madman’s offer is a reason to have the attitudes that violate the structural requirement, then it seems like it should be strong enough to, in principle, be decisive; a conflict would then follow. Whether this is a successful example of a conflict case turns on whether there are so-called “wrong kinds of reasons” for attitudes that are nevertheless genuine reasons for those attitudes (as opposed to something else, such as wanting to have the attitude or trying to get yourself to have it).42
Secondly, if one thinks that it’s incoherent to have inconsistent beliefs, then preface paradox-style cases, where one’s evidence seems to support inconsistent beliefs (due to the agglomeration of risk of error), might be examples of conflicts between (evidential) reasons and structural requirements. Whether this is a successful example of a conflict case turns on whether it really is incoherent (even in preface-style cases) to have inconsistent beliefs.43
Overall, then, the two ways of trying to head off the superfluousness problem have different virtues and vices. The existence of permissive cases is relatively uncontroversial (at least in the practical domain), but the argument from permissive cases to the claim that structural requirements are not superfluous turns out not to be straightforward. By contrast, the existence of conflict cases is more controversial and takes some work to establish, but the argument from conflict cases to the claim that structural requirements are superfluous is straightforward. Either way, if I’m right, those who (like both Broome and I) want to stress the importance of the distinction between reasons and structural requirements have substantive work to do here.
3. The threat of disappearing subject-matter
As Broome clarifies several times, his use of ‘requirement’ is not such that a requirement is necessarily normative in a robust sense.44 In a robust sense of ‘normative’, something is normative if it helps determine what you ought to do. It follows from Broome’s discussion of the relationship between reasons and ‘ought’ that reasons are normative in this sense. But -and here we find another important element of the distinction between reasons and structural requirements- it is not guaranteed that structural requirements are normative in this sense. Broome says that he although believes that rational requirements (i.e., structural requirements) are normative, he “has no argument” for this claim.45
This may leave some readers puzzled. In what sense could a structural requirement be a requirement if it were not normative? But as Broome points out, the idea of a non-normative requirement can be made sense of. To use his example, the claim that Catholicism requires you to abstain from eating meat on Fridays is both intelligible and true even from the point of view of someone who thinks that we have no reason whatsoever to comply with the requirements of Catholicism.46 Similarly, we can talk about requirements of etiquette, of grammar, of Mafia morality, and so on, without committing ourselves to thinking of these requirements as normative.
However, there is a major disanalogy between all of these (potentially) non-normative requirements, on one hand, and structural requirements of rationality on the other. The difference is that every clear example of the former seems to be in some way fixed by facts that are (in a broad sense) conventional. For example, what 19th century British etiquette requires is simply a matter of what the conventional practices and beliefs of 19th century Brits are. By contrast, I want to suggest, we do not have clear examples of requirements that are both non-normative and non-conventional. Yet structural requirements of rationality do not seem to be conventional. So perhaps it is, after all, hard to make sense of the idea that structural requirements of rationality are non-normative.
This creates a puzzle even if one, like Broome, suspects that structural requirements are, ultimately, normative. For Broome still takes it to be a possibility worth taking seriously that structural requirements are not normative. But he never seems to take it to be a possibility worth taking seriously there simply are no genuine structural requirements. Assuming that structural requirements are not conventional, this combination of attitudes requires us to be able to at least make sense of the idea of a non-normative, but also nonconventional requirement.
The problem would go away if we took the specification of structural requirements to be essentially a matter of stipulation. Some philosophers do talk about ‘coherence’ this way. For example, epistemologists typically call any set of credences that do not obey the axioms of the probability calculus “probabilistically incoherent”. In doing so, they are just stipulatively using the term ‘incoherent’ to refer to any set of credences with those probabilities. Indeed, some of these philosophers go on to deny that probabilistic incoherence is in fact irrational, or that there is any genuine requirement on us to avoid it.47 Similarly, some may treat it as a matter of stipulation that any set of deductively inconsistent beliefs is incoherent. Broome, however, doesn’t treat the specification of structural requirements as a matter of stipulation. Chapters 9 and 10 of Rationality Through Reasoning are concerned with trying to determine, as a substantive matter, which putative structural requirements are genuine requirements of rationality, and which aren’t. For example, for Broome, it’s a substantive question for debate whether (e.g.) there is a genuine structural requirement of rationality forbidding all inconsistent beliefs (in fact, he thinks there is not).48
But, to reframe the problem, what does that substantive question come to? One might think the question of whether there is a genuine structural requirement forbidding inconsistent beliefs just is the question of whether we really ought (or at least, have some reason) to refrain from having inconsistent beliefs. But as we’ve just seen, for Broome, this can’t be so, since saying that something is a (genuine) structural requirement of rationality is not supposed to commit one to the claim that this requirement is normative. (He turns to the question of whether the requirements of rationality are normative only in Chapter 11, independently of and after asking which putative requirements of rationality are genuine requirements.) Broome’s method of uncovering requirements of rationality is to “appeal largely to our intuitions”.49 But our intuitions about what? It’s not clear how to have an intuition about whether a putative structural requirement is genuine without at least tacitly asking ourselves questions like “is this really a (putative) requirement that I ought to comply with?”.
Of course, you could ask yourself “is this a (putative) requirement that it would really be irrational to violate?”. But, as I’ve already argued in section I.1, the bare concept of rationality (as a status that supervenes on the mind) doesn’t automatically zero in on the relevant notion: given the right interpretation, the notion of a failure to respond to one’s reasons can be thought of as irrational, in a good sense that respects the supervenience constraint. Rather, we have to zero in on structural rationality, or coherence, specifically. But then the question is just: is this a requirement that it would be structurally irrational to violate?. And it’s hard to know how to have intuitions about that question without an answer to the question of what makes a putative requirement a requirement of structural irrationality.
The final threat to the depth of the distinction between reasons and structural requirements, then, is that when the two are cleaved apart so that calling something a structural requirement does not in itself commit one to saying there is reason to comply with it, the very substance of the subject-matter of a theory of structural requirements (that is, a theory of which structural requirements are genuine and which aren’t) may seem to disappear.50 Heading off this problem, I think, requires us to give a more substantive account of what structural rationality, or coherence, is.51 It’s not enough to have a purely formal account -that is, one that tells us only the form of structural requirements, for example that they govern combinations of attitudes, or are “wide-scope”. For this, at most, enables us to see which things are of the right form to be putative structural requirements.52 The putative requirement not to have inconsistent beliefs, for example, clearly has the right form. So do absurd putative requirements, like the claim that it’s rationally required not to both hope that p and simultaneously believe that if p then q. This purely formal account, then, doesn’t help us say which requirements are actually genuine and which aren’t. Especially if it is not necessarily normative, we need a substantive account of what structural rationality is in order to know what we are looking for when we accept or reject putative structural requirements as genuine.
II. Rational permissions and correct reasoning
For someone who wants to separate reasons and structural requirements sharply, as both Broome and I ultimately do, a further question is how the notion of correct reasoning, and of a rule of correct reasoning, fits in with these notions. A paradigm rule of correct reasoning is:
Modus Ponens Rule. From the belief p and the belief if p then q, derive the belief q.
It’s part of the concept of correct reasoning that reasoning can be correct even if the attitudes involved are not supported by reasons. For example, even if the belief p and the belief if p then q are both hopelessly unsupported by the evidence, it’s still true that if one were to reason from these beliefs to the belief q, one would be reasoning correctly. This might make us suspect that correct reasoning is tied, not to reasons, but to structural rationality.53 And that’s exactly what Broome claims. It’s this link that I want to put pressure on in what follows. I’ll be suggesting that correct reasoning and structural rationality are less intimately related than Broome thinks, and that we may thus need to make room for correct reasoning as a third sui generis normative notion, alongside reasoning and structural rationality, that cannot be analyzed in terms of the other two.
In attempting to tie correct reasoning to structural rationality, one might think that the place to start is with the claim that reasoning is correct just when it brings one into satisfaction of structural requirement of rationality.54 However, Broome argues (correctly) that this not right: it is neither necessary nor sufficient for reasoning to be correct that it bring one into satisfaction of a structural requirement.55 One reason that it isn’t necessary is that there are many instances of correct reasoning that are trivial extensions of one’s existing beliefs (for example, inferring from a belief that p to the disjunction of p and any other proposition). Such reasoning is correct when one does it, but there is no rational requirement to believe all these trivial extensions of one’s beliefs, and so the reasoning does not bring one into satisfaction of any structural requirement. One reason that it isn’t sufficient is that, if rationality requires you to hold some attitude given other attitudes you have (for example, if it requires you to intend to Ψ, given that you intend to Ф and believe that to Ф you must Ψ), and you currently lack the intention to Ψ, it’s always possible for some other, incorrect process of reasoning to happen to lead you to form the intention to Ψ. This would bring you into satisfaction of a structural requirement, but would not count as correct reasoning thereby.
Broome attempts to deal with this problem by tying correct reasoning not to structural requirements but to structural permissions. As he claims, “if it is correct to reason to some conclusion, that is because rationality permits you to reach that conclusion” (Broome (2013: 219). So correctness of reasoning is understood in terms of structural rationality: but not structural requirements; rather, structural permissions. According to Broome, (correct) reasoning will often bring one into satisfaction of rational requirements, and is one of our main ways of doing this.56 But equally, “in many cases, you commit no offense against rationality by failing to do a piece of reasoning that would have been correct had you done it”.57
Broome develops his account by appealing to the notion of a “basing permission”, which specifies that it is rationally permissible to base some particular attitude on some other attitude or attitudes.58 Though basing permissions make appeal to a notion of basing that has not featured in the structural requirements we have considered so far, they are still permissions of structural rationality: they pertain to the rationality of how one combines one’s attitudes.59
According to Broome, “each permission will determine a rule, and reasoning by correctly following that rule will be correct”.60 For instance, reasoning by the modus ponens rule is correct because it is always (structurally) rationally permissible to base a belief that q on believing that p and believing that if p then q. Here is the way Broome writes out the basing permission corresponding to the modus ponens rule (with the notation slightly adjusted):
Modus Ponens Permission. Rationality permits N that (N believes q at some time on the basis of believing p at some time and believing (if p then q) at some time).61
It’s crucial to understand that this is a supposed to be a schema that holds for all possible propositions p and q. Thus, it claims that it’s always structurally permissible to base a belief q on a belief p and a belief if p then q.62 This is a very strong claim. I will argue that, even confining ourselves to purely structural rationality (and so bracketing whether your beliefs in p and (if p then q) are themselves justified), there are cases where, in believing q on the basis of believing p and believing (if p then q), you are not structurally rational.
For a first example, suppose that you believe that it is raining, and believe that if it is raining then it is not raining. And suppose that on the basis of these two beliefs, you believe that it is not raining. This combination of states is structurally irrational, since it involves your believing that it is raining and believing that it is not raining; that is, believing contradictory propositions. For all that, though, the application of the modus ponens rule is still correct reasoning here.63 So we have a failure of correspondence between correct reasoning and structural rationality: there is a rule of correct reasoning for which there is no genuine corresponding basing permission.
In conversation, Broome suggested two lines of response to this objection. The first is this. As he sets out basing permissions in his book,64 the basing permission says that it’s (structurally) permissible to be in the following combination of states in this case: {believing that it’s raining, believing that if it’s raining then it’s not raining, believing that it’s not raining on the basis of the first two beliefs}. However, Broome denies the axiom of deontic logic that says that if p is permitted, and p entails q, then it follows that q is permitted. So, Broome says, he can deny that the basing permission entails that it’s permitted to be in the following combination of states: {believing that it’s raining, believing that it’s not raining}. I do not agree with Broome’s rejection of the relevant axiom here. Though he has independent purported counterexamples to this axiom and other closely related ones, I think these counterexamples can be easily dealt with by accepting a standard contextualist semantics for deontic modals.
But it would take me too far off-track to argue for this here.65 Fortunately, I have another response, which is that it is already an unacceptable result to say that any total state involving a belief that p and a belief that not-p is structurally rational. Even if Broome denies the relevant axiom of deontic logic, he is still committed to this unacceptable result.
Broome’s second line of response is different. It is to revise his statement of the relevant basing permission, so that it says the following: rationality permits that (if you believe that it’s raining, and you believe that if it’s raining then it’s not raining, then you believe that it’s not raining on the basis of the first two beliefs). So what’s permitted is not a conjunction but a (material) conditional. But once we switch to this reading of basing permissions, they become too weak to vindicate the purported connection between basing permissions and rules of reasoning. Consider the following permission:
Rationality permits of you that (if you believe that it’s raining, and you believe that it’s not raining, then you believe that pink elephants will invade China on the basis of your belief that it’s raining and your belief that it’s not raining).
If we read the conditional here as material, it is permissible to be such that this material conditional is true. Specifically, it’s permissible to make it true by making its antecedent false: by not having the contradictory beliefs in question. Of course, there’s also a way of making the material conditional true that is not structurally permissible, namely making both the antecedent and the consequent true. But no-one accepts the principle that if it’s permissible to Ф only if any way of Ф-ing is permissible. So this is no objection to the basing permission as stated. But clearly, this basing permission does not correspond to a rule of correct reasoning. It definitely isn’t correct reasoning to go from the belief that it’s raining and the belief that’s it not raining to the belief that pink elephants will invade China.
A better option for Broome might be simply to claim that it’s the basing on its own that is permissible. Both attempts at stating the basing permission just considered mention the state of basing the belief that it’s raining on the belief that it’s raining and the belief that if it’s raining, then it’s not raining. So perhaps we can say that this state on its own is permissible. But even this doesn’t strike me as the right result. There is something distinctively (structurally) irrational about basing a belief that it’s not raining, in part, on the belief that it is raining. This is not a structurally rational way to be.
So much for the first example of a case where it’s not structurally rational to base a belief q on a belief p and a belief (if p then q). For a second example, consider a long chain of modus ponens deductions, like so:66
P1
If P1, then P2
So, P2
If P2, then P3
So, P3
[…]
So, Pn-1
If Pn-1, then Pn
So, Pn
Suppose that, although the agent believes each (underived) premise in this long chain of deductions (i.e. both P1 and each conditional premise), there is a slight risk of error associated with each premise. These risks aggregate so that eventually the probability of the conclusion, Pn, is low, and as such doesn’t warrant belief. And let us suppose that the agent herself recognizes these risks of error, and even recognizes how they aggregate. Would she be even structurally rational to base belief on the conclusion of the basis of these premises? I think not.
But consider what would be so if it were always structurally rational to base a belief q on a belief p and a belief (if p then q). It would be structurally rational for her to believe P2 on the basis of P1 and (if P1, then P2). And it would be structurally rational for her to believe P3 on the basis of P2 and (if P2, then P3). And so on, until it would be structurally rational for her to believe Pn on the basis of Pn-1 and (if Pn-1, then Pn). There would be no structural irrationality in the chain. But that isn’t correct. So it cannot be that it is always structurally rational to base a belief q on a belief p and a belief (if p then q).
I therefore conclude that Broome’s basing permissions are too strong to be plausible. There are rules of correct reasoning (such as the modus ponens rule) with no corresponding basing permissions. And so the purported connection between rules of correct reasoning and basing permissions fails.
My tentative suggestion is that rules of correct reasoning are a sui generis notion that can’t be understood either in terms of reasons or in terms of structural rationality.67 Thus, the kind of separation between these latter two notions that Broome so importantly draws our attention needs to be effected again between each of them and the notion of a rule of correct reasoning. It may be hard to hear claims about correct reasoning in a way that doesn’t involve any claim about structural rationality. But then again, many philosophers have found it hard to hear claims about (structural) rationality in a way that doesn’t involve any claim about reasons. We need to expand our conceptual horizons here.
If I’m right about this, it’s a problem for Broome’s project of intimately tying reasoning and rationality together, reflected in the title of his book. But in a broader sense, I hope this conclusion is a fitting tribute to somehow who’s done so much throughout his career to insist that philosophers of normativity not treat different normative notions as interchangeable, and to show us why separating them matters. In my view, it turns out that we need one more such separation.