I. SOME SHARED PREMISES

Alexander and I begin from premises that are strikingly similar to the premises Shapiro sets out in Legality. What we call the “Circumstances of Law”Footnote ¹¹ are much the same as Shapiro's “Circumstances of Legality.”Footnote ¹² We describe a community whose members agree on the general principles of morality but disagree about specific applications of those principles;Footnote ¹³ likewise, Shapiro imagines a community of morally virtuous persons who nevertheless face moral “doubts and disagreements” that cannot effectively be resolved by nonlegal means.Footnote ¹⁴ Accordingly, the fundamental aim of law (or legal planning) is the same for us and for Shapiro: our lawmaker, Lex, is chosen by general accord to provide authoritative settlement of moral controversy;Footnote ¹⁵ Shapiro's lawmaker, also called Lex, is vested by general accord with authority to make plans that “rectify the moral deficiencies associated with the circumstances of legality.”Footnote ¹⁶ Authoritative settlement, we argue, improves the moral prospects of the community by coordinating individual conduct, providing expertise, and increasing the efficiency of moral decision-making;Footnote ¹⁷ Shapiro, too, relies on the need for coordination and efficiency as well as “an ethic of mutual respect” arising from shared agency.Footnote ¹⁸

Alexander and I identify the principal means for settling moral controversy within a society as legal rules; Shapiro helpfully redescribes laws as plans—a term that captures the cumulative, interconnected, and open-ended nature of a system of laws.Footnote ¹⁹ The essential characteristics we ascribe to rules, however, are very similar to the characteristics Shapiro ascribes to legal plans. In particular, we maintain that effective legal rules must be general in application and also reasonably determinate in form. Addressing determinacy, we say that “if uncertainty and disagreement about how moral principles apply in concrete situations are what give rise to the need for Lex and Lex's rules, Lex's rules are responsive to this need only if they themselves do not engender uncertainty and disagreement about their concrete applications.”Footnote ²⁰ Shapiro makes the point perhaps more elegantly in the form of a principle he calls the “Simple Logic of Planning” (“SLOP”): if a legal plan is to settle moral controversy, “[t]he existence and content of a plan cannot be determined by facts whose existence the plan aims to settle.”Footnote ²¹

For Shapiro, SLOP prohibits moral content in legal plans. In contrast, we take the view that the use of moral or evaluative terms in legal rules is acceptable as long as the terms have an accepted meaning within the community.Footnote ²² Although this difference may have some theoretical significance, the underlying idea is the same: if rules or plans are to settle moral questions effectively, those who act under them must be able to follow their prescriptions without further deliberation on the questions the rules or plans are designed to settle.

Another point of agreement is that we, like Shapiro (and Hart), recognize the importance of a hierarchy of rules (or plans). A legal system must be founded on a master rule that exists by virtue of general recognition and renders the system authoritative. We also agree that an effective legal system needs secondary rules that order the process of making and applying rules and plans.Footnote ²³

Finally, Shapiro appears to share our sense of when a legal system is worthy of acceptance by those who are subject to its laws. I note above that rules (or plans) are designed to settle controversy by translating agreed moral principles into determinate prescriptions for action. Determinate prescriptions, if generally accepted and obeyed, can improve on unguided decision-making through the effects of coordination, decision-making efficiency, and, depending on the wisdom of the authority that issues them, comparative expertise. At the same time, the determinacy of rules makes them overinclusive and underinclusive in application: they will sometimes dictate outcomes that are incorrect when judged by the underlying principles on which they are based. Thus, in the course of reducing error, rules introduce errors of another kind.Footnote ²⁴

This means that the desirability of rules (or plans) is contingent: they count as good rules worthy of acceptance when those who follow them consistently will commit fewer errors or errors of lesser magnitude than they would commit in the course of unguided practical reasoning. For ease of reference, I give this criterion for rule acceptance a name (the Net Error Reduction Principle) and an acronym (NERP).

Notice that NERP requires the agent to accept R as a rule for all cases. If the agent were permitted to accept R provisionally, as a rule to be followed only when the agent judged that the outcome the rule calls for was correct, the rule would exercise no constraint on the agent's reasoning and yield no benefits. Thus the agent must endorse the rule as a rule to be followed in all cases that fall within its terms.

NERP also implies that the master plan of a legal system is worthy of acceptance when the overall effect of its present and future rules, taken together, is to reduce the level of error by agents. As a strict legal positivist, Shapiro would say that the question of whether a legal system succeeds in reducing the overall level of error has no effect on its status as a system of law. Nevertheless, as a criterion for external evaluation, including deliberation by potential participants in the system about whether to accept or reject the system's master plan, NERP is consistent with the primary moral aim of law.

II. DIFFERING CONCLUSIONS

Thus Alexander and I agree with Shapiro about the aim and benefits of law and, for the most part, about what is needed to obtain those benefits. At this point, however, our view of law diverges sharply from Shapiro's. For Shapiro, once a master plan for law wins general acceptance, the feasibility of legal planning falls into place.Footnote ²⁵ The capacity to plan is part of human nature, and compliance with plans follows from the fact of acceptance. Those who accept must obey, because commitment to the system makes compliance the only rational option. Those who do not accept can be brought into line through sanctions for disobedience. The main challenge, as Shapiro sees it, is how to interpret the plans made by various officials who play a part in the planning process. Shapiro proposes that interpretive methodology should vary according to the level of trust that those who accept the master plan repose in the relevant officials.Footnote ²⁶

In our view, the central problem in legal theory is how law can be effective in guiding action. Summarizing our position, we agree that it is rational for agents to endorse the system of legal rules (or legal planning) and by doing so to endorse all rules that issue from it, intending to follow them in all cases. But it does not follow for us that it is rational—let alone rationally compelled—for agents to conform to the rules in particular cases when those cases arise. Rules translate moral principles into concrete directions for classes of cases. They are accordingly different from and simpler than the moral principles that lie behind them and will sometimes dictate a morally wrong result. When agents consider rules in the abstract, the balance of reasons requires them to accept any rule that meets the test of NERP and to accept it as a rule to be followed in all cases. But when an agent must apply the rule to a set of facts, and the rule, so applied, prescribes an outcome that appears wrong according to the underlying principles the rule is designed to clarify and promote, the balance of reasons requires the agent to disobey. There is thus a gap between what the agent ought to endorse (an exceptionless rule) and what the agent ought to do (make an exception).

If agents were always correct in their assessment of what action shared moral values require in a particular case, this gap might not be of serious concern: an agent's decision not to follow the rule would be a moral improvement over the outcome indicated by the rule. But one of the principal reasons that rules (or plans) are necessary is that agents are often mistaken in their assessment of what to do, either because they act on insufficient evidence or because they are unable, without the aid of reliable rules, to coordinate their decisions with those of others. Therefore, it remains the case that the rule-maker ought to demand full compliance in all cases and that the agent ought to accept the rule on these terms.

A partial solution is what Frederick Schauer calls rule-sensitive particularism.Footnote ²⁷ Rule-sensitive particularism treats the decision whether to follow a rule as an all-things-considered decision based on current evidence about reasons for action but counts the benefits of settlement as reasons to comply with the rule. At least in some circumstances, rule-sensitive particularism will lead an agent who might otherwise defect from the rule to follow it instead. Specifically, the agent may reason that his own defection will lead others who previously were disposed to follow the rule to reconsider their stance of unthinking obedience. The agent's disobedience, in other words, will set a potentially harmful example. If so, the others may decide to follow their judgment and disobey in cases in which their judgment is wrong. This prospect, in turn, becomes a reason for the original agent to obey the rule.

The difficulty is that under rule-sensitive particularism, the potential effect of a bad example is not a conclusive reason to obey: the agent may believe that the moral error the agent will commit by following the rule more than offsets any negative consequences of defection and so may defect after all. And again, the agent's calculation may be wrong. Thus rule-sensitive particularism fails to close the gap between the exceptionless rule the agent has reason to endorse and the action the agent has reason to take.Footnote ²⁸

III. RATIONALITY

Shapiro believes that those who accept a system of legal plans are practically committed to following the rules the plan imposes on them: rationality entails compliance.Footnote ²⁹ Alexander and I believe that an agent can have undefeated reasons to accept a rule and commit to follow it in all future cases and yet also have undefeated reasons to disobey the rule in some of those cases when they arise. What explains this divergence?

Evidently we are working under different understandings of rationality. We can probably agree that rationality involves an appropriate response to applicable reasons for action or decision. For Alexander and me, this means an appropriate response to the full set of reasons that apply to the agent at the time the agent acts. Among possible acts—such as following a rule or not following a rule—it is rational for the agent to do what the weight of current reasons for action recommends.

Shapiro, in contrast, takes a view of rationality that is both broader and narrower than ours. It is broader in a temporal sense: rationality is a property of agents that cuts across time.Footnote ³⁰ At the same time, Shapiro limits his attention to instrumental rationality: rationality requires agents to perform those actions that will best promote the agents’ ends.Footnote ³¹ Applying Shapiro's temporally expansive notion of instrumental rationality to law, an agent may determine that given the fallibility of unguided reasoning, the best way to accomplish his or her ends is to commit to follow a legal plan and to carry through on the commitment in the future. In committing to follow the rule, the agent is “strategically interacting with another agent—his later self,” with the terms of interaction fixed by the present self.Footnote ³² This means that from the time of commitment (T_C), the agent's decision is binding on future selves. The effect is to eliminate the option of future noncompliance: the whole self has committed and the future self has no independent choice in the matter. Departing from the rule's prescription is simply not “feasible.”Footnote ³³

The term “feasible” may seem odd in this context because breaking a rule is surely something the agent can do. Shapiro, however, adopts a special notion of feasibility that he ties to reasons for action.Footnote ³⁴ An action is feasible, in Shapiro's sense, if the agent can perform it for a reason; conversely, if the agent has no reason to perform the action, it is not rationally available to him. Applying this notion of feasibility to the problem of rule-following, Shapiro draws a distinction between “deliberative reasons,” which are reasons to decide upon an act (including the act of committing to a rule), and “implementation reasons,” which simply dictate that the agent should do what is necessary to execute a previous decision. Shapiro's key point is that an agent who has committed to follow a rule can no longer choose noncompliance because there is no choice left to be made. All decisions to be made were made at T_C, based on a full consideration of “deliberative reasons.” Now, at the time of action (T_A), deliberative reasons are no longer in play, and the only reason applicable to the agent is the implementation reason provided by the rule itself. It follows that the only feasible act—the only act the agent can perform for an extant reason—is to comply. This feature of rationality gives agents the power to bind themselves for the future through volitional acts and gives law the power to govern. Shapiro also maintains that rational agents must have this power in order to negotiate the outcome that is best for all present and future selves.Footnote ³⁵

Shapiro's argument for the “causal” power of agents to bind their future selves has the air of wishful thinking: if the power is important enough to realizing one's ends, then it must exist in some literal sense.Footnote ³⁶ Yet I do not think that Shapiro intends to say that agents who have committed to a plan actually lack the capacity to defect. Instead, I take him to mean that a rational agent will carry through on prior commitments. That is, an agent who holds himself or herself to the standards of rationality will carry through, and an agent who does not carry through is subject to criticism for acting irrationally.Footnote ³⁷

Viewed as an argument about rational deliberation, Shapiro's argument is difficult to defeat on its own terms because it takes off the table anything one might say about deliberation at T_A. At this point, deliberation has ceased, and implementation is the only option. In the discussion that follows, I first point out some potential difficulties that are internal to Shapiro's account. Then, in the final section of the essay, I pose a broader challenge to Shapiro's understanding of rationality.

The most significant problem that arises within Shapiro's account of rational commitment to legal plans is whether and when the agent can escape the effects of commitment. Shapiro suggests at several points that the effect of commitment to a plan on a rational agent is less than absolute. “[C]hoosing a plan,” Shapiro says, “does not set it in stone. Reconsideration is rational when, but only when, there is good enough reason for it.”Footnote ³⁸ What counts as a good-enough reason is not precisely defined, but Shapiro suggests that the reason for reconsideration must be “compelling:” “it would defeat the purpose of having plans if [the agent] were to review their wisdom without an otherwise compelling reason to do so.”

An initial question is whether Shapiro really means what he says about reconsideration of plans, at least in the context of legal plans. He evidently does not want to suggest that all laws are absolutely binding on those who accept the basic plan for a system of law. But he equivocates about whether exceptions to the binding effect of laws must be built in to the laws themselves or instead stand as qualifications to the commitment that follows from an agent's acceptance of the basic plan. For example, he states that settlement by means of legal planning:

This passage can be read to mean that the defeasibility of commitment is not a feature of rationality but a contingent feature of the law itself. Legal plans may or may not permit agents to reconsider their commitment in exceptional circumstances; when they do not, an agent's acceptance of a legal plan continues to foreclose consideration of any and all deliberative reasons for failing to comply with its terms. Alternatively, the passage can be read to mean either that when an agent commits to follow the basic plan, the agent's former self adds a caveat for problem cases that may arise in the future, or that rationality itself incorporates a caveat for problem cases.

The possibility of absolute commitment is not a problem for Alexander and me because we do not believe in commitment.Footnote ⁴⁰ Shapiro does believe in commitment and so must deal with its effects. For one who, like Shapiro, accepts that the prior self can bind the future self, it would seem to follow that the prior self can bind the future self without exception. If Shapiro is willing to bite this bullet, fine. But if not—if he means to say that some sort of escape route from commitment is built into rationality itself—he must spell out in more detail when and why the agent is free to defect.

I assume that the reason for providing an escape route is that it is not rational for the former self to foreclose all consideration by the future self of deliberative reasons for action, no matter how pressing. If this is what Shapiro intends, his first task is to define the type of reason a rational former self would treat as a compelling reason for future defection from a plan. To make the problem concrete, suppose that a group of library patrons wants to maximize their enjoyment of reading and their ability to study, so they appoint a director to make a plan.Footnote ⁴¹ The director determines that quiet is important to study and enjoyment, that speech within the library is less important, and that if some patrons speak, so will other patrons who would otherwise prefer a norm of silence. For these reasons, the director adopts a rule forbidding speech in general terms: “No talking in the library.”

Assuming that Shapiro intends that the “compelling reasons” exception is a built-in qualification on commitment, one way to understand the exception is that patrons have reason to reconsider their commitment to the plan when the plan is not (or has ceased to be) instrumentally justified as a rule. If circumstances have changed so that the plan no longer satisfies NERP, those who have committed to follow it are released from rational obligation. Perhaps the plan for the library worked well for some time, but it is now 2020, and the library no longer contains books (it contains rows of computers that display digital material and are equipped with noise-blocking headphones). As a result, the overall benefits of consistent application of the rule no longer outweigh the benefits of speech, and patrons are no longer bound to comply. This version of the compelling-reason exception, however, is quite narrow, particularly if the NERP test is applied not at the level of specific laws but at the level of the master plan for the legal system.Footnote ⁴² Shapiro seems to have in mind a more general exception that would apply to justified as well as unjustified rules.Footnote ⁴³

Another way to interpret the compelling-reason exception is as a safety valve for extreme cases in which compliance could have disastrous consequences. A library patron, noticing that the person seated next to him has suffered a heart attack, decides to call 911. This is certainly a worthy reason to disregard the no-talking rule, and perhaps the “disaster” category can be cabined in a reasonable way. It does not seem necessary, however, to invoke an exception to the patron's commitment to follow the rule in order to avoid truly disastrous results. The same result can be reached by interpretation, at least under an interpretive theory in which the meaning of a rule turns on the rule-maker's intent.Footnote ⁴⁴ One can easily say, without engaging in creative or purposive interpretation, that a library director who issues a rule “No talking in the library,” with the general aim of preserving a quiet environment for study, does not intend to forbid lifesaving phone calls. Again, however, I doubt that this understanding of compelling reasons for defection would satisfy Shapiro: he seems to anticipate that the former self would give the later self a somewhat longer leash than life-threatening emergency.

The more difficult case, and the case Shapiro probably has in mind when he speaks of compelling reasons to reconsider, is one in which the rule is well justified as a rule, and no dire emergency has arisen, but the agent's reasons to defect strongly outweigh reasons to honor a commitment to follow the plan. A patron learns while browsing the news on a cell phone that the president of the United States has just been assassinated. This is not a personal emergency, and announcing the event will create a disturbance, but the patron may judge that most others in the room would wish to know. So she speaks. Arguably, at least, this patron has a compelling reason both to reevaluate her reasons for accepting the rule and to violate the no-talking rule by announcing what has occurred. I am not sure whether Shapiro would accept this as a case for reconsideration of commitment, but he might: something unusual has happened that significantly alters the balance of the reasons on which the rule is based. The patron is not simply recalculating but responding to an unforeseen contrary reason of considerable strength.Footnote ⁴⁵

Another preliminary issue is whether the existence of a compelling reason to defect from commitment is an objective question or a question to be answered from the internal perspective of the agent. I am already assuming that Shapiro is applying a standard of rational deliberation to which agents will attempt to conform and by which they can be criticized if they fail to conform. Within this framework, it seems appropriate that the exception for compelling reasons should at least track the evidence available to the agent. Whether it should also allow for idiosyncrasies in the agent's evaluation of evidence is not so clear.Footnote ⁴⁶

I believe the best reading of Shapiro's comments about exceptions to commitment is that agents are permitted to defect when they believe or reasonably believe that the reasons to defect strongly outweigh the reasons to obey (including the value of binding commitment). Permission to defect derives from the standards of instrumental rationality, which dictate that the self who makes a commitment can impose a substantial obligation of compliance but not an absolute obligation of compliance on future selves. Given these assumptions, the most serious problem Shapiro faces is that an exception that allows agents to defect from commitment in circumstances not authorized by the rule itself is hard to reconcile with Shapiro's explanation of the causal force of commitment. In particular, an exception of this kind appears to undermine the distinction Shapiro draws between “deliberative” reasons and “implementation” reasons. Shapiro asserts that at the point of application of a plan (T_A), there is no decision open to the agent; therefore deliberative reasons are no longer in play. Yet, if the agent believes that there may be reasons to defect in a particular case, it seems that the agent must engage with deliberative reasons in order to determine whether those reasons are strong enough to support a release from commitment.

This is true under any of the various definitions of “compelling” reasons discussed above. Reconsideration of the rule or plan under NERP requires the agent to review the set of deliberative reasons supporting the plan as well as the errors the plan may cause in application. Reconsideration of commitment in case of emergency requires the agent at least to assess the deliberative reasons for avoiding disaster and probably also to compare them to the deliberative reasons supporting adherence to the plan. Reconsideration of commitment when some new application of a justified rule or plan creates a strong reason to depart from the plan requires the agent to consider both the deliberative reasons underlying the rules and the deliberative reasons for disobedience in order to determine whether the reasons for disobedience are comparatively strong. Thus, unless Shapiro takes the strict view that only the law itself can determine when reconsideration is allowed, he must concede that some sort of decision involving deliberation goes on at the point of applying a plan.

Further, even if Shapiro limits defection to cases in which authorization to reconsider for compelling reasons is written into the plan, there are questions about how a vague exception of this kind can remain stable. Suppose the stated exception tracks the third possibility described above: agents have a compelling reason to depart from the plan if the circumstances of particular case generate a strong reason to do so. The difficulty is that agents will often believe that they have strong reason to defect, and they will sometimes be wrong. Further, even a correct application of the exception can rapidly undermine the coordination effects of a rule. For example, consider the rule “No talking in the library,” now amended to read “No talking in the library without a compelling reason to speak.” Patron A sees what appears to be a heart attack, determines that he has a compelling reason to speak, and phones 911. Patron B receives a call from home, where a babysitter is watching her child, and considers whether she has a compelling reason to speak. B's reason to speak may be less compelling than A's reason to speak, but at this point the threshold is lower because there is already some noise, or a recent history of noise, in the library. If the pattern continues, the level of cell phone chatter may reach a point at which almost any reason to answer a call looks compelling when compared to potential harm to the rule's value in preserving an environment conducive to study.

IV. RATIONALITY PLUS

Suppose the problems I have raised so far can be ironed out. There remains a basic divide between Shapiro's understanding of rationality and the understanding of rationality that leads Alexander and me to question the ability of rules to constrain conduct. Shapiro views constraints put in place by an agent's former self as limits on the reasons for action available to the agent's future selves, while Alexander and I believe that agents can, and as a matter of rationally should, respond to all reasons that apply to them at the time of action.Footnote ⁴⁷ In our view, the final decision to act is always a decision, and there is no mechanism by which prior intentions can curtail deliberation about that decision. I refer to Shapiro's view as the model of constraint,Footnote ⁴⁸ and to Alexander's and my view as the model of full deliberation.

Shapiro's model of constraint has the advantage of enabling human agents to adhere without the aid of psychological distortions to plans that will advance their ends more effectively than unconstrained decision-making at the time of action. If true, this is good news: reason itself can maximize success in pursuit of one's ends. I argue, however, that the model of full deliberation gives a better picture of rationality, because it acknowledges not only standards of instrumental rationality but also standards governing the broader process of deliberative decision-making. In particular, full deliberation takes account of epistemic responsibility, a consideration that Shapiro does not address.

I begin by noting a move that Shapiro does not make. It can be tempting to think that commitments, promises, and the like create additional reasons to honor the commitment or promise over and above whatever initial reasons the agent had to do the planned or promised act.Footnote ⁴⁹ In my view, this is a mistake.Footnote ⁵⁰ Suppose, for example, that the balance of reasons narrowly favors that I work tonight rather than watch a movie. I then make a plan to rent a DVD and commit to following through on the plan. It is difficult to see how my commitment can add anything to my reasons for watching a movie, which continue to be outweighed by my reasons to work.Footnote ⁵¹

Shapiro's analysis is consistent with this observation. Shapiro claims that commitment generates an “implementation” reason to follow through on plans, but, as I interpret his argument, the implementation reason does not add to the balance of reasons favoring conformity to a particular plan. Instead, the new implementation reason displaces the set of deliberative reasons that previously favored the plan. If these reasons were outweighed initially by the reasons against following the plan, the implementation reason that arises from commitment would also be outweighed. Nothing has changed, and it is not rational for the agent to follow the plan.Footnote ⁵²

Shapiro's mistake lies instead in his failure to consider the rationality of beliefs. An agent facing the question whether to follow or defect from a plan or rule must form a belief about which course of action the agent ought to pursue. Perhaps discussion of belief formation simply shifts the original debate from the level of reasons for action to the level of grounds for belief. Thus Shapiro might say that an agent who has committed to follow a rule will believe that he or she ought to follow the rule given that deliberative reasons to the contrary are not available. Alexander and I would say that the agent should form whatever belief the full array of reasons recommends. We remain at an impasse. Arguably, however, focusing on the agent's beliefs changes the picture of rule-following in two ways.

First, standards of epistemic responsibility may not correspond to standards of instrumental rationality. According to traditional accounts of epistemic responsibility, beliefs are justified when they are supported by the evidence available to the agent.Footnote ⁵³ The traditional view has met with some dissent, but most would agree that the immediate purpose of epistemic standards is not to enable the agent to advance his or her ends but to enable the agent to acquire true beliefs and avoid false beliefs. Consequently, standards for justified belief will be different from and may not always be consistent with standards for instrumental rationality.

Along these lines, Thomas Kelly makes a strong argument that the fact that an agent's holding a certain belief will serve the agent's ends cannot justify the agent in adopting the belief. Kelly assumes that it is possible to affect one's own beliefs, for example by seeking out evidence that confirms a particular belief. Yet he argues, plausibly I think, that a belief acquired in this way is based not on the instrumental value of the belief but on the new evidence the agent acquires in support of the belief.Footnote ⁵⁴ Kelly gives content to the “based-on” relation by asking whether the agent would continue to hold the belief if its instrumental value decreased. Thus, if Pascal thinks it may be useful to believe in God, he may seek out evidence that supports belief in God and eventually come to believe in God. But if he later changes his mind about the advantages of believing in God, he will not simply stop believing in God; to stop believing, he will need to seek contrary evidence that actually alters his belief. Accordingly, Pascal's belief in God is not based on the benefits of holding the belief. Kelly concludes that “[w]ith respect to beliefs, rationality just is epistemic rationality”Footnote ⁵⁵—which he takes to mean responsiveness to evidence rather than responsiveness to consequences.Footnote ⁵⁶

Assuming that standards for justified belief are different from standards for instrumental rationality, as Kelly's argument indicates, they are also naturally prior to standards for instrumental rationality. If an agent fails to conform to standards of belief formation that are reasonably calculated to promote true belief, deliberation about the relations of means to ends is unlikely to result in practical success. Thus, to put practical instrumental reasoning on a firm footing, the agent must first form relevant beliefs in an appropriate way. For present purposes, I am assuming, in keeping with traditional views of epistemic responsibility, that this means the agent's beliefs are supported by the agent's evidence.

The second way in which focusing on epistemic responsibility alters the debate over rule-following is that it illuminates the distinction between general and particular responses to plans or rules. When an agent deliberates about whether to commit to follow a rule, the agent considers the rule in the abstract as it will apply to the full set of cases that may arise over time. When the question is whether to follow the rule, the agent considers only one application of the rule. At this point, the array of evidence available to the agent as well as the relevance of individual items of evidence is different from what it was at the earlier time, when the agent committed to follow the rule.

To see this, suppose again that the director of the library where agent A is a frequent patron has issued a rule, R (“No talking in the library”). Implicit in rule R is the further instruction that R is to be applied in all cases covered by its terms. At time T_C, when A considers whether to accept this rule, the only relevant evidence—that is, the only evidence that could make a difference to A's decision—is evidence that bears on R's long-term effects. For example, suppose A knows that A and other library patrons prefer quiet to speech within the library; that library patrons are unable to coordinate their behavior spontaneously; that A's reasoning and the reasoning of other patrons are fallible; and that the time A can devote to future deliberation is limited. A also knows that R is imperfect: in some future cases, speech will be more important than both silence and the settlement benefits that may be lost through defection. The errors R may generate, however, matter only insofar as they affect R's satisfaction of NERP. If NERP is satisfied, A knows that despite some errors, A will do better over the long run by following R consistently than by following her own judgment. Based on this evidence, A concludes that she can best advance her ends by accepting rule R and committing to follow it in all cases that come within its terms. This conclusion, it is fair to assume, is well supported by A's evidence.

At time T_A, A receives a phone call that appears to be from her babysitter. A must (quickly) form a belief about whether she ought to answer the call immediately or follow the rule and call back from another location. A now has new evidence: she knows that she has received a phone call, believes the phone call may be urgent, and believes she can handle it quickly. After glancing around the room, she perceives that there is only one other patron present, which indicates that a quick violation of R would do minimal damage to the coordination powers of the rule. So A concludes that following R would be an error.

In a sense, A's evidence is not new—or at least not unanticipated—because A has known from the outset that following R in every case would cause some errors. A also knows at T_A what she knew at T_C about the long-term benefits of following R. But the change in her vantage point has altered the weight of this evidence. The long-term benefits of R are pertinent, but only in comparison to the particular evidence of error A now has (a call from the babysitter, etc.) and only to the extent that the particular violation A is considering will undermine the settlement value of R. In other words, NERP is not relevant to the question whether, in this case, the error of following R exceeds the error of violating R. A's commitment to follow R continues to be instrumentally rational, but if A concludes that the particular evidence before her favors a violation of R, she cannot believe, consistent with standards of epistemic responsibility, that she ought to follow R.

At this point, Shapiro is likely to say that A's beliefs, like A's actions, should conform to the model of constraint: because A has committed to follow R, R now dictates what A should believe about what she ought to do. Following R is instrumentally rational; holding the belief that she ought to follow R will enable A to follow R; therefore holding the belief that she ought to follow R will advance A's ends; therefore A should believe that she ought to follow R. But this takes us back to Pascal's belief in God: from the fact that A can advance her ends over time by believing that she ought to comply with R, it does not necessarily follow that A is epistemically justified in believing that she ought to comply with R. As Kelly shows, justification for belief depends on evidence, not consequences.

Accordingly, I do not think that Shapiro can avoid the problem of epistemic responsibility at T_A. When an agent faces a particular application of a plan or rule, the difference in the agent's vantage point alters both the evidence available to the agent and the importance of evidence about the long-term benefits of the plan or rule. In the library example, it is hard to see how A, as an epistemically responsible agent, can ignore the evidence that following R may lead her to err. If, taking account of this evidence, A comes to believe that she ought not follow R in this case, her belief may be wrong but it will be justified. If, instead, she forms the belief that she ought to follow R in this case because she has committed to follow R, and acts on her belief, her act of rule-following may be instrumentally rational but the belief on which it was based will not be justified.

My own sense, at least, is that the second of these—lack of justified belief—represents a more serious flaw in the overall process of A's reasoning. Instrumental reasoning that rests on irresponsible beliefs should not count as rational, no matter how useful or even essential it is to successful management of social life. Consequently, I adhere to the position Alexander and I have taken—that although it may be eminently rational to endorse a rule as a rule to be followed in all cases, it is not always rational to follow it in fact.

If I am right that epistemic standards prevent agents such as A from following rules, this is not just a complication for Shapiro's model of constraint but a challenge to the core idea of rational commitment. If A deliberates carefully at time T_C, A will understand that she cannot form a justified belief that she ought to follow R when the evidence available to her in a particular case indicates that following R would be an error. Accordingly, she will understand that in some future cases she will not be able to follow R at T_A, despite her commitment at T_C.Footnote ⁵⁷ This makes the commitment itself futile and therefore irrational by Shapiro's own standards of instrumental rationality.