The Possibility of Practical Reason, 2nd Edition

Behavior, Activity, Action

Philosophers of action have traditionally defined their topic by quoting a bit of Wittgensteinian arithmetic: “What is left over if I subtract the fact that my arm goes up from the fact that I raise my arm?”[2] The difference between my arm’s rising and my raising it is supposed to illustrate the difference between a mere occurrence involving my body and an action of mine. And the difference between mere occurrences and actions is what the philosophy of action seeks to explain.

Yet there is reason to doubt whether Wittgenstein’s computation has a unique result. As Harry Frankfurt has pointed out, my raising an arm may be something less than an action:[3]

Frankfurt’s distinction between action and mere activity reveals a potential ambiguity in the above quotation from Wittgenstein. “The fact that I raise my arm” can denote an instance of action, such as my signaling a request to be recognized by the chair of a meeting; but the same phrase can also denote an instance of mere activity, such as my idly and inattentively—perhaps even unwittingly—scratching my head while engrossed in a book. When the fact that my arm goes up is subtracted from something called “the fact that I raise my arm”, what is left will depend on whether the minuend is a case of action or mere activity.

Unfortunately, most philosophy of action is premised on the mistaken belief pointed out by Frankfurt, that human events can be divided without remainder into actions and mere happenings. The result is that the prevailing theory of action neglects the difference between action and activity.[4]

This difference is also illustrated by behaviors that call for psychoanalytic explanation.[5] Consider an example from Freud’s Psychopathology of Everyday Life:[6]

This explanation is simpler than many of Freud’s, in that it portrays his action as a realistically chosen means to a desired end, rather than a symbolic wish-fulfillment or the enactment of a phantasy. The agent wanted to destroy the inkstand so as to make way for his sister to give him a new one; and his desire to destroy the inkstand moved him to brush its cover onto the floor, thereby destroying it.

Freud’s point about bungled actions is that they are no accidents: they serve an intention or purpose. Because there was a purpose for which the agent brushed the inkstand’s cover onto the floor, his doing so cannot be classified as something that merely happened to him. He didn’t just suffer or undergo this movement of his hand; he actively performed it.

Nevertheless, Freud acknowledges that a bungled action somehow differs from a normal attempt to accomplish the same purpose with the same bodily movement. This admission is clearest in Freud’s explanation for a famous slip of the tongue:[7]

In commenting on this case (which he does several times during his career), Freud sometimes emphasizes the similarity between the President’s initial slip and his subsequent correction.[8]

Here Freud implies that the President’s utterance of the word “closed” was motivated by a desire to close the sitting, just as his subsequent utterance of the word “open” was motivated by a desire to open it. In other passages, however, Freud draws a contrast between the two utterances. In the first case, he points out, the President “said the contrary of what he intended”, whereas “[a]fter his slip of the tongue he at once produces the wording which he originally intended”—and which he now presumably intends again.[9] The correction is therefore intentional in a sense that the slip was not. Indeed, Freud ultimately implies that the slip was committed not only unintentionally but unwillingly, since he says that the desire to close the session “succeeded in making itself effective, against the speaker’s will.”[10]

Thus, Freud’s explanation of the slip as purposeful leaves unchallenged the speaker’s own sense that his power of speech ran away with him, or that his words “slipped out”. The explanation would contradict the speaker only if he went to the extent of denying that it was indeed his power of speech and his words that were involved. The Freudian explanation would then force the speaker to admit that “I declare the sitting closed” was something that he said—not, for example, a noise forced from his throat by a spasm. But the Freudian explanation still allows him to claim that he said it despite himself, and that it was therefore a slip, however motivated.

Such cases require us to define a category of ungoverned activities, distinct from mere happenings, on the one hand, and from autonomous actions, on the other. This category contains the things that one does rather than merely undergoes, but that one somehow fails to regulate in the manner that separates autonomous human action from merely motivated activity. The philosophy of action must therefore account for three categories of phenomena: mere happenings, mere activities, and actions.

Making Things Happen

The boundaries separating these categories mark increments in the subject’s involvement as the cause of his own behavior. A slip of the tongue differs from a spasm of the larynx, we observed, in that it doesn’t just issue from the subject: he produces it. But then, of course, there is also a sense in which his utterance is produced despite him, by a desire that he didn’t intend to express. Similarly, a person can knock something off a desk in a manner that is adroitly clumsy—perfectly aimed, on the one hand, and yet also out of his conscious control, on the other.

Mere activity is therefore a partial and imperfect exercise of the subject’s capacity to make things happen: in one sense, the subject makes the activity happen; in another, it is made to happen despite him, or at least without his concurrence. Full-blooded human action occurs only when the subject’s capacity to make things happen is exercised to its fullest extent. To study the nature of activity and action is thus to study two degrees in the exercise of a single capacity.

This capacity merits philosophical study because it seems incompatible with our conception of how the world works more generally. We tend to think that whatever happens either is caused to happen by other happenings or just happens, by chance: events owe their occurrence to other events or to nothing at all. But if we make things happen, those events owe their occurrence to us, to persons. How can people give rise to events?

On the answer to this question hangs the viability of innumerable concepts indispensable for everyday life—concepts of human agency, creativity, and responsibility. Nothing that happens can genuinely be our idea, our doing, or our fault unless we somehow make it happen. Without a capacity to make things happen, we would never be in a position to choose or reject anything, to owe or earn anything, to succeed or fail at anything. We would simply be caught up in the flow of events, and our lives would be just so much water under the bridge.

We don’t seem to be adrift in the flow events: we seem to intervene in it, by producing some events and preventing others. Yet our intervention invariably consists in thoughts and bodily movements, which either happen by chance or are caused to happen by other thoughts and movements, which are themselves events taking place in our minds and bodies. Our intervening in the flow of events is just another part of that flow. So how can it count, after all, as an intervention—or, for that matter, as ours?

The Standard Model

The standard answer to this question goes like this. We want something to happen, and we believe that some behavior of ours would constitute or produce or at least promote its happening. These two attitudes jointly cause the relevant behavior, and in doing so they manifest the causal powers that are partly constitutive of their being, respectively, a desire and a belief. Because these attitudes also justify the behavior that they cause, that behavior eventuates not only from causes but for reasons. And whatever we do for reasons is consequently of our making.

Thus, for example: I want to know the time; I believe that looking at my watch will result in my knowing the time; and these two attitudes cause a glance at my watch, thus manifesting their characteristic causal powers as a desire and a belief. The desire and belief that cause my glance at the watch are my reasons for glancing at it; and because I engage in this behavior for reasons, I make it happen.[11]

This model seems right in several respects. To begin with, it treats my making something happen as a complex process composed of simpler processes in which events are caused by other events. I can make something happen even though it is caused by other events, according to this model, because their role in its production can add up or amount to mine. If the model identifies events whose causal role really does amount to mine, then it will have succeeded in reconciling my capacity to make things happen with the causal structure of the world.

The model is at least partly successful on this score. The events that it picks out in the causal history of my behavior are closely associated with my identity, and the causal operations of these events consequently implicate me, at least to some extent. What I want and what I believe are central features of my psychology, which is central to my nature as a person. My wantings and believings are therefore central features of me, and whatever they cause can be regarded as caused by me, in some sense.

The question remains, however, whether the causal role of my desires and beliefs adds up to the role that I play in producing an action or whether alternatively, it amounts to the role that I play in producing a mere activity. The claim made for the standard model is that it is a model of action, in which my capacity to make things happen is exercised to its fullest extent. Is this claim correct?

The standard model is at least correct, I think, about what this claim will require for its vindication. The model assumes that the processes constituting a person’s role in producing an action must be the ones that connect his behavior to reasons in such a way that it is based on, or performed for, those reasons. If a person’s constitution includes a causal mechanism that has the function of basing his behavior on reasons, then that mechanism is, functionally speaking, the locus of his agency, and its control over his behavior amounts to his self-control, or autonomy.[12]

Why would behavior produced by such a mechanism be any more attributable to the person than that produced by other causes? The answer is that a person is somehow identified with his own rationality. As Aristotle put it, “Each person seems to be his understanding.”[13] Hence causation via a person’s rational faculties qualifies as causation by the person himself. Of course, this statement raises more questions than it answers; but I hope to answer those questions, too, by the end of this Introduction. For now, I simply want to endorse this inchoate intuition underlying the standard model of agency.

One might object, at this point, that responding to reasons is the function of an entire person, not of a causal mechanism within him. The phrase “responding to reasons”, one might insist, already describes something done by the agent and hence cannot describe a mere chain of events.

To be sure, the concept of basing behavior on reasons belongs to the same conceptual vocabulary as that of performing an action or making things happen, and so it cannot provide the desired reduction of those concepts into the vocabulary of event-causation. But it isn’t meant to provide that reduction. “Basing behavior on reasons” is not proposed as an event-causal replacement for agential concepts; rather, it is proposed as that agential concept whose reduction will be the key to reducing the others. In order for a chain of events to constitute a person’s making things happen, in the fullest sense of the phrase, it will have to constitute, more specifically, his doing something for a reason.

So says the standard model of action—rightly, in my view. But my agreement with the standard model ends here. The model goes on to say that the chain of events constituting a person’s doing something for a reason is that in which his behavior is caused by a desire and belief in the manner that’s characteristic of those attitudes. I think that this aspect of the model runs afoul of obvious counterexamples.

Failings of the Standard Model

The standard model already contains a clause designed to rule out some counterexamples, in which behavior is caused by a desire and belief but fails to constitute an action performed for reasons. This clause appears in my formulation as the requirement that the desire and belief causing behavior must exercise the causal powers that are characteristic of those attitudes.

Here is an example in which desire and belief operate uncharacteristically. A speaker’s desire to win the sympathy of his audience, and his belief that nothing short of tears would suffice, might frustrate him to the point of tears. In causing behavior through the medium of frustration, his desire and belief would not manifest their characteristic causal powers. Characteristically, these attitudes cause whatever behavior is specified in the content of the belief as conducive to the outcome desired.[14] But the speaker in this case could have been frustrated to the point of tears by many different beliefs about the difficulty of attaining his goal. It’s just an accident that the belief frustrating his desire, and thereby producing tears, is a belief about the necessity of tears. The mechanism thus actuated—that is, the mechanism of frustration—would not in general conform the subject’s behavior to the instrumental content of his belief. Hence his motives do not exercise their characteristic powers in causing his behavior.

Adherents of the standard model believe that by ruling out such cases, in which behavior is caused but not motivationally guided by a desire and belief, they have succeeded in narrowing their analysis to behavior that is caused in the right way to qualify as an action performed for reasons. I think that they have made considerable progress in narrowing their analysis to behavior that qualifies as motivated activity. But I do not think that motivated activity necessarily constitutes an action performed for a reason.

Recall the first Freudian slip examined above.[15] The agent wants to destroy his inkstand and he is thereby moved to do what he knows will destroy it. His behavior thus satisfies the standard model, but it doesn’t qualify as an action: it’s a defective instance of the agent’s making something happen.

Note that this example is not ruled out by the requirement that desire and belief exert their characteristic powers in causing behavior. In Freud’s explanation of his mishap these characteristic powers are indeed at work. It’s no accident that the agent is caused to do what’s specified in the content of his belief as conducive to the desired outcome of obtaining a new inkstand: the instrumental content of his belief is what’s governing his behavior. So the agent really does brush the inkstand’s cover off the desk for the purpose of destroying it, unlike the frustrated speaker imagined above, whose tears are shed out of frustration and hence not for any purpose.

In sum, the agent’s movement is caused in the way that’s designated as right by the standard model; and yet it is only an activity. The standard model thus appears to specify the wrong “right way” for behavior to be caused. It specifies the way in which behavior must be caused in order to qualify as a purposeful activity, but not the way it must be caused in order to qualify as an autonomous action.

If we want to know why the standard model has failed to specify the right way for autonomous action to be caused, we need look no farther than the requirements that the model set for itself. The idea behind the model, remember, is that the causal processes constitutive of action will be the ones in virtue of which behavior is based on or performed for reasons. Those processes are the ones that the model aspires to specify as the right way for action to be caused. But the model has not lived up to its own aspirations: it hasn’t specified the processes in virtue of which behavior is based on reasons.

A reason for acting is something that warrants or justifies behavior. In order to serve as the basis for the subject’s behavior, it must justify that behavior to the subject—that is, in his eyes—and it must thereby engage some rational disposition of his to do what’s justified, to behave in accordance with justifications. When someone just knocks over something that he unconsciously wants to destroy, or blurts out something that he unconsciously wants to say, he has not necessarily seen any justification for his behavior, nor has his rationality been engaged, although he has indeed been motivated by a desire. So although his behavior has been caused by something that may in fact be a reason, it has not been caused in the right way to have been done for that reason.

This flaw in the standard model is papered over, in some versions, by a characterization of desire itself as entailing the grasp of a justification for acting. Engagement of the agent’s rationality is thus claimed to be inherent in the very nature of desire.

Desiring something entails regarding it as to be brought about, just as believing something entails regarding it as having come about, or true. And regarding something as to be brought about sounds as if it entails seeing a justification for acting. Proponents of the standard model therefore claim that if a subject desires something, and believes some behavior conducive to it, then he already sees a justification for that behavior, and his responsiveness to reasons is thereby engaged.

Unfortunately, this argument trades on confusions in the language of “seeing” and “regarding as”. To say that desiring something entails regarding it as to be brought about is simply to describe the so-called direction of fit that’s characteristic of desire. Desire has what is called a mind-to-world direction of fit, in that its propositional object functions as a model for what it represents rather than as modeled after it. When the President wants the session of the Senate to be closed, for example, he has a mental representation of the session’s being closed, and that representation serves as an archtype for the state of affairs that it represents rather than as an ectype of it. But to regard the session’s closure as to be brought about in this sense is not to think of it as appropriate or fitting or correct to bring about: it is not to judge that the session’s closure is desirable or good. It’s simply to hold the thought “session closed” in a conative rather than cognitive mode. Thus, wanting the session to be closed does not entail seeing any justification or warrant for behavior conducive to closing it.

The objection therefore stands that the standard model fails to specify the way in which action involves causation by reasons, although it succeeds in specifying the way in which purposeful activity involves causation by desires and beliefs. The standard model is a model of activity but not of action.[16]

Let me pause for a brief summary. I began by drawing a distinction between mere activity and action, which differ with respect to the degree of the subject’s agency—the degree to which he makes things happen. I then posed the question how a person can make things happen in a world where events are caused by other events. An answer to this question, I suggested, would have to show how causation by events could add up to or amount to causation by a person.

I next examined a standard model of agency, which rests on the premise that a person causes his behavior when it is caused by reasons in such a way that it is based on or performed for those reasons. The model claims that behavior is performed for reasons whenever it is caused by desire and belief in the characteristic way. But some instances of characteristic desire-belief causation yield no more than mere activity, because the resulting behavior is not based on the desire and belief as reasons. Hence the standard model is sufficient for motivated activity but not for autonomous action.

I shall now consider a proposal for improving the standard model by adding to it. This proposal will bring us closer to an account of agency, but still not close enough. My critique of this proposal will suggest a third—and, in my view, correct—account of agency.

Adding to the Standard Model

I have argued that when desire and belief cause behavior in such a way as to operate as its motives, they do not necessarily operate as its reasons—that is, as reasons for which the behavior is performed. But I do not claim that their being motives for acting somehow excludes their also being reasons; nor do I claim that their operating as motives somehow excludes their also operating as reasons. All I claim is that their operating in the one capacity doesn’t amount to their operating in the other. Autonomous action requires something more than motivation by desire and belief, as is demonstrated by motivated slips that are not autonomous; but the “something more” that’s required can be provided by the same motivating desire and belief, operating in an additional capacity.

Consider an alternative version of Freud’s story, in which the agent not only is motivated by his desire to destroy the inkpot but also acts on the grounds of that desire, in its capacity as a reason. In acting on that desire as a reason, let us suppose, he is aware of the desire and regards it as justifying a movement of his hand; and he makes the movement partly because of seeing it as justified. His desire thus causes his behavior via his disposition to behave in accordance with perceived justifications—a mechanism that wasn’t operative in the original story, where the agent was unaware of the justifying desire.

Yet even in the alternative version of the story, where the desire influences the agent via his perception of it as justifying action, it can continue to operate as a motive, as it did when it was hidden from view. The new influence that it now exerts in its capacity as a reason can be to enlist some reinforcement for, or remove some inhibition of, its own motivational force. Even when the subject is persuaded by rational reflection on his desire—a process different from simply being moved by its inherent force—his response to being persuaded can be to acquiesce in being so moved. On the grounds of his desire conceived as warrant, he may accede to its impetus as a motive.

The interaction of these causal mechanisms is not as mysterious as it may sound. Suppose that you were charged with the task of designing an autonomous agent, given the design for a mere subject of motivation.[17] If you like, you can imaginatively assign yourself to divine middle-management as project leader for the sixth afternoon of creation; or you may prefer to take the role of natural selection over the corresponding millennia. In either case, you face a world already populated with lower animals, which are capable of motivated activity, and your task is to introduce autonomous agents.

In neither case would you start from scratch. Rather, you would add practical reason to the existing design for motivated creatures, and you would add it in the form of a mechanism modifying the motivational forces already at work. You would design practical reason to survey a creature’s motives, to block or inhibit some of them, and to reinforce others.

A creature endowed with such a mechanism would reflect on forces within him that were already capable of producing behavior by themselves, as they do in nonautonomous creatures or in his own nonautonomous activity. His practical reasoning would be a process of assessing these springs of action and intervening in their operations—which intervention would require an additional, rational spring of action capable of modifying or redirecting the force exerted by the other springs.

The Hierarchical Model

Actually, you would already have come close to equipping motivated creatures with such a mechanism as soon as you had endowed them with self-awareness. For if creatures already have the capacity to desire—that is, to represent things as to be brought about—and if they subsequently gain the capacity to represent their own desires, then they will be able to desire states of affairs involving their own desires. They will be capable of wanting to have some desires but not others, and to be actuated by some desires but not by others. And then the latter, lower-order desires, upon coming to their notice, will be either reinforced or opposed by the force of the higher-order desires directed at them.

The result would be the hierarchical model of agency, as it is often called.[18] In the hierarchical model, the behaviors that a person makes happen, in the fullest sense, are the ones that are caused by his first-order motives as reinforced by higher-order motives. Autonomous action, according to this model, is behavior motivated by the desires and beliefs by which the subject wants, or is at least content, to be motivated.

I regard the hierarchical model of agency as an improvement on the standard model, because it requires the subject to be reflectively aware of his motives in order to act autonomously. A Freudian slip takes its agent by surprise, thereby casting him in the passive role of observer. This behavior takes its agent by surprise because its motive is unconscious: he is not aware of wanting to do what he does. Such a lack of self-awareness would not have disqualified the resulting behavior from being an autonomous action according to the standard model, but it is indeed disqualifying according to the hierarchical model. For an agent cannot want or be content to be motivated by a desire that he is unaware of having.

Even so, the hierarchical model of autonomous action still seems inadequate. In this model, the subject’s awareness of his first-order desires arouses second-order desires as to whether they should motivate him, but the latter desires are not necessarily a response to the rational force of the former: they aren’t a response to first-order desires in their capacity as reasons for acting. Hence the subject’s higher-order desires play a causal role from which he can once again be dissociated.

Suppose that the President in Freud’s second example had been aware of his desire to close the session rather than open it; and that he had even liked the idea of being motivated by that desire as he spoke. His desires might have been due to an overwhelming sense of depression or ennui at the prospect of the new session.[19] Also out of boredom or depression, the President might have taken perverse satisfaction in being moved to speak inappropriately; he might even have taken a further, perverse satisfaction in the perversity of his own satisfaction; and so on. According to the hierarchical model, his indifferent mood would have enhanced his autonomy, by forestalling any higher orders of dissatisfaction with his first-order motives. He would have been motivated by a first-order desire with which he was satisfied, and satisfied to be satisfied, and so on. The hierarchical model would therefore have classified his behavior as autonomous.

As this example suggests, however, higher-order satisfaction with one’s motives doesn’t necessarily make for autonomy. Insofar as the President’s satisfaction with his own motivational state was attributable to depression or ennui, it would not have been an expression of his own will. If anything, it would have expressed a lack of will on his part, under the weight of a psychic force that is usually regarded as pathological or alien.

The reason why the hierarchical model would classify this behavior as autonomous is that the model doesn’t distinguish sufficiently among the subject’s responses to his own motivation. A favorable disposition toward his first-order motives automatically implicates the subject in causing their behavioral results, according to this model, irrespective of why or how he is favorably disposed toward those motives. It doesn’t matter, in the hierarchical model, whether the subject is satisfied with his first-order motives because of depression or boredom or laziness—or, alternatively, because he is responding to their force as reasons.

But this distinction ought to matter in the constitution of autonomous action. A motive cannot be taken up into the subject’s will by just any favorable response to it. It can be taken up into the subject’s will only by a favorable response to it as a reason for acting—a response mounted by the subject’s rationality. What autonomy seems to require, then, is not just the capacity for higher-order motives in general but particular higher-order motives, which would reinforce the agent’s first-order motives insofar as the latter were perceived as reasons.

We thus arrive at a third model of action, avoiding the faults of the other two.[20] Unlike the standard model, this third model would exclude from the category of actions those unforeseen movements to which a person is impelled by motives of which he is unaware. Unlike the hierarchical model, it would also exclude movements produced by those motives which the agent endorses without regarding them as reasons. This model would define action as behavior whose first-order motives are perceived as reasons and are consequently reinforced by higher-order motives of rationality.

Higher-Order Motives of Rationality

What sort of higher-order motives can we imagine for this role? One possibility would be to posit a higher-order desire, on the part of every agent, to be actuated by those of his lower-order motives which constitute the best reasons for him to act. This desire would move a person to survey and evaluate his motives as reasons for acting, and it would then add its motivational force to whichever combination of motives impressed him as rationally superior.

I am going to argue in favor of positing something like a desire to be actuated by the best reasons—something that might imprecisely be described as such. But first I am going to argue against positing the desire that would fit the description precisely.[21]

In order to fit this description precisely, a desire would need to have a de dicto content that included the concept of motives that constitute reasons for acting. It would have to be a desire to be actuated by reason-constituting motives so described—described, that is, as constituting reasons. And there are serious problems with the notion that such a desire mediates an agent’s response to reasons.

One problem with this notion is that it would require a person to have the concept of a reason for acting in order to have the desire whose reinforcement of his other motives would turn their behavioral output into full-blooded actions. The notion would thus require a person to have the concept of a reason in order to be capable of acting at all. Indeed, it would require him to have, not only the generic concept of a reason, but a specific conception of what counts as a reason, and what makes some reasons better than others. Without such a conception, a person would never draw any conclusions about which motives were rationally superior, and his desire to be actuated by rationally superior motives would never be engaged. Agency itself would therefore require a fairly advanced level of intellectual sophistication.

A further problem emerges from consideration of how a person might form his conception of what counts as a reason for acting. What does count as a reason for acting, and how can a person tell that it does? In order to answer this question, we shall have to suspend our inquiry into the causal structure of practical reason, so that we can consider its logic.

I should pause, at this point, to beg for patience with the arguments that follow. When the dust settles, these arguments will have yielded a new account of agency. But the dust won’t settle for several pages, during which the outlines of the final product will remain obscure.

The Logic of Reasons

A reason is a consideration that justifies, and to justify something is to show that it is just, in the old-fashioned sense meaning “correct”. Something is subject to justification only if it is subject to a jus, or norm of correctness; and it is then subject to reasons in the form of considerations showing it to satisfy the norm.

Thus, a belief can be justified only because it can be correct or incorrect by virtue of being true or false. Reasons for a belief are considerations that show the belief to be correct by this standard, insofar as they show it to be true. The question, then, is what serves as the standard of correctness for action, in the same way as truth serves as the standard of correctness for belief.

There is a temptation to think that the norm of correctness for actions is that they should be supported by the strongest reasons. But this thought leads in a vicious circle. What counts as a reason for acting depends on what justifies action; which depends on what counts as correctness for action; which cannot depend, in turn, on what counts as a reason. Action must have an independent norm of correctness—a standard not dependent on the concept of reasons—before it can provide the sort of normative context in which reasons exist.[22]

Ideally, the norm of correctness for action should be exempt from deliberative criticism. That is, the norm should not leave open any question about whether to act in accordance with it. If such a question could be raised, it would have to be answered by appeal to reasons for acting in accordance with the norm; whereas the norm is supposed to determine what counts as a reason for acting, in the first place. If there had to be reasons for acting in accordance with the norm that determines what counts as a reason for acting, then our account of practical reasoning would again be circular.[23]

Here the analogy with theoretical reasoning suggests a solution. The norm of correctness for belief is not open to question because it is internal to the nature of belief itself. The concept of belief just is the concept of an attitude for which there is such a thing as correctness or incorrectness, consisting in truth or falsity. For a propositional attitude to be a belief just is, in part, for it to be capable of going right or wrong by being true or false. Philosophers have traditionally accounted for this feature of belief by saying that belief constitutively aims at the truth.[24]

If there were something at which action constitutively aimed, then there would be a norm of correctness internal to the nature of action. There would be something about behavior that constituted its correctness as an action, in the same way as the truth of a propositional attitude constitutes its correctness as a belief. This standard would not be open to question: actions meeting the standard would be correct on their own terms, so to speak, by virtue of their nature as actions, just as true beliefs are correct by virtue of their nature as beliefs. And this norm of correctness for action would in turn determine what counts as a reason for acting.[25]

Let me be clearer about the relation between the constitutive aim of belief and the norm that applies to belief in light of that aim.[26] To say that belief aims at the truth is not simply to re-express the norm stipulating that a belief must be true in order to be correct; rather, it is to point out a fact about belief that generates this norm for its correctness.

Belief aims at the truth in the normative sense only because it aims at the truth descriptively, in the sense that it is constitutively regulated by mechanisms designed to ensure that it is true. Belief also bears a more fundamental relation to the truth, in that it is an attitude of regarding a proposition as true; but in this respect it is no different from other cognitive states, such as assuming or imagining, which share the same direction of fit, in that they take their objects as true. What distinguishes a belief from other states that take their propositional objects as true is that, unlike assumption or fantasy, belief tends to track what is true, when its regulatory mechanisms are functioning as designed. If a cognitive state isn’t regulated by mechanisms designed to track the truth, then it isn’t a belief: it’s some other kind of cognition. That’s why aiming at the truth is constitutive of belief.

Belief thus aims at the truth in the same sense that the circulation aims to supply body tissues with nutrients and oxygen. Not just any movement of fluids within the body counts as the circulation, but only those movements which are under the control of mechanisms designed to direct them at supplying the tissues. Hence the aim of supplying the tissues is constitutive of the circulation, just as the aim of being true is constitutive of belief.

If action were to have a constitutive aim in the same sense, that aim would have to be a function of mechanisms that produce and control action, and constitutively so. What could those mechanisms be?

Well, we have already envisioned such mechanisms, in our third model of action, outlined above. That model characterizes action as behavior that is motivated by lower-order desires and beliefs as regulated by a particular higher-order motive. The model thus implies that it is constitutive of action to be regulated by a particular motive. The object of that motive, whatever it is, should qualify as action’s constitutive aim.

Our inquiry into the logic of practical reason has thus led us back to the causal mechanisms involved. Action would be logically subject to justification, we find, if it had a constitutive aim in relation to which it could be correct or incorrect merely by virtue of its nature as action. But the constitutive aim of action would have to be something at which it was in fact aimed; and its being aimed, in some direction or other, would be a fact about the mechanisms causing and controlling it—in particular, the mechanisms whose causing and controlling it were constitutive of its being action. And we previously posited such a mechanism, in the form of a higher-order, rational motive.[27]

We were wondering whether this higher-order, rational motive could consist in a desire to be actuated by whichever motives provide the best reasons for acting. That question prompted our digression into the logic of reasons; and the digression has now put the answer within reach.

What’s in question is the content of that higher-order motive which turns mere behavior into autonomous action by regulating how it is motivated. We have now seen that this higher-order motive, in constitutively regulating action, would lend it a constitutive aim in relation to which it would be subject to justification and reasons. But if this motive consisted in a desire to be actuated by reasons, then being actuated by reasons would be the constitutive aim of action—which is impossible, as we have also seen. For if the constitutive aim of action were that it be actuated by reasons, then the aim of action and the nature of reasons for acting would be caught in a vicious circle. What counted as a reason for acting would depend on what counted as correctness for actions as such; but what counted as correctness for actions would depend on what counted as a reason for acting.

We are hoping to find that action is constituted by a substantive aim, just as belief is constituted by the substantive aim of being true. The aim of belief is substantive in the sense that it is conceptually independent of reasons for believing; our hope for action must be, similarly, that it turn out to be constituted by an aim conceptually independent of reasons for acting. What turns out to lend action its constitutive aim, we must hope, is a higher-order motive whose content does not include the concept of a reason.

Yet we have also said that the higher-order motive constitutively regulating action must be a rational motive, embodying the subject’s responsiveness to reasons, so that its role in producing behavior will be inescapably his. The question therefore arises how a motive can be rational, and embody the subject’s responsiveness to reasons, if its content doesn’t include the concept of a reason. The answer is that the motive constitutively regulating action will constitute the subject’s responsiveness to reasons simply because the considerations to which it responds will ipso facto count as reasons for acting.

Let me resort again to the analogy with belief. Indicators of truth count as reasons for belief because truth is the aim of belief; but truth’s being the aim of belief just consists in the way that the mechanisms of belief are designed to regulate it—which they do by responding to indicators of truth. Ultimately, then, indicators of truth count as reasons for belief because they are the considerations in response to which belief is designed to be regulated. The aim of belief and reasons for belief are fixed simultaneously, the one being determined by the way in which belief is constitutively regulated in response to the other.

So, too, we are hoping that considerations of a particular kind will count as reasons for acting because of action’s constitutive aim; but action’s having that aim must consist in the way that its mechanisms are designed to regulate it—which they must do in response to considerations of that kind. Thus, we must hope for considerations of a kind that count as reasons for acting because they are the kind in response to which action is designed to be regulated. The aim of action and reasons for action must be fixed simultaneously, the one being determined by the way in which action is regulated in response to the other.

Again, the considerations in response to which action is constitutively regulated will need to have something in common other than being reasons for acting, since their being reasons for acting will just consist in action’s being constitutively regulated in response to them. Whatever regulates action in response to these considerations may turn out to be describable, imprecisely, as a desire to be actuated by reasons. But there will have to be something else about them to which this regulatory mechanism responds, since its responsiveness to them will be what constitutes them as reasons, and hence cannot depend on their already being reasons.

Indeed, the analogy between action and belief suggests that this mechanism need not literally be a desire.[28] We conceived of it as a desire because we appended it to the hierarchical model of agency, which characterizes action as regulated by higher-order motives. The problem with the hierarchical model, we found, is that it can be satisfied by any higher-order motive at all, whereas the mechanism that constitutively regulates action must somehow connect it to reasons for acting. That’s why we posited a higher-order motive of rationality.

Yet consider the case of belief. Here the analogous role may be played by the subject’s desire to arrive at the truth; but it may also be played by sub-personal cognitive systems that are designed to track the truth, independently of the subject’s desires. Truth must be the aim of belief, but it need not be an aim on the part of the believer; it may instead be an aim implicit in some parts of his cognitive architecture. When his beliefs change in the face of evidence or argument, he might be described as trying to arrive at the truth, as if he were motivated by a desire. But this description might be a personification of aims that are in fact sub-personal.

Similarly, an agent’s responsiveness to reasons for acting may only sometimes be due to a desire for the relevant aim while, at other times, being due to psychological mechanisms in which that aim is implicit. Whenever someone acts, he might then be described as trying to arrive at that aim, as if motivated by a desire for it. As in the case of belief, however, this description might sometimes be a personification of aims that are in fact sub-personal. I shall therefore stop speaking of a higher-order, rational motive and speak instead of a higher-order, rational aim, which may or may not be imparted to action by a motive on the part of the agent.

The Constitutive Aim of Action

I have now arrived at a highly abstract schema for a theory of agency—so abstract, I fear, as to lack any intuitive appeal. So let me return to a more concrete level of thought. Let me return, in particular, to the examples with which I initially illustrated the difference between action and mere activity. I want to see whether a theory of agency can be developed directly from reflection on those examples. (The dust should finally begin to settle.)

Consider the slip of the tongue committed by the President in Freud’s anecdote: “I herewith declare the sitting closed!” This behavior was less than an autonomous action precisely because it was a slip. But what do we mean in calling the President’s utterance a slip, or in saying that it slipped out? What we mean, I think, is that it slipped past something—that it gave the slip to something that should have held it back. To what restraint did this utterance give the slip?

The relevant restraint is indirectly revealed, I believe, in the next sentence of the story: “His attention was only drawn by the general merriment and he corrected his mistake.” What this sentence tells us is that the speaker didn’t know what he had said until the laughter of his audience brought it to his attention. He came to know what he had said only by having his attention drawn to what he had heard himself say. And the reason why he came to know it in this manner is that as he said it, he didn’t know what he was saying.

The same is true of any slip of the tongue, including one that immediately draws the speaker’s attention without any help from the audience. Even when a speaker immediately catches his own mistake, the fact remains that he catches it: he receives it passively, in the manner of a surprised audience. And he is obliged to catch his faulty remark precisely because he let it fly without knowing what it was.

Unwitting speech of this sort is often prevented by an inhibition. We often proceed under an inhibition with a content something like this: “Keep your mouth shut unless you know what you’re going to say.” This inhibition, I suggest, is what our words slip past on those occasions when they slip out: they slip past our inhibition against saying what we don’t yet know that we’re going to say.

Not saying what we don’t know we’re going to say is clearly a means of ensuring that when we open our mouths to say something, we already know what it is. Our inhibition against saying we-know-not-what is thus the negative manifestation of a positive aim: the aim of knowing what we’re (already) saying. The words that slip out are the ones that escape regulation by this aim—which explains why we learn of them only by hearing ourselves say them.

Knowing what we’re saying is rarely our end-in-view, of course, but I’m not suggesting that it is; I’m claiming only that it is an aim regulating our verbal behavior. It is what might be called a sub-agential aim,[29] which is not represented in our practical reasoning. Our behavior is regulated by many sub-agential aims. For example, our movements through the world are generally regulated by the aim of avoiding pain, even though pain-avoidance becomes our end-in-view only on rare occasions, when we are deliberately being careful. We are usually inhibited from bumping into furniture or stepping on sharp objects, without thinking about the pain to be avoided. Our behavior is thus regulated by an aim that isn’t our end. I am suggesting that our behavior is similarly regulated by the sub-agential aim of knowing what we’re saying, which inhibits us from speaking until we know what we’re going to say.

In many cases, of course, we speak without already knowing the precise words that we’re going to use, or even the precise meaning that we’re going to express. All that we already know, in these cases, is the gist of what we’re about to say, or perhaps some mental images associated with the thought to be articulated. Our precise wording is then something that we learn only by listening to ourselves.

But the reason why we have to listen for our words on such occasions, I suggest, is merely that the cognitive aim regulating our speech has been scaled back, not that it has disappeared. The aim that’s operative on these occasions is merely to know what thought we’re expressing, rather than the words in which we’re expressing it; and so we are inhibited from speaking only until we know which thought we’re going to express. At that point, we simply let the words come out, without knowing which words they’ll be. But our verbal behavior has already been regulated by the cognitive aim of knowing our own drift, which is still, in modest form, the aim of knowing what we’re saying.

A natural thing to observe about these occasions is that we speak without choosing our words: to produce words without knowing what they’ll be is, in effect, not to choose them. The times when we choose our words are times when we don’t utter them until we have them properly in mind—which, I have suggested, are times when they are regulated by the aim of knowing what we’re saying.

This observation suggests that there may be a connection between choosing our words and uttering them under the regulation of this sub-agential aim. In fact, it suggests that there may be a connection between choosing our behavior in general and producing it under the regulation of the corresponding, general aim. The general aim corresponding to that of knowing what we’re saying would be this: the aim of knowing what we’re doing.

I suggest that you go back to the drawing-board and add this aim to your design for autonomous creatures. To their existing capacities for motivation and self-awareness, add the sub-agential aim of knowing what they’re doing. Now what sort of creatures have you designed?

Of course, your creatures still have first-order motives for doing various things on particular occasions. But because their behavior is now regulated by the aim of knowing what they’re doing, they will be inhibited from doing anything until they have an idea of what they’re going to do. And once they have an idea of what they’re going to do, they will gain an additional inclination to do it, rather than anything else, since doing something else would result, at least momentarily, in their not knowing what they were doing. Their antecedent motives for doing various things will thus be restrained, until there’s something that they expect themselves to do; whereupon the balance of forces will be decisively tilted in favor of doing the expected thing. Your creatures are therefore in a position to expect themselves to do any one of the things toward which they are antecedently motivated, in the confidence that they’ll do whatever they expect.[30]

Your creatures now seem to share our capacity for choice or decision-making. When we choose or decide what we are going to do, we settle that question in our minds and we thereby settle the same question in fact. We make up our minds that we are going to do something, and we thereby make it the case that we’re going to do it. Similarly for your redesigned creatures: settling in their minds what they are going to do is a way of settling that question in fact. For they will be inhibited from doing anything until they think they are going to do it, and then they will be prompted to do what they think.

Your creatures should also share our sense of having an open future or a metaphysically free will.[31] I don’t say that they have a metaphysically free will or open futures—only that they should share our sense of having them. I assume that your creatures will be governed by deterministic laws of nature, under which their futures are fixed by facts about the past. I therefore assume that their futures are not in fact open, and their wills are not in fact free in the metaphysical sense.

But when these creatures contemplate what they are going to do next, there is nothing that they must think they’ll do, in order to think correctly. There are many different things that they would be correct in expecting themselves to do, because they would do whichever one they expected. There is of course only one of these things that they are actually going to do; but they are going to do it only because they are going to expect so. They would do otherwise if they expected otherwise, and so they aren’t epistemically barred from those alternative expectations.

There is consequently an epistemic sense in which the future is open, from these creatures’ perspective. When they imagine various alternative futures for themselves, they would be correct to believe in any one of them, since the future that they end up believing in will be the future that they end up having. And being in an epistemic position to believe correctly in any one of several different imaginable futures feels like being in a position to have any one of several different imaginable futures. Thus, an epistemically open future is easily misperceived as metaphysically open. In reality, your creatures aren’t in a position to have any one of several different futures, only to believe in them correctly. But the confusion is understandable.

The fact that your creatures would now perceive themselves as free, or their futures as open, is indirect evidence for the hypothesis that they have in fact been endowed with a capacity for choice or decision-making. What makes us feel free, after all, is our own capacity to make choices. If you have given your creatures the same feeling by endowing them with a particular higher-order aim—which, as it happens, also enables them to settle what they are going to do—then you must have endowed them with something very like our own capacity for choice.

My examination of a Freudian slip has now led me to an hypothesis about the nature of choices or decisions. I first identified the inhibition that fails when we commit a slip: it’s the inhibition against doing things without knowing what they are. I then imagined this inhibition as being exerted by a second-order aim of knowing what we’re doing, which would regulate what we do, by guiding us toward things that we already know about. In the mechanism thus envisioned, we would have a kind of expectation that functioned in a peculiar way: it would settle in our minds the question what we were going to do, and it would thereby settle the same question in fact. Because this is exactly the function of a choice, I have arrived at the hypothesis that a choice consists in just such a self-fulfilling expectation.

One might wonder how we can come to expect behavior that isn’t going to occur unless we already expect it. There would seem to be, antecedently, nothing to expect.

But that’s the beauty of it: there isn’t antecedently anything to expect, if “something to expect” means some future event waiting for us to expect it. Our expectation of doing something embodies an invention rather than a discovery. For we can simply adopt the expectation that we’re going to do any one of the things for which we have some antecedent motives, and this expectation will modify the balance of forces so as to make itself true. We are thus in a position to make up our forthcoming behavior. Making up what we will do is, in fact, our way of making up our minds to do it.

Choice and Belief

I may at this point seem to have misappropriated terms like ‘expectation’ and ‘knowledge’. An expectation, after all, is a belief; and knowledge is true and reliably justified belief. How can we simply make up true and reliably justified beliefs? In order to answer this question, I shall need to examine the similarities and differences between the mental states of belief and choice.[32]

The relevant similarities and differences between these attitudes occur on three related dimensions, two of which I have already mentioned—namely, the attitudes’ directions of fit and their constitutive aims. I’ll begin with the first of these dimensions.

An attitude’s direction of fit, remember, consists in whether the attitude treats its propositional object as true or to be made true. The desire to act, on the one hand, and the belief that one will act, on the other, are attitudes toward the same proposition—i.e., that one will act—but the belief accepts the proposition as a representation of how things are arranged, whereas the desire projects it as a representation of how things are to be arranged. Their different directions of fit are what distinguish belief as cognitive from desire as conative.

Because choosing entails settling a question in one’s mind, it requires more than representing an answer as to be arranged. If it were still to be arranged that I was going to act, then it would not yet be settled that I was going to act; and insofar as I regarded it as to be arranged, I would not yet regard it as settled. Settling on a future action thus requires representing the action as arranged: my choice makes it true that I’m going to act, by representing it as true that I’m going to act. It therefore has the same direction of fit as a belief.

Where choice differs from ordinary beliefs is in a feature that might be called its direction of guidance. An attitude’s direction of guidance consists in whether the attitude causes or is caused by what it represents.

Now, there is a temptation to assume that if a mental state is cognitive, representing how things are, then it must be caused by how they are; whereas if it causes what it represents, then the state must be conative, representing how things are to be. This assumption implies that a cognitive direction of fit entails a passive direction of guidance; and, conversely, that only states with a conative direction of fit can be active or practical.

But when I make a choice, a question is resolved in the world by being resolved in my mind. That I am going to do something is made true by my representing it as true. So choice has the same direction of fit as belief but the same direction of guidance as desire: it is a case of practical cognition.

Choice has a third essential feature, which it shares with belief. Like belief, choice aims at the truth.

Note that accepting a proposition as true and aiming at the truth are two distinct features of belief. Mental states such as imagining and assuming also regard their propositional objects as true—that is, as representing how things are arranged rather than how they are to be arranged. But belief is regulated by the aim of regarding something as true only if it really is true; whereas imagining and assuming entail accepting a proposition fancifully or hypothetically, with some aim other than getting at the truth. The latter states thus share belief’s direction of fit but not its constitutive aim.

When I form a choice or decision, however, I aim to settle a question in my mind only insofar as I can thereby settle it in fact; I aim, that is, to avoid representing an arrangement that I am not thereby managing to make. In the face of compelling evidence that I’m not going to do something, my mind cannot remain made up that I’m going to do it: the decision to do something, like the belief that I’m going to, cannot withstand evidence to the contrary. It thus represents things as having been arranged in some way, with the aim of representing how they really have been arranged, albeit by this very representation. It has the cognitive direction of fit and the associated aim of being true, despite having a practical direction of guidance.

A state that represents something as true, with the aim of so representing what is true, ought to count as knowledge if it attains its aim in the right way. And it ought to count as belief whether or not it attains its aim.

I do not want to quibble over the terms ‘knowledge’ and ‘belief’, however, if anyone insists on withholding them from states with a practical direction of guidance. By the same token, I can expect to be allowed the provisional use of these terms, given that I have explained how they can be replaced by descriptions of the mental state in question. The full description of this state is that it represents as true that we are going to do something; that it aims therein to represent something that really is true; and that it causes the truth of what it represents.

Reasons for Acting

I have suggested that what you should add to subjects of motivation, in order to create agents, is the higher-order aim of knowing what they’re doing. Just design your creatures to gravitate toward knowing what they’re doing, and they will do only those things which they have made up their minds that they’re going to do; and so they will act by choice.

This design specification implies that self-knowledge is the constitutive aim of action. And in my schema for a theory of agency, the constitutive aim of action determines an internal criterion of success for action, in relation to which considerations qualify as reasons for acting. The question therefore arises what sort of reasons apply to action as constituted by this aim.

The answer is this: the considerations that qualify as reasons for doing something are considerations in light of which, in doing it, the subject would know what he was doing. They are, more colloquially, considerations in light of which the action would make sense to the agent.

Now, there is a definition of ‘making sense’ under which it is a term of practical rationality. What makes sense for someone to do, by this definition, is whatever he has reason for doing. The statement that reasons for an action are the considerations in light of which the action would make sense can therefore sound like a tautology. But I do not mean to speak tautologically.

When I speak of “making sense”, I am borrowing the phrase from the domain of theoretical reason, where it is used to characterize phenomena as susceptible to explanation and understanding. What makes sense to someone, theoretically speaking, is what he can explain. This is what I mean when I say that reasons for doing something are considerations in light of which it would make sense. I mean that they are considerations that would provide the subject with an explanatory grasp of the behavior for which they are reasons.

Here I may seem to have changed my mind about the cognitive aim under discussion. I initially said that the aim constitutive of action is to know what we are doing; I have now suggested that considerations qualify as reasons insofar as they provide an explanatory grasp of what we are doing. And knowing what we’re doing may seem slightly different from being able to explain it.

But the difference is more apparent than real. The ostensibly descriptive terms in which we aim to know what we’re doing are in fact descriptions of activities—that is, of behavior as actuated by particular motives under particular circumstances—and so they are in fact explanatory as well. The sort of self-knowledge at which we aim is embodied in descriptions such as “declaring the session closed” or “getting rid of an old inkstand”, and these descriptions carry implications as to the causes and conditions of the bodily movements to which they apply.[33] To know what we are doing is thus to grasp our bodily movements under concepts that set them in an explanatory context of motives and circumstances.

Considerations of these motives and circumstances are what qualify, in my view, as reasons for acting. They are the considerations out of which we can fashion a description that would embody a knowledge of what we were doing, if we applied that description to ourselves in the way that would prompt us to behave accordingly. Reasons provide us with an account of what we could be doing and, indeed, would be doing if we adopted an expectation to that effect.[34]

In this sense, reasons provide us with a rationale under which we can choose to act.[35]

Another way of putting the point is this: reasons for acting are the elements of a possible storyline along which to make up what we are going to do.[36] The story might be this: “My sister has just pointed out that my inkstand seems out of place beside my new antiquities, and I want to make way for a new one, which I suspect that she plans to buy for my birthday; so I’ll knock the old inkstand off my desk.”

Of course, this particular story is borrowed from an historical incident in which it notably failed to play the role of rationale. In this incident, recounted by Freud, the agent was indeed moved by the attitudes and circumstances mentioned in the story, but he was thereby moved to act without knowing what he was doing. The story itself occurred to him only after the fact. That’s why the activity motivated by his desire and belief did not amount to an autonomous action. He didn’t make up this particular piece of his personal history; he was obliged to discover it.

In order to have acted autonomously, the agent would need to have been actuated not only by the desire and belief mentioned in the story but also by the story itself, serving as his grasp of what he was doing—or, in other words, as his rationale. He would need, first, to have been inhibited from acting on any of his desires and beliefs until he knew what he was up to; and then guided to act on the relevant ones once he had adopted this story. He would then have acted autonomously because he would have acted for a reason, having been actuated in part by a rationale.

This conception of reasons as rationale requires considerable elaboration. What makes one reason better or stronger than another? How do we deliberate with the reasons so characterized?

Answers to these questions will have to come from the philosophy of action, if my schema for a theory of agency is correct. Reasons will have to qualify as better or stronger in relation to the constitutive aim of action, which lends reasons their normative force. Roughly speaking, the better reason will be the one that provides the better rationale—the better potential grasp of what we are doing.

If the agent in Freud’s anecdote had become aware of a desire to destroy his inkstand, he would immediately have realized that such an activity was contrary to various other motives of his, as well as to some of his customs, emotions, and traits of character. So even though he would subsequently have known that he was destroying the inkstand in moving his arm, he might still have wondered, “What am I doing?” That is, he might still have been puzzled as to how a person like him, with a makeup like his, would come to act on such motives; and so he wouldn’t really or fully have known what he was up to.

This lack of self-knowledge would have indicated to the agent that he would have had a better idea of what he was doing if he had chosen to do something else instead. That is, he could have adopted, and consequently enacted, a more intelligible story. And insofar as there was a more intelligible story for him to enact, by choosing to do something else, there was a better rationale for doing that thing instead.

Conclusion

I have now brought my discussion back to its point of departure: the difference between autonomous action and mere purposeful activity. Purposeful activity is motivated by desire and belief, but it may or may not be regulated by the subject’s grasp of what he is doing. Autonomous action is activity regulated by that reflective understanding, which constitutes the agent’s rationale, or reason—the reason for which the action is performed, and whose role as its basis is what makes it an action rather than a mere activity.

The arguments by which I have reached this conclusion are developed further in the essays reprinted here. They expand on ideas originally contained in my book Practical Reflection.[37]

One of the following chapters departs from the present line of argument in an important respect. In the title essay, “The Possibility of Practical Reason” (Chap. 7), I identified the constitutive aim of action as autonomy itself, partly because I liked the Kantian ring of that claim and partly because I had hopes of forestalling criticism of my view as oddly intellectualist, or as portraying an autonomous agent to be unduly self-absorbed. Unfortunately, I think that the resulting version of my view is unworkable, as becomes evident, I think, in the final, tortured sections of that chapter.[38]

Perhaps this result is not so unfortunate, after all. The intuition with which I began, in writing about the philosophy of action, is that autonomy is an expression of the drive to wrap your mind around things—an expression, in particular, of that drive as directed at yourself. You govern yourself, it seems to me, when you seek to grasp yourself as part of an intelligible world and consequently gravitate toward being intelligible.

The appeal of this view, for me, is that it locates autonomy in a part of the personality from which you truly cannot dissociate yourself. This part of your personality constitutes your essential self, in the sense that it invariably presents a reflexive aspect to your thinking: it invariably appears to you as “me” from any perspective, however self-critical or detached.[39]

That’s what Aristotle means, I think, when he says that each person seems to be his understanding.[40] You can dissociate yourself from other springs of action within you, by reflecting on them from a critical or contemplative distance. But you cannot attain a similar distance from your understanding, because it is something that you must take along, so to speak, no matter how far you retreat in seeking a perspective on yourself. You must take your understanding along because you must continue to exercise it in adopting a perspective, where it remains identified with you as the subject of that perspective, no matter how far off it appears to you as an object. Your understanding is therefore like that point between your eyes which constitutes the visual standpoint from which you see whatever you see, even when you view that point itself in the mirror, at a distance. Just as that point is always “here”, at the origin of your visual images, even when it’s also “over there”, in the mirror; so your understanding is always “me” in your reflective thinking, even when you regard it externally, as “it”. It’s your inescapable self, and so its contribution to producing your behavior is, inescapably, your contribution.

It’s an intellectualist view, all right. But we are intellectual creatures, and our autonomy may well be a function of our intellect.