Originally published in Freedom and Determinism, edited by Keith Lehrer (New York: Random House, 1966): pp. 141-74

Fatalism and Determinism

by Wilfrid Sellars

I. SOME TRADITIONAL CONSIDERATIONS

It will be useful to begin our discussion of the reconcilability of the concept of “action of one’s own free will” with scientific determinism by a discussion of the so-called Hume-Mill solution. In this tradition, it will be remembered, the problem appears as one concerning the compatibility of personal responsibility with the principle that every event has a cause. The fundamental move is to distinguish between causation and compulsion, and to argue that, while action of one’s own free will is, of course incompatible with compulsion, it is thought by Libertarians to be incompatible with causation because they have taken causation to be a form of or to involve compulsion. Action of one’s own free will is no more to be equated with uncaused action than causation is with compulsion. As the Hume-Mill tradition sees it two pairs of opposites, “free” - “compelled,” “uncaused” - “caused” are collapsed by the Libertarian into one.

Now there is certainly something of value in this way of approaching the problem, but it simply won’t do as it {142} stands. A series of comments, most of which are recognized, either explicitly or implicitly, by the tradition, will set the stage for a discussion of the problem as it exists today.

The first thing to be noted is that the term “cause,” as it appears in this argument, has a technical meaning, which by no means corresponds to that which it has in everyday life. The point is a familiar one, but it should at least be mentioned because it is a paradigm example of the role of ambiguity in the free-will issue.

As Collingwood and others have pointed out, causation in the ordinary sense is the idea of the intervention "of an agent in a system, thereby bringing about changes which would not otherwise have occurred. The root metaphor, to use Pepper’s invaluable expression, is that of a person bringing it about that another person or a group of persons does something which the first person wishes to have done. It is worth noting, therefore, that this root metaphor is essentially that of compulsion. That this root metaphor is by no means completely “frozen” undoubtedly accounts, at least in part, for the fact that many philosophers find it simply “absurd” that an action which was caused could be done of the agent’s own free will. It might also be added that, in this sense of causation, it is simply not the case that every episode has a cause. The king who causes a prisoner to be brought before him is not himself caused to make this move.

The practical sense of causation in its application to nature was gradually “depersonalized” into the analogical concept of a non-person—a merely material thing—intervening in a system of non-persons—other merely material things—bringing about changes which would not otherwise have occurred. Familiar examples are to be found in the areas of mechanics and electrodynamics: the cue ball in a break; a thermostat in a heating system. Notice, however, as Collingwood has emphasized, that persons hover in the background of even this ostensibly depersonalized sense of the term. The cause may not be a {143} person, but it is conceived of as manipulable by a person, at least in principle, and the objects involved are, cut up, from this point of view, into that which intervenes and the remainder which constitute the system under intervention. This fact, as I see it, is essential to understanding what is ordinarily meant by a causal explanation. Not even in this extended sense is it true that every event has a cause. Furthermore, an event which is not in this sense caused need not therefore be a brute matter of fact, an unintelligible happenstance. Common sense and science alike make constant use of the concept of a system which changes in intelligible ways in the absence of external intervention.

We shall have more to say at a later stage in the argument about the intelligibility of episodes which do not have intervening causes. Our next move must be to give a more adequate formulation of the Hume-Mill approach against the background of the above distinctions. In this reformulation, the intelligibility of episodes becomes their predictability, and determinism, although still couched in terms of “cause,” becomes the thesis that all episodes are in principle predictable.

It is often thought that universal predictability is an incoherent notion. The problem arises at two levels: (a) the level defined by the common sense framework of persons; (b) the level defined by the framework of microphysical theory. I shall be concerned with the place of predictability in the framework of persons shortly. As for (b), and abstracting from scientific issues pertaining to quantum mechanics, conceptual difficulties do arise about universal predictability if we fail to distinguish between what I shall call epistemic predictability and logical predictability. By epistemic predictability, I mean predictability by a predictor in the system. The concept of universal epistemic predictability does seem to be bound up with difficulties of the type explored by Gödel. By logical predictability, on the other hand, is meant that property of the process laws governing a physical system which involves {144} the derivability of a description of the state of the system at a later time from a description of its state at an earlier time; without stipulating that the latter description be obtained by operations within the system. It can be argued, I believe, with considerable force, that the latter is a misuse of the term “predictability,” but it does seem to me that this is what philosophers concerned with the free will and determinism issue have had in mind, and it simply muddies up the waters to harass these philosophers with Gödel problems about epistemic predictability.

Returning now to the Hume-Mill gambit, we see that the point is well taken that, if causality is construed in terms of predictability rather than intervention, then the fact that an action is caused does not imply that it is compelled. On the other hand, the overtones of the ordinary use of the word “cause,” pointed out above, make it unlikely that the libertarian will be persuaded by one who puts the matter in this way.

There is, as we have seen, a tension between the idea of “acting out of one’s own free will” and “being caused to do something.” It is therefore important to note that there may be a sense of “cause” which is stronger than that of mere predictability, and in which it is true to say that actions are caused, but not that persons are caused to do them. I shall be arguing shortly that there is such a sense and that in this new sense volitions are, at least on occasion, the causes of actions. That causation in this further sense does not imply compulsion is indicated by its departure in grammar from the form “thing (or person) impinging on things (or persons) bringing about a change, etc.”

But the Hume-Mill substitution of predictability for causation raises problems of its own, problems which in one way or another will be with us throughout the remainder of this paper. The crucial issue concerns the conceptual framework in which (or, as it used to be put, the level of explanation at which) the predictability is {145} supposed to obtain. I have distinguished in a number of papers between what I call the “manifest image” and the “scientific image” of man-in-the-world.[1] Roughly, the manifest image corresponds to the world as conceived by P. F. Strawson—roughly it is the world as we know it to be in ordinary experience, supplemented by such inductive procedures as remain within the framework. The manifest image is, in particular, a framework in which the distinctive features of persons are conceptually irreducible to features of nonpersons, e.g. animals and merely material things.

The scientific image, on the other hand, is man-in-the world as we anticipate he would he conceived by a unified scientific account, which makes use of the familiar techniques of theory construction. The scientific image, although methodologically rooted in the manifest image, freely transcends the conceptual framework of the latter, introducing new concepts, as it is said, by “postulation,” rather than by explicit definition, however widely construed. To articulate the philosophical tensions aroused by “scientific determinism,” let us suppose that the scientific image contains a picture of man as part and parcel of a deterministic order. To make our thought experiment as forceful as possible, let us suppose that, sooner or later, quantum phenomena will be found to have a deterministic substructure. Only by pressing this assumption can the irrelevance of quantum indeterminacy to the problem of free will be made manifest.

An indefinite amount of time could be spent in giving these two images of man-in-the-world a higher polish: I think however, that the above remarks suffice to mobilize a useful conception of the contrast I have in mind. The question I want to raise concerns the force of the term “predictable” in these two radically different conceptual {146} frameworks. Thus we have supposed it to be a framework principle of the scientific image that every episode, including the scientific counterparts of human thoughts and actions, is predictable. Is it, correspondingly, a framework principle of the manifest image that all human actions be predictable? Could it perhaps be a framework principle of the manifest image that not all actions are predictable, although the very same actions that are “manifestly” unpredictable are, as projected in the scientific image, in principle predictable? And, if so, would we have to say that one or the other image, no matter which, gives a false account of what there really is?

If we now turn our attention to the manifest image, and examine the use of the term “predictable” in connection with persons and what they do, a number of points stand out quite clearly—provided, that is, that we are careful to avoid mixing in elements from the scientific image. To say of a person that his actions are predictable is not always a compliment. Even a person who can be counted on to do what is right is marked down a little when it is said that one can predict exactly what he will do. For to say this implies that he meets situations in routine ways, never thinking things through afresh or gaining new insight.

To be predictable, in this image, means that a person’s actions are habitual, inferable by inductive reasoning based an observation of his past behavior. For an action to be predictable is for it to be an expression of character. It is therefore essential to note that, however predictable a person may become, not all of his actions can be expressions of character. Actions form character as well as express it, and the idea that actions which, in one respect, form character must, in another respect, he expressions of character can be traced to a confusion of the concept of the nature of a person (which concept belongs to the scientific image) with the concept of character.

Even within the manifest image, of course, things have natures. The natures of material things are their abilities {147} and predictabilities. The “realm of nature” consists traditionally of those things the observable behavior of which is either predictable, even if only “for the most part,” or unintelligible because random: Their intelligibility is a matter of dispositions and propensities working themselves out and being called into play.[2]

In the post-renaissance period, there began that blurring of the distinction between the manifest and the incipient scientific image that has been, ever since, a source of philosophical confusion. The combination of the failure to draw a clear distinction between inductive generalization and theory construction with an increasing appreciation of the power of scientific thought, led to an exaggerated conception of the place of predictability within the manifest framework of perceptible things. And since, whatever else persons may be, they are perceptible things, the modern form of the problem of free will began to take shape.

It is therefore important to realize that, even with respect to merely material things, the manifest image is not able, out of its own resources, to generate a deterministic picture. A deterministic picture arises at best indirectly, by the correlation of perceptible things with systems of imperceptible scientific objects, which are metaphorically said to be “in” them.

To explain what an observable thing does by reference to theoretical processes that are taking place ‘within’ it is, a radically different sort of thing from explaining the same behavior in terms of observable antecedent states.

A parallel, I believe, can be drawn between the above two modes of explanation and two modes of explaining {148} the observable behavior of persons: (a) in terms of other observable behavior; (b) in terms of inner mental states. From this point of view, the explanation of behavior in terms of character would be analogous to explaining the observable behavior of material things by means of inductive generalizations. And when what a person does is not in this sense predictable—i.e. not an expression of “second nature” or character—it would nevertheless not be an “unintelligible” episode, as would be true if it were the behavior of a merely material thing, which could not have been predicted without appealing to theoretical considerations. The role of inner episodes in the manifest image can be compared to, that of theoretical entities in the scientific image. They are, however, elements of an autonomous framework—not a speculative extension of microphysics—which carries the imprint of the specifically human observable behavior they are designed to explain.[3]

Thus, whereas the inductively unexplainable behavior of a merely material thing would be said to be unintelligible, a matter of chance, an inductively unexplainable action of a person—i.e. one which could not be explained in terms of character—can be intelligible in terms of being capable of explanation with reference to the practical reasoning and, ultimately, the volition of which it is the expression.[4] Needless to say, character defined in terms of dispositions and propensities pertaining to overt behavior will have its counterpart in the framework of mental episodes, just as causal properties operationally {149} defined in the observation framework have their counterparts in the theoretical framework which explains them.

Suppose it is granted that, in the manifest image, actions which are not intelligible in terms of character can nevertheless be intelligible in terms of the practical reasoning which is its cause.[5] But what, it may be asked, of this reasoning itself? Must not the occurrence of those inner episodes that explain the behavior of a rational being have a kind of intelligibility? And must not this intelligibility be a matter of predictability? The answer to the first question is, of course, yes. But to the second it is no! The intelligibility of thoughts is no more a matter of predictability than is the intelligibility of moves in games of chess—and for essentially the same reason.

I have been arguing that, in the manifest image, not everything that even a routine person does is predictable in the sense of being an expression of character, and that its intelligibility does not require predictability in this sense. I want now to make the stronger point that when, in ordinary discourse, we speak of predicting what, a person will do, we have in mind what I referred to above (p. 143) as epistemic predictability. I want therefore to point out that, given that this is the sense of predictability involved, it is indeed a framework principle of the manifest image that the behavior of rational beings cannot be supposed to be universally predictable. This point transcends the more elementary one that acquiring a character presupposes actions which are not themselves expressions of character. For to be a rational being is to be being who is capable of action that is not in character, and hence cannot be predicted within the framework of the manifest image.

But might not all human action be predictable if no holds are barred—if, that is, we take into account its projection {150} in the scientific image? And if so, would this predictability be compatible with any genuine sense of individual responsibility? (That the predictability in question would have to be logical rather than epistemic is clear.) But, before we take up these questions, it will be helpful to introduce some concepts pertaining to voluntary action and put them to use in defining additional dimensions of the problem of free will.

II. VOLITION

The central concept pertaining to practical reason is that of volition. A volition is an inner episode, a mental act, which is, in the absence of paralysis and granted the existence of favorable circumstances, the cause of the corresponding action. Thus, in the absence of paralysis or constraint, the volition to raise one’s right arm causes at raising of that arm. Before key issues in the free will-determinism syndrome can be properly formulated and resolved, it will be necessary to justify this concept.

Volitions are a subclass of thoughts, which can be initially characterized as mental episodes having a propositional form. As I have argued in “Empiricism and the Philosophy of Mind” (and Peter Geach has also so argued in Mental Acts), the concept of the propositional form of a thought is to be construed as an analogical extension of the concept of propositional form as it applies to meaningful overt speech.

The relation of a thought-episode to what is thought is correspondingly analogous to the relation of a meaningful utterance to what is said. Here the key is that to say of an utterance U that

U says that-p

is to classify the utterance as a ‘that-p’ utterance, in terms of its role in the language to which it belongs. ‘That-p’ is a class term, formed from the expression, in the language in which the classification is made, which plays the same {151} role. Thus statements about what expressions mean classify rather than relate: a token which means that p is a ‘that-p’ token.[6] It is often said that it is people rather than utterances which mean. But utterances are people uttering; the claim in question is true only in the trivial sense in which certain movements are a waltz only in so far as a person moving in certain ways is a person waltzing.

By analogy, to say of a thought-episode that it is a thought that-p is to classify the episode as a ‘that-p’ episode. The framework of thoughts is to be construed as an explanatory framework of entities postulated by a commonsense theory, and the analogical character of thoughts is to be be compared to the analogical character of the conceptual framework postulated by, say, micro-thermodynamics.

The model for this common-sense theory is what I shall call “thinking-out-loud.”[7] Here it is essential to distinguish between thinking-out-loud and verbal performances or actions. Telling someone something, or telling him to do something, are actions; thinking-out-loud is not. Imagine a person who wonders, observes, reasons, deliberates, decides, and initiates action “out loud.” That thinkings-out-loud are not actions, in the sense of pieces of conduct, follows from the simple but basic fact that they are not the sort of thing that is voluntarily brought about. There are, indeed, actions which are initiated by utterances; these, if they were spontaneously or candidly made, would be thinkings-out-loud. Thus, telling someone that there is a snake in the grass might be phonetically identical with thinking-out-loud that there is a snake in the grass. But to classify it as a telling is to characterize it as the sort of thing that could be initiated by deciding- {152} out-loud to do it. And an utterance thus initiated is, just for this reason, not a thinking-out-loud.

One must be careful here, for there are actions pertaining to thinking-out-loud. But these actions are not such items as thinking-out-loud that-p, but rather such items as

thinking-out-loud about whether 39 × 7 = 273

or

deliberating-out-loud whether to go to a movie.

Thus, consider the person who notices-out-loud that the table is green. His utterance

Lo, the table is green!

would not be a noticing-out-loud, unless it were a spontaneous reaction to the visual stimulus. An utterance like it in other respects, but which was brought about in the manner

“I shall utter ‘Lo, the table is green,’
so, lo, the table is green”

would not be a thinking-out-loud, nor would

“. . . therefore, it will rain,”

if its context was

“I shall utter ‘therefore, it will rain’;
so, therefore, it will rain.”

as contrasted with

“It is cloudy;
therefore, it will rain.”

Thinkings-out-loud can, of course, be indirectly brought about through self-training. But then the actions involved are not these thinkings-out-loud themselves, but rather the bringing about of propensities to think-out-loud in these ways in specified kinds of circumstance.

To sum up, then, thoughts-out-loud are not actions, {153} where “action” is understood to mean the sort of thing that one can appropriately be said to decide to do. The same is true of covert thoughts. They are “acts,” not in the conduct sense in which “action” was just used, but only in the sense that they are actualities. The same is true of such nonconceptual “inner episodes” as sensations, feelings and sense-impressions. These are “doings," only in that broad sense in which any verb in the active voice that stands for an episode expresses a doing. This is not to say that there are no such things as covert mental actions in the conduct sense. But, as in the case of actions pertaining to thinking-out-loud, these actions are not such items as thinking that Tom is tall, but rather such items as working on a mathematical problem, or deliberating about whether to go to a movie.

Thoughts, then, are, in the first instance, theoretical episodes designed to explain how people can behave intelligently, not only when their behavior is permeated by thinking-out-loud, but when they are silent.[8] That covert thoughts are to be reconstructed, in the first instance, as theoretical entities, carries with it the consequence that what determines the frequency and rapidity with which thoughts can legitimately be said to occur is the character of the overt phenomena which they are designed to explain. Roughly, we have reason to suppose that people do that amount and kind of thinking in a given timespan that is necessary to account for such of their behavior as could be imagined to be thought-through-out-loud. But if thoughts are in the first instance “theoretical,” they are in the second instance accessible to direct self-knowledge. We have acquired the ability (within limits) to monitor our thoughts—i.e. to respond to them, and set ourselves to respond, with second-level thoughts which are true of them in a way analogous to that in which, in perception, {154} we respond, and, in observation, set ourselves to respond, to perceptible things with true first-level thoughts. This monitoring makes it possible for our thought processes to be under our voluntary control, to the extent that they are in mental action proper. But I have, I hope, said enough to make it clear that unmonitored thinking is thinking in a straightforward sense. The important distinction here is the depth-psychological one, between thinkings which we are able to monitor (preconscious) and thinkings we are not (unconscious).

Notice that there are two senses in which an utterance can be said to “express” a thought. (a) In the first sense, by “thought” (cf. “Gedanke”) is meant a proposition, and the utterance expresses what it says.[9] (b) There is the radically different sense in which one characterizes an utterance as the “expression” of an inner episode, i.e. as the spontaneous culmination of a process initiated by the inner episode. (That the episode itself is classified as, for example, a thought (i.e. thinking) that-p, tempts unwary philosophers to run the two senses of “thought" and “expression” together.) That an utterance expresses in the first sense a certain thought (i.e. proposition) is a semantical fact. That it expresses in the second sense a certain thought-episode is a psychological fact which explains its occurrence. These two senses of “express” can be distinguished, respectively, as the logical or semantical, on the one hand, and the psychological or causal, on the other.

So far we have been discussing the status of episodes of thinking that-p. The concept of believing that-p can be explicated as a related disposition. Very roughly, to believe that-p is to be disposed to have thoughts that-p, rather than thoughts whether-p, let alone thoughts that-not-p.

All of the above distinctions apply to intentions. In the first place, we must distinguish between episodes of intending {155} (a species of thought-episode) and the corresponding dispositions. Dispositional intentions are the counterparts of believings. Intendings are, in the classical sense, practical thoughts, and the reasonings in which they occur are practical reasonings. Again, we must distinguish between intention as what is intended—the counterpart of a proposition—and intention as an act of intending or a disposition to intend.

The model for covert intendings is thinkings-out-loud of the kind illustrated by:

I shall raise my hand in ten minutes.
The victim shall be avenged.

The variety of intendings-out-loud is comparable to the variety of thinkings-out-loud that something is the case. I shall not, however, attempt to botanize them, save by implication. The “practical” character of intendings is a matter of their logical relation to action. In terms of intendings-out-loud, this can be put by saying that intentions of the form

“I shall now do A”

grow, in the absence of paralysis or hindrance or unfavorable circumstances, into doings of A. Intentions of other forms have a more or less indirect logical relation to action, by virtue of their logical relation to intentions that have a more direct logical relation to action. Intendings-out-loud of the form

“I shall now do A”

can be called willings-out-loud. They are the model for volitions. Thus, covert volitions are the inner episodes which find expression (in the psychological sense) in candid resolutive utterances, pertaining to action here and now.

But volitions are not only expressed in the psychological sense by overt speech (if one is in a thinking-out-loud frame of mind); they also find expression in a different, {156} but equally psychological sense in action. Thus, ceteris paribus, the volition “I shall now raise my hand” finds its expression, in one sense, in the overt resolutive

“I shall now raise my hand,”

and its expression in another sense in the action of raising my hand.

Shall we say that the volition is the cause of the action? I see no harm in doing so, if one is sufficiently sensitive to the variety of relationships that can be called causal, and avoids construing this case in terms of a paradigm that would lead one to say that the volition causes the person to do the action. In voluntary action, we are not caused to act.

Notice that volitions are not tryings. To try to do A is to do one or more things which one thinks likely in the circumstances to constitute doing A. Nor are volitions choosings. One can will to do A without choosing to do A rather than something else—even refraining from doing A. Nor are volitions decisions: a volition need not be the culmination of a process of deliberation or practical reasoning.

The idea that there are such things as volition—mental episodes that cause actions—has recently been under severe attack. Thus it has, in the first place, been subjected to the general attack on the idea of mental episodes proper—i.e. episodes that are not short-term dispositions pertaining to over behavior. It has become increasingly clear, however, that this attack was based on the overly narrow interpretation of the limits of concept-formation and rational belief, which takes as its paradigms ostensive teaching and induction by simple enumeration. To follow this road with single-minded conviction is to end in absurdity.

In the second place, there is the objection that, if volitions are episodes, they must be actions—that is, the sort of thing which can be done voluntarily. The conclusion is drawn that a vicious regress is involved in explaining voluntary {157} action in terms of volitions. But the premise of this argument is simply false. Thus, feelings of warmth are episodes, but they are not actions. Furthermore, we have already seen that thoughts that something is the case are not actions. It is a logical fact about thoughts that they are not directly brought about.[10] It is equally a logical fact about our intendings-at-the-moment that they are not under our voluntary control, just as it is a logical truth that our noticings-at-the-moment are not under our voluntary control.[11] I have no doubt that some philosophers have thought of volitions as actions, indeed as what we really do. Prichard’s curious notion that what we directly do is “set ourselves to bring something about” is a conflation—grounded in the thesis that we can be said to do, strictly speaking, only what we know that we can do—of the concept of volition as causing an action, with that of the minimal action by doing which, in the circumstances, we bring it about that we are doing such complex actions as shooting a person or paying a debt.

Finally, there is the objection that there would be a logical connection between these supposed episodes of willing to do A and the actual doing of A, which would be consistent with the Humean principle that it is logically possible for anything to cause anything. Could a willing to raise one’s hand cause a wiggling of the ears? The objection is a naive one, which overlooks the distinctive features of the level at which the causality obtains. The point stands out most clearly if we turn our attention to thinkings-out-loud, the primary mode of conceptual activity. Thus, to think out loud,

{158} “I shall now do A”

is not simply to utter noises. One hasn’t learned to think-out-loud in this mode—i.e. to use resolutive sentences—unless one has acquired a disposition to do A on saying “I shall now do A.” In the absence of this propensity, such utterances could simply not be said to express an intention or volition. The logical connection between “I shall now do A” and the doing of A is a presupposition of the parental training which results in the propensity. Thus, the characterizing of the utterance as a willing-out-loud logically implies the propensity. A willing-out-loud is the sort of thing that plays a specific role in language-guided behavior. In the absence of the conceptual commitments involved in classifying an utterance as a willing-out-loud, there would, of course, be no reason why

“I shall now raise my hand”

should be connected with hand-raisings rather than ear wigglings, or, indeed, with any particular form of behavior.

The same point stands out in the context of perception. Someone might argue: How can a green object in standard conditions be said to cause a person to be under the visual impression that he is confronted by a green object in standard conditions? Why shouldn’t green objects in standard conditions cause persons to be under the impression that a brass band is playing the Star Spangled Banner? The answer, once again, stands out clearly at the level of thinking-out-loud. To learn how to perceive-out-loud involves learning to respond with such utterances as

“See: this is a green object in daylight”

or, schematically,

“Lo: this is a φ object in perceptual circumstance C”

to green objects in daylight or to φ objects in perceptual circumstance C. Furthermore, to learn to use perceptual {159} predicates in the full way in which people who see use color words involves learning to use them in some such rubric.

Much more, of course, would have to be said in order to give even an elementary account of perceiving-out-loud, but for our present purposes the important thing to note is that the existence of these response propensities is a necessary condition of being able to perceive-out-loud. Notice that the logical connection with which we are concerned does not consist in the fact that ‘S perceives-out-loud that-p’ implies ‘It is true that-p’. For the connection in which we are interested is one which permits of error. This corresponds to the fact that, if a person is unwittingly paralyzed, no amount of his willing to raise his hand will cause his hand to rise.

It is perhaps worth noting here that both in the framework of thinking-out-loud and in that of covert thought, it makes sense to suppose that training (e.g. learning to play a composition for the piano) is able to link behavior into complex patterns which can be conceptualized and performed thinkingly as units, without each component being separately conceptualized and performed thinkingly as such. Thus, if I seem to be picturing waking life as threaded on a steady stream of thoughts—in particular volitions—this proliferation of mental acts must not be taken to imply that every motion that is under our voluntary control springs from a separate volition.

III. VOLUNTARY ACTION AND COMPULSION

In the primary sense of the phrase, a “voluntary action” ins one that is caused by an act of will. In a derivative sense, it is equivalent to the phrase “action under the agent’s voluntary control.” To say of an action, A, that it is under the agent’s voluntary control is to say that if, just before doing A, the agent had willed not to do A, he would not have done A.

If an action is not voluntary, at least in the latter sense, {160} then it is not really an action, but rather behavior of a sort which would be an action if it were voluntary. An involuntary wink is not a wink at all, but rather a blink. Even a compelled action must be voluntary in the sense defined. To go out the window propelled by a team of professional wrestlers is not to do but to suffer, to be a patient rather than an agent. Thus we can distinguish a narrower sense of “voluntary,” in which it means “voluntary in the previous sense and not compelled.”

The concept of compulsion is closely related to that of excuse. To say of a person that he was compelled to do something is to imply that the degree of duress to which he was submitted was sufficient to constitute an excuse for doing what he did. It is a familiar fact that the degree of duress which would be sufficient to excuse a civilian, for having performed or not having performed a certain act (thus betraying a secret), would not excuse a soldier.

Compulsion being a paradigm case of excusable circumstance, the term has naturally been extended to other cases of excusable behavior. Thus, for example, we find the term “inner compulsion” used when it is a question of excusing the behavior of mentally disturbed persons. The extensions are instructive, because they focus our attention on the relative thinness of the concept of voluntary action, as defined above. Consider, for example,the case of the narcotics addict who succumbs to clear and present temptation. If we say that the action was not under his voluntary control, the concept of compulsion is misapplied, and we should rather say that the addict didn’t do anything but rather, so to speak, blinked or twitched. If, on the other hand, we say that the action was under his voluntary control, in the sense that if he had willed to refrain he would have refrained, we are struck by the question “Could he have willed to refrain?” It occurs to us that there is a striking difference between those cases of compulsion in which it is true nevertheless that a person could have willed to do otherwise, and those in which it is not. In the latter case, action seems to {161} be excusable in what one is tempted to call an “absolute” rather than a “relative” sense. Indeed, as we shall see, it is just because determinism seems to imply that no one could have willed to do what he did not will to do, that it seems to give everybody an absolute excuse for all his wrongdoing, however relatively inexcusable he may be, as well as to strip him of all genuine merit, however relatively meritorious his actions may have been.

One often gets the impression that philosophers influenced by psychoanalytic theory have two models for human behavior, and believe there to be little behavior, indeed, which fails to conform to one or the other of these models. The first is the model of addiction. Its promiscuous use implies that psychoanalytic theory provides us with reasons for supposing that we are often unable to will other than we do. The other model is that of unavoidable ignorance, and, in particular, unavoidable ignorance of our preferences and desires. The primary value of John Wisdom’s thought-experiment,[12] concerning a person the strength of whose desires is fixed by the Devil, is that it makes us see that to be a free and responsible agent is to be able to bring it about (indirectly, by a program of self-training) that one’s desires are different or of different strengths. And it is indeed true that if one is unavoidably ignorant of what these desires and preferences are, one can scarcely be held responsible for not undertaking to change them. But under what circumstances are people thus unavoidably ignorant? It is of the utmost importance to keep metaphysical issues from confusing this scientific question.

IV. COULD ONE HAVE DONE OTHERWISE?

If microdeterminism is true, then, when X does A at t, this means that there was an antecedent partial cross-section of the universe such that, relatively to that cross-section, {162} it is physically necessary that X do A at t; or, equivalently, such that it is physically impossible, relatively to that cross-section, that X do something other than A at t—i.e. something which he did not in point of fact do.

It has seemed to many to follow from determinism, thus understood, that X could not at t have acted differently—that he was unable at t to do anything but A.

If it does have this consequence, then microdeterminism is, indeed, incompatible with our framework conceptions pertaining to action, for in that framework the following implications obtain:

X did A at t → X was able to do A at t
X was able to do A at t → X was able to do something else at t
X did A at t → X was able to do something else at t

These implications can be spelled out in more detail as follows:

The first two steps, (1) and (2), are justified on pp. 159-160 above. The crucial step is (3). Were does it come from? The answer is to be found by exploring what it means to say of a person that he is able to do a certain action. I shall first discuss the problem in a highly oversimplified context, and then introduce the complications necessary to give it a reasonably realistic flavor and to bring out its full relevance to the problem of free will.

I shall first introduce some familiar concepts and distinctions pertaining to the physical or “causal” modalities. Since at no time will I be concerned with the logical modalities, {163} I shall use familiar symbols from modal logic without any subscript to indicate that it is the physical modalities which are in question. Thus I shall use “N” for physically necessary, and “M” for physically possible. By means of this symbolism, I shall draw the familiar distinctions between absolute, hypothetical and relative modalities as follows:

(a) Absolute or intrinsic: an absolutely or intrinsically necessary state of affairs is one the necessity of which depends only on the conceptual content of the state of affairs as formulated. Thus (to use crude, but indicative examples):

It is physically necessary that whenever there is lightning, it thunders.
N[(t) L(t) → T(t + Δt)]
It is physically possible that lightning occur at t₁
M[L(t₁)]

(b) Hypothetical: a state of affairs q is necessary on the hypothesis that-p, if the proposition p → q is absolutely or intrinsically necessary. Thus:

That thunder occur at t₁ is necessary on the hypothesis that lightning occur at t₁ - Δt
N[T(t₁)] / L(t₁ - Δt)

(c) Relative: a state of affairs q is necessary relative to the fact that p, if q is necessary on the hypothesis that-p, and it is the case that-p. The basic form for representing relative necessity will be the same as that for hypothetical necessity, except that an exclamation mark is added at the end to indicate that the state of affairs to which the necessity is relative actually obtains; thus:

N[T(t₁)] / L(t - Δt)!

It is also useful to introduce quantified forms for hypothetical and relative necessity, thus

and

{164} In putting these concepts to use, the first simplification I shall make, which, in one form or another, will persist throughout the paper, is a limitation of the problem to the case of “minimal actions”—actions, that is, that do not have other actions as parts.[13] At least some minimal actions can be characterized as “bodily changes under direct voluntary control.” That the concept of action is being simplified in other useful ways will come out as the argument progresses. I shall use the following symbolism for statements pertaining to minimal actions (A):

‘A(x, t),’ for ‘x does A at t’
‘V_A(x, t)’ for ‘x wills at t to do A’
‘t′’ for ‘t - Δt’
‘ABLE[A(x, t)]’ for ‘x is able to do A at t’[14]

I shall next put these resources to use in presenting an oversimplified but essentially sound account of the compatibility between determinism and “could have done otherwise.” The key to this account is the following definition:

ABLE[A(x, t)] =_df N[V_A(x, t′) → A(x, t)]

i.e. x is able to do a certain minimal action at time t if, were he to will at t′ to do A, he would do A at t—i.e. willing to do A at t′ physically implies doing A at t.

Notice that while

ABLE[A(x, t)]

entails

M[A(x, t)]

{165} i.e. that x’s doing of A at t is intrinsically or absolutely possible (thus A is not the action of drawing a square circle in the air or completely rotating one’s head) it does not entail

(p) M[A(x, t)]/p!

where ‘p’ is a variable ranging over states of affairs assumed to be logically compossible with x’s doing A at t. If we represent this statement of relative possibility by

MR[A(x, t)]

we see that

ABLE[A(x, t)]

entails

M[A(x, t)]

but not

MR[A(x, t)].

Now, if determinism is true,

~A(x, t)

entails

(EK) K(x, t′) · N[K(x, t′) → ~A(x, t)]

i.e. that there is a kind of state, such that the universe was in that state just prior to t, and that it is absolutely physically necessary that, if the universe is in that state at a certain time t′, then x does not do A at t. If now we use ‘α’ to represent the state which satisfies this quantified formula we see that if determinism is true, from

~A(x, t)

we can infer

~M[α(x, t′) · A(x, t)]
~M[A(x, t)]/α(x, t′)!
~MR[A(x, t)].

{166} The last of the preceding formulas, although it asserts the relative impossibility of a doing, turns out, when unpacked, to assert he absolute impossibility of a complex state of affairs, of which x’s doing A is one conjunct. It does not assert that x was unable to do A at t, for

~MR[A(x, t)]

does not entail

~ABLE[A(x, t)].

Thus, although “it was (relatively) impossible that x do A at t” and “x was unable to do A at t” look deceptively alike in the above representation, which pictures a classic confusion, the former has the form

φ(x, t′) · N[φ(x, t′) → ~A(x, t)],

while the latter has the form

~N[Φ(x, t′) → A(x, t)].

The former denies the possibility of A(x, t) relative to (and, hence, on the hypothesis of) α(x, t′), whereas the latter asserts the possibility of ~A(x, t) on the hypothesis V_A(x, t′).

To put it somewhat differently, both determinism and the statement that x could have done otherwise tell us that something is hypothetically necessary, the former that

the latter that

where A is the action that was not done. It is clear that these statements are in no way incompatible.

V. A MORE REALISTIC ACCOUNT

I pointed out above that the above discussion would be oversimplified in a number of respects. The complexity I {167} shall now take into account involves the concepts of “circumstance,” “being in a position to do something,” “having the ability to do something,” and “being able to do something at a certain time.” Let us add the following symbols to our previous lists:

‘Γ’ for kinds of circumstance
‘A′’ for a kind of action incompatible with A
‘Π’ as a variable for periods of time
‘t’ for moments of time (in a vague sense of ‘moment’);

thus one speaks of during Φ and at t.

As sentence forms we have:

‘Γ’(x, Π) for “x is in Γ throughout Π”
‘A(x, t)’ for “x does A at t”
‘V_A(x, t)’ for “x wills at t to do A”
‘A(x, Π)’ for “x does A at some time during Π”
‘V_A(x, Π)’ for “x wills to do A at some time during Π,” (or, where appropriate, “x wills to do A at any time during Π”)
‘P ♢ Q’ for “P is physically compossible with Q.”

In terms of these, we introduce the following definitions:

1. PVT (read ‘prevents’)
PVT[Γ, A(x, Π)] =_df [Γ(x, Π) ♢ V_A(x, Π)] · N[Γ(x, Π) → ~A(x, Π)]

2. POSIT (read ‘is in a position to’)
POSIT[A(x, Π)] =_df ~(EΓ) PVT[Γ, A(x, Π)]
POSIT[A(x, t)] =_df (EΠ) t ∈ Π · POSIT[A(x, Π)]

3. CAN (read, alternatively, as “has the ability”)
CAN[A(x, Π)] =_df POSIT[A(x, Π)] → N[V_A(x, Π) → A(x, Π)]
CAN[A(x, t)] =_df (EΠ) t ∈ Π · CAN[A(x, Π)]

4. ABLE (read “is able during Π (or at t) . . .”)
ABLE[A(x, Π)] =_df POSIT[A(x, Π)] · CAN[A(x, Π)]
ABLE[A(x, t)] =_df (EΠ) t ∈ Π · ABLE[A(x, Π)]

In terms of these definitions, we can now set down the following action framework principles:

of which the latter is the more important for our purposes. From it follow:

Since

A(x, t) → ~A′(x, t),

we have, using PA-I, the consequence that

A(x, t) → (EA′) ~A′(x, t) · ABLE[A′(x, t)]

Let us return, this time in a more complex framework, to the question of the compatibility of determinism with the conceptual framework of action. The thesis of determinism applied to actions can be formulated as the principle:

where ‘α(x, t′)’ expresses the unspecified antecedent state of the universe relevant to x’s doing A at t. To build up the picture which generates the puzzles I am trying to clarify, let us transpose α, as a function of x with respect to t′, into a function γ which holds x with respect to t. We then have

In accordance with our conventions, this can be reformulated along the lines:

so that we have as our third formulation of the principle of determinism {169}

Let us now compare the consequent of PD-III

~(EA′) M[A′(x, t)]/γ(x, t)!

with the consequent of PA-I

(EA′) ABLE[A′(x, t)].

If we put these in words along the following lines:

It is not the case that there is an alternative action which it was possible [relative to γ(x, t)] that x do at t,

There is an alternative action which x was able to do at t,

they might look like contradictories. If we reformulate the latter to read:

There is an alternative action which it was possible for x to do at t,

it might be thought that the consequent of PD-III differs from the straight contradictory of the consequent of PA-I, by mentioning in addition the ground on which the possibility of x doing A is denied. The consequent of PD-III is then interpreted as:

It is not the case that there is an alternative action which it was possible for x to do at t—because γ(x, t).

Notice, however, the shift from

. . . which it was possible that x do at t

to

. . . which it was possible for x to do at t

This shift is a telltale symptom of the confusion which is being made.

The solution of the puzzle is essentially the same as that which was given above in the simpler framework, as becomes clearer when we turn our attention to PA-III: {170}

A(x, t) → (EA′) N[V_A′(x, t′) → A′(x, t)]

It might be thought that by virtue of the fact that, from PD-II it follows that

it also follows that alternative actions are prevented, for one is tempted to rewrite PD-IV as

A(x, t) → (A′)(EΓ) N[Γ(x, t) → ~A′(x, t)]

and to use the definition of ‘PVT’ to get

A(x, t) → (A′) PVT[A′(x, t)].

I am certain that something like this move has been the source of many misconceptions. This line of thought amounts to construing the antecedent state of the universe which makes A(x, t) relatively necessary, as a circumstance in which A is done! When it is put thus baldly, it is intuitively evident that this is a mistake.

The mistake is highlighted by two features of the definition of ‘PVT’ which the above move violates: (a) that which concerns the role of a period of time in the concept defined; (b) the neglected requirement that, in order to prevent an action, a circumstance must be compatible with willing the action. The requirement imposed by (a) can be only verbally satisfied bythe antecedent state of the universe. Thus, if an object y is φ at a certain time t, we can not only introduce a new predicate ‘φ′’ which applies to things at a time such that

φ′(y, t + Δt) =_df φ(y, t),

it being true at t + Δt that y was φ at t; we can also introduce the predicate φ′′ which applies to things during periods such that

φ′′(y, Π) =_df (Et) t ∈ Π · Φ(y, t).

The former was the move which took us in our discussion of PD-I from α(x, tʹ) to ‘γ(x, t)’. The latter is a similar move to satisfy a stronger requirement.

{171} The second feature (b) of the concept “prevents,” which precludes the possibility that the antecedent state of the universe envisaged by scientific determinism should prevent the agent from doing anything other than he did, concerns the requirement that a preventing circumstance be physically compatible with willing to do the prevented action. This requirement is incompatible with the fatalistic use of the principle of determinism; for, given that x did not will to do A′, then it is physically impossible, relative to the antecedent state of the universe, that x have willed to do A′.[15]

VI. COULD ONE HAVE WILLED OTHERWISE?

This brings me to the final section of my argument. Someone might grant that determinism is compatible with

which tells us that having done A at t implies having been able to do something else instead, but yet he might argue that this is a superficial truth. If one cuts deeper, one finds—so he claims—that determinism is incompatible with the idea that one could have willed to do anything other than what one did. And if so, is there not a deeper sense in which, if determinism is true, one could not have done anything other than one actually did?

We noted above (pp. 153 and 156-7) that, although to will to do something is a “doing,” in the broad sense in which any episode expressed by a verb in the active voice is a doing, it is not a doing in the conduct sense. And, since “willing” is always “willing to do” something in the conduct sense of “do,” it follows that

{172} x wills to will to do A

makes sense only as elliptical for something like

x willed to develop a frame of mind conducive to his willing to do A.

Now it might be claimed that our concept of ability to do is too narrow, since according to it ability applies only to doings in the conduct sense (actions), and not for example, to volitions. Is there not, indeed, a broader sense of “able,” in which persons are able to will as well as to act? And, if this is recognized, is not the way open for an argument to the effect that determinism is incompatible with the idea that a person could have willed to do something other than, in point of fact, he did will to do? Surely, it might be added, there is a correspondingly broad sense of “prevent” in which people can be prevented from willing. Does not determinism imply that, in this sense, people are prevented from willing to do what they did not will to do, and, hence, in a deeper sense than previously examined, from doing what they did not do?

I have already tipped my hand by pushing this line of thought to the conclusion that the antecedent state of the universe is the sort of thing that can prevent one from willing to do something. The absurdity, however, doesn’t stand out so clearly if we focus our attention on expressions such as

x was able to will to do A′
it was possible for x to will A′.

Once again the temptation is present to confuse

it is not possible that x at t willed to do A′ [because, according to the principle of determinism, etc.]

with

x was not able at t to will to do A′.

The confusion is so radical that it appears in the context of purely material things, as well as persons. The problem: does it make sense to say:

x at t is able to will to do A and also able to will to do A′[16]

is paralleled by the problem: does it make sense to say, using a familiar type of example:

x at t is able to go five mph and able to go ten mph?

What we want to say, roughly, is that circumstances do not prevent in the one case the willings, in the other the goings. But what is to count as a circumstance? The familiar phrase “standing conditions”[17] gives the clue". The concept of being in a circumstance is, as already pointed out, a period concept. We can introduce this broader sense of “able” for which we are looking, as, follows, representing it by ‘CAP’:

CAP[φ(x, Π)] =_df ~(EΓ) Γ(x, Π) · N[Γ(x, Π) → ~φ(x, Π)].

As before, ability at a time is derivative from ability over a period.

Is determinism compatible with

~φ(x, t) · CAP[φ(x, t)]?

Well, if determinism is true,

~Φ(x, t) → (Eγ) γ(x, t) · N[γ(x, t) → ~φ(x, t)],

and it might look as though we could treat the consequent as a substitution instance of

(EΓ) Γ(x, t) · N[Γ(x, t) → ~φ(x, t)],

{174} and conclude that, according to determinism,

~φ(x, t) → ~CAP[φ(x, t)],

which is compatible with

~Φ(x, t) · CAP[φ(x, t)].

Well, why not? What is the objection to taking

x’s having participated in an antecedent state of the universe which physically implies that x is not φ at t

as a circumstance which holds of x during a period to which t belongs? After all, it might be said,

x’s having heard an explosion at t

defines a physically relevant circumstance which obtains with respect to x during a subsequent period of time. There are various things which, he may not be capable of during that period—e.g. feeling peaceful and unafraid.

But surely this very example brings out the essential difference between deterministic pseudo-circumstances and genuine circumstances. For the law which relates hearing explosions at t to not being in certain states during the subsequent period is a specific law of nature, an “historical law” which is paralleled at the physical level by historical laws pertaining, for example, to elastic substances. It is with reference to “real” circumstances that abilities and hindrances are defined.

Thus, as I see it, we are often prevented by real circumstances from willing as we did not will. But the “metaphysical” circumstances implied by determinism do not render us unable to will to do what we did not do, and therefore do not, indirectly, render us unable to do what we did not do.

NOTES: