17

Click here to load reader

Alvin I. Goldman, A Theory of Human Action

Embed Size (px)

Citation preview

Page 1: Alvin I. Goldman, A Theory of Human Action

METAPHILOSOPHY VoI. 5, No. 4, October 1974

REVIEWS ALVIN I. GOLDMAN, A Theory of Human Action. Englewood

Cliffs, N. J.: Prentice-Hall, 1970. 230 pp. Alvin Goldman’s recent book, A Theory of Human Action is

not, perhaps, the only book-length study of the concept of action (there is Melden’s, which sets a great many of the central, most discussed puzzles: Anscombe’s and Kenny’s, which somewhat more tangentially explored unavoidable complexities; Shwayder’s, Louch’s, D’Arcy’s, Taylor’s and others’, all interesting and focussed more or less convergently on what one senses to be the heart of the matter). But Goldman’s is undoubtedly the most ambitious, sustained, and detailed effort in recent years to present a single articulated theory of action, often containing surprising doctrines, always intriguing, carefully formulated, and distinctly original. It is, accordingly, a not-to-be neglected item in the critical literature concerning action-probably for some time to come. With due respect to its importance, therefore, I shall, in reviewing its claims, attempt to point out in a serial way what I take to be essential mistakes and weaknesses in the argument.

The ground commitments are made in the first chapter. There, Goldman provisionally attacks what he calls the identity thesis (p. 2 ) (attributed principally to Donald Davidson and G. E. M. Anscombe, as in certain of their well-known statements). But he actually never says what the identity thesis is. For instance, i t might seem plausible to hold that, on the identity thesis, one and the same act may be identified under alternative descriptions (a seemingly irresistible thesis) or, that an act may be identified as one and the same in accord with this or that criterion (which is neither supplied by Davidson nor Anscombe nor formulated by Goldman) or, that, in context, it is reasonable to construe these or those descriptions as descriptions of one and the same act (though, since the determining criteria are not supplied, there is no relevant thesis to be attacked by Goldman). As a matter of fact, Goldman simply affirms: “In general, if X and Y are identical, then X must have all and only the properties that Y has”-suggesting that “we shall find . . . that some of the pairs of acts which are alleged to be identical do not share all the same properties” (p. 2). But this is baffling because 1) everything is

348

Page 2: Alvin I. Goldman, A Theory of Human Action

REVIEWS 349 self-identical and, necessarily, as such, has whatever properties it has; but 2) to speak of “X’ and “Y” being identical is, precisely, to concede that one and the same thing may, at times, be identi- fied under alternative descriptions or proper names or indexical designators or the like. Consequently, we are led to suppose that Goldman has a particular thesis in mind about how to reidentify one and the same act under different descriptions (that he means to challenge), wishes to say more than that putative such re- identifications (as perhaps Davidson and Anscombe provide) are merely wrong (if they are), and does not wish to affirm that it is never the case that one and the same act can be identified under alternative descriptions.

Goldman essentially argues that he has shown (in Ch. 1) that “many pairs of acts which are alleged to be identical fail to satisfy the principle of the indiscernibility of identicals : they fail to have all properties in common” (p. 4); and, although he recognizes the possibility of referentially opaque contexts, he will not be put off by the claim that such relevantly obtain-he opposes “proliferating opaque contexts’’ (p. 6 ) (without providing any rule for determining when they obtain and when they do not obtain) and even claims that, in context, “the proof of referential opacity could only be accomplished by begging the question” (p. 7). I’m afraid I think this is just utterly wrongheaded, and I stress the counterconsiderations because there is reason to think that the rest of this closely reasoned book depends entirely on these initial errors.

One absolutely central confusion in Goldman’s presentation concerns conflating causal contexts and contexts of causal ex- planation. If, say, John’s pulling the trigger and John’s killing Smith are said to be one and the same action under alternative descriptions (according to the identity thesis) then, why is it that, if it is true that “John’s act of pulling the trigger . . . caused the event in question, i.e., . . . caused the gun to fire”, “it would be extremely odd to say that John’s killing Smith caused the gun to go off’ (p. 2). The answer is elementary: i t would be odd to say so because so saying would not serve to explain why the gun went off; it would not be odd in the sense that the action in question (however described or identified) was not the cause (or part of the cause) of the gun’s going off (it was the cause). Causal contexts are extensional and contexts of causal explanation are not; and Goldman’s case does not in the least demonstrate that “if X and Y are [said to be] identical . . .”, they do not (in this instance) have the same properties in common, namely, the

Page 3: Alvin I. Goldman, A Theory of Human Action

3 50 REVIEWS

property of causing the gun to go off: “they” do have that property.

Goldman fails to see that if events or actions are taken as values in causal relations, statements or propositions or facts are taken as values in causal explanations (though, as Vendler has demonstrated, we may use mixed statements, that combine features of both sorts of relationship). The mistake is straight- forwardly compounded in instances like the one in which if John’s playing the piano = John’s putting Smith to sleep = John’s awakening Brown, one wonders how it could be that John’s playing the piano caused Smith‘s falling asleep but that John’s putting Smith to sleep did not cause Brown’s waking up; as well as in instances in which one holds that the wood‘s burning yellow and the wood‘s burning simpliciter must be different events (sic) because what explains the wood’s burning and what explains its burning yellow are different factors (pp. 2-3)! The mistake is puzzlingly pervasive. Also, of course, if causal con- texts (as distinct from contexts of causal explanation) behaved intensionally, Goldman would have not the slightest basis for denying opacity (without any questionbegging maneuver) and, hence, would not have the slightest basis for rejecting the so- called identity thesis. He does not, therefore, have an argument, whether or not causal contexts are intensional. Goldman says, further, that the identity theorist may wish to hold that the property of being supererogatory holds of actions “under certain descriptions”; but if so, then “we shall be forced to deny that actions cause events, and . . . say instead that only actions under descriptions cause events . . . [which1 has the unattractive conse- quence of committing us to the view that causation is somehow Zanguage-dependent” (p. 7). But the easy way out is to take i t that all evaluative discourse is intensional (all justificatory discourse, all discourse of appraisive and justifying reasons) and that this admission does not affect in the least the question of the exten- sionality of causal contexts. Goldman concludes that, since i t js “more natural” to “speak of acts per se as causes, as effects, as supererogatory, etc. . . . we must abandon the identity thesis [and say, for instance] that John’s pulling the trigger is a different act from John’s killing Smith” (p. 8). But there is a little of the “natural” on both sides and the denial of the identity thesis (still, by no means clear) is simply a non sequitur.

The rest of Chapter 1 concerns the distinction of act-types and act-tokens, which play a role in the rest of Goldman’s theory. Goldman explicitly states that “Being a token of a property

Page 4: Alvin I. Goldman, A Theory of Human Action

REVIEWS 351 should not be confused with exemplifying a property”: “act- tokens, like anything else, may exemplify indefinitely many properties, although they are tokens of only one property each”; and “an act-type is simply an act-property, a property such as mowing one’s lawn, running, writing a letter, or giving a lecture” (pp. 10-11). Goldman warns, however, that, say, “John’s moving his hand (at t ) is a token of the property ‘moving one’s hand’, but it does not exemplify that property. (John exemplifies that property)” (p. 11). The import of this, as I understand it, is that to exemplify a property is equivalent to the true predication of a property, of something-with referential emphasis, so to say, on the property rather than on the subject of predication. And, to be a token is to be an instance of a kind. To say so, of course, is to insure the intensionality of the concept token: ‘John’s moving his hand (at t)’ is, as such, a token of the property ‘moving one’s hand’; but if John’s act were differently described, it might well be reidentified as one and the same act under different descrip- tions, though it could not (ex vi termini) be the same act-token as a token of another type (which is doubly incoherent) or even be a token at all. (Actually, it is the occurrence of the act itself and not an expression identifying it that Goldman regards as the token, but he himself is quite casual about such contrasts). The confusion here has to do with reidentifying acts under alternative descriptions (without reference to tokens and types) and dis- tinguishing (in an entirely independent context) between act- tokens of different kinds. The reason this is important is simply that it shows the general tendency in Goldman’s account to conflate intensional and extensional distinctions. Goldman simply rejects, on the strength of the putative “counter-intuitive conse- quences” of the cases previously exposed, the elementary reason Davidson would adhere to a version of the identity thesis (that, say, John’s killing Smith [at t ] and John’s moving his finger [at t ] are the same act-namely, that Davidson (rightly) treats the phrase ‘John’s killing Smith‘ “as a definite description” (p. 12).

The mere mention of this obvious alternative, noting the con- fusion of acts and act-tokens, obviates the entire maneuver that Goldman is at such pains to provide. For, it no longer matters that (putatively different) act-tokens of different types cannot be the same act-tokens. The original question-utterly unrelated- had to do only with whether an act could be reidentified as one and the same under different descriptions! Goldman worries a great deal about the identity of properties, finally settles for an inconclusive synonymy test (pp. 12-14); and he is prepared to hold

Page 5: Alvin I. Goldman, A Theory of Human Action

3 52 REVIEWS

that it is not essential to his theory that, if the tallest man in the room (at t ) is the wealthiest man in the room (at t) , then (at least on the synonymy test) John’s hitting the tallest man in the room (at t ) would not be the same act as John’s hitting the wealthiest man in the room (at t ) (p. 13-14). But, first, Goldman himself asks whether these are one and the same act (finds that the denial is somewhat unwelcome); and second, Goldman holds that “John’s hitting the tallest man in the room (at t)” is not the same act- token as “John’s hitting the wealthiest man in the room (at t)”. He clearly confuses the two issues since he supposes that he has shown, by the cases he advances, that the identity thesis is unten- able. Also, the thesis that act-tokens can instantiate one and only one act-property is both counter-intuitive and extraordinarily ex- pensive as a way of obviating the identity thesis. For, consider that a table is a “token” of the type TABLE; it might very well also be a “token” of the type RENAISSANCE TABLE and of the type TABLE PURCHASED BY MY BROTHER and so on. But we would not wish to say, for that reason, that there are inde- finitely many table-tokens to be admitted in speaking of what purports to be a single table. We seem to be forced back to the description thesis.

If the distinctions here provided are fair, then we are forced to the surprising conclusion that Goldman has neither formulated the identity thesis that he wishes to attack, nor has he attacked any version of what might plausibly be construed as the identity thesis. He has instead turned to elaborate the distinction of act- tokens and act-types-that would normally have been conceded, without an argument, on the strength solely of the intensionality of talk of kinds. And this proves important, dialectically, because, at the close of Chapter 1, he introduces the concept of a “basic act-token” (adjusted from Danto’s concept of a basic act) in the following way (allowing for difficulties posed by verbs like ‘coughing’ that signify either deliberate acts or befallings that are not acts at all): “a token of a basic act-type is a basic act-token only if it is intentional” (p. 18). Nevertheless, symptomatically, Goldman himself concedes that “if the identity thesis is correct, then the distinction between basic actions and non-basic actions must be abandoned”, since our intuitions about which actions are basic will be undermined by the identities admitted (p. 6): here again, the argument has turned back to actions.

The underlying implausibility of Goldman’s thesis lies in this: we can hardly suppose that every particular act possesses one and

Page 6: Alvin I. Goldman, A Theory of Human Action

REVIEWS 353 only one property qua act, namely, that in virtue of which it is a token of a given act-type. On the other hand, if we grant that a given act-token may be re-identified under different actdescrip tions, we shall be faced with the analogue of the difficulty Gold- man thinks confronts the identity thesis and we shall be obliged therefore (if we allow Goldman’s argument, provisionally) to begin a regress. Goldman himself cuts the dilemma by conceding straightforwardly that “a person performs indefinitely many acts a t any time (at least whenever he is performing any acts a t all)” (p. 33f). But surely, this too is not entirely “natural” : we do not normally suppose that agents are performing indefinitely many- perhaps infinitely many-acts when they perform any acts, and we cannot restrict the number of acts (on Goldman’s account) to whatever descriptions we may supply. That there will be inde- finitely many-perhaps infinitely many- act-tokens whenever any act is performed is a trivial consequence of the definition of an act-token. I stress these considerations because, in Chapter 2, Goldman provides us with his ingenious account of the relation- ship between acts that, on the identity thesis, would normally have been taken as identical--justified, as far as I can see, solely on the grounds that he has already shown, in Chapter 1, the anomalies that result from holding the identity thesis (pp. 29, 32-3 3, 37-38). Nevertheless, there are serious difficulties confront- ing the analysis, even if we suppose that Goldman has disposed of the identity thesis.

Here, I must warn the reader that, in Chapter 2, Goldman de- fines a number of technical notions, with great care, that would require too much space to include in this report. The most im- portant is that of “level-generates”. Roughly, act-token A level- generates act-token A’ if and only if A and A’ are distinct act-tokens of the same agent performed at the same time (Gold- man adjusts the temporal conditions carefully) and A and A’ do not differ only “in containing different individual concepts of the same object’’ (as in the instance of hitting the tallest man in the room and hitting the wealthiest man in the room-where the same man is both the tallest and the wealthiest) and there is a set of conditions C* such that the conjunction of A and C+ entails A’ (though neither conjunct does), the agent would not have done A’ if he had not done A, and if C+ had not obtained then he would not have done A’ (even if he had done A) (p. 43)-there is a noticeable casualness here about specifying acts and statements about acts. In effect, the definition permits Goldman to articulate his alternative to the identity thesis. It appears to be coherent,

Page 7: Alvin I. Goldman, A Theory of Human Action

354 REVIEWS

but the question remains whether it is plausible and compre- hensive.

Level-generation is “an asymmetric, irreflexive, and transitive relation” (p. 21). We must distinguish generated and generating acts, according to a variety of processes; the contrast suggests the intuition of “level-generation”, which Goldman wishes to capture in a variety of “act-trees” (a heuristic diagrammatic device, which provides at its base at least one “basic act-token” and exhibits by vertical, horizontal, and branching diagonal lines various ways in which act-tokens are generated and in which act-tokens are de- pendent on, or independent of, one another). Goldman identifies four categories of generation : causal generation, conventiona2 generation, simple generation, and augmentation generation (p. 22). He holds that alternative classifications may be provided and may be as useful, but he does insist that the alternatives given are “exhaustive” of the processes of generation (p. 30). Regarding the first, we are warned that causal generation and causation “are mutually exclusive. Two acts can never be related both by causal generation and by causation” (p. 23). In fact, “Act-token A of agent S causally generates act-token A’ of agent S only if (a) A causes E, and (b) A’ consists in S’s causing E” (p. 23). So, if S flips the switch (A) causing the light to go on (E), then A causally generates the act of causing the light to go on (A’). This has the useful consequence of obviating Arthur Danto’s well-known mistake about basic actions not being caused by other acts; Goldman would say that a basic act may be caused by another act but not causally generated by another act (p. 24). Still, the very notion of basic and non-basic act-tokens depends, as we have seen, on the repudiation of the identity thesis and a shift from talk of acts (which Danto himself does not seem prepared to give up) to talk of tokens. The characterization of the second process is as follows : “Act-token A of agent S conventionally generates act-token A’ of agent S only if the performance of A in circum- stances C (possibly null), together with a rule R saying that A done in C counts as A’, guarantees the performance of A’ ” (p. 26). So, for example, S’s extending his arm out the car window “generates” S’s signalling for a turn; S’s moving his queen to king’s-knight-seven, S’s checkmating his opponent. Simple genera- tion involves no causal connection of the sort already mentioned and no rule. For instance, S’s jumping 6 feet 3 inches, in certain circumstances, simply generates S’s outjumping George (pp. 26- 27). The fourth process is contrasted with the others because it seems to fail the preposition-test, that the generated act is said to

Page 8: Alvin I. Goldman, A Theory of Human Action

REVIEWS 355 be performed by or in performing the generating act; nevertheless it is said to be “superior” to the other alternatives-presumably in avoiding the alleged anomalies of the identity thesis (p. 29). The augmentation cases are such that the generating act is en- tailed by the generated act but the reverse is not true (p. 28). Thus, for instance, S’s running is entailed by S’s running at 8 m.p.h. and S’s extending his arm, by S’s extending his arm out the car window. It does look as if, on the conditions given, if S performs any act a t all, then he performs not only indefinitely many acts but infinitely many acts. Surely, for the kinds of qualification of actions provided in a natural language (notice that we normally speak of qualifying one and the same action in a variety of ways, which forces us to attend to the distinction be- tween alternative definite descriptions of actions and the unique and essential properties of act-tokens), there are an infinite number of qualifications that can be provided. Perhaps only conventional generation precludes an infinite number of acts, though no act could generate other acts only conventionally. Again, augmentation generation has the counterintuitive conse- quence that we cannot acknowledge the adverbial qualification of a given act; for what purports to be such a qualification proves, on the thesis, to be part of an indissoluble act-property that is simply different from another (equally indissoluble) act-property. But if, on the basis of exemplification, we can individuate act- tokens as entities, there is no reason (assuming the identity thesis to be defensible) why we should not also be able to individuate acts as entities and attempt to preserve the logical features of adverbial qualification. Also, Goldman does not ever tell us what properties are actually full-fledged “act-properties’’ as opposed to properties that given acts may exemplify.

Goldman provides, in passing, details of considerable interest- for instance, an interpretation of the identity theorist’s view of a single action in terms of a single act-tree (p. 37) and a reasoned denial that the irreducibility of higher-level acts to lower-level acts (or of acts to movements) entails that higher-level acts (or acts as such) are not susceptible of causal explanation (against, say, philosophers like R. S. Peters) (pp. 40-41); also, an account of compound acts generated by each of two independent acts (pp. 34-35), an unresolved difficulty about “overdetermination” by two independent acts each of which would have sufficed to generate some A’ (pp. 43-44), and a clarification to the effect that, on the theory advanced, not every act-token need be intentional nor

MPH F

Page 9: Alvin I. Goldman, A Theory of Human Action

356 REVIEWS

every act, one for which the agent would be held responsible (p. 46).

There is, in Chapter 2, also provided a brief sketch of what is to be excluded from the class of acts: states, what happens to a person, what he suffers or undergoes, and the like (pp. 46-47). But there are other exclusions and anomalies that are not so easily accounted for. Goldman’s definition of acts “makes no provision for disjunctive acts, conditional acts, biconditional acts, or other ‘truth-functional’ acts, with the exception of conjunctive (com- pound) acts and negative acts . . .”; truth-functional connectives are not, Goldman says, “appropriate for forming new acts from other acts” (p. 47). But this cannot be a satisfactory reply, for an obvious reasons. If we merely christen any would-be acts of these sorts as distinct act-types, we shall have to admit all of them in specifying act-tokens. For instance, if S smokes or drinks, let us say that S drokes; then if S smokes (A), then A generates the distinct act-token A’ (S’s droking). But if this holds for disjunc- tion, i t will hold as well for all the connectives; also, Goldman seems to have ignored the formal relations that hold among the connectives themselves, that appear to outflank his own restric- tions. On the other hand, if we disallow (not unreasonably) all such “acts”, why should we not also exclude any number of others that Goldman admits as distinct acts (e.g., S’s saying ‘hello’ loudly as distinct from S’s saying ‘hello’)? There appears to be no uniform principle for justifying the constraint. Another difficulty (that Goldman acknowledges) concerns omissions (that is, negative act-tokens that are not intentional). Goldman is in- clined to disallow these as acts, though he admits that (since people are held responsible for omissions) omissions appear to be instances of negative acts (p. 48). He thinks that negative acts are difficult to characterize in terms of their level-relationship with other acts, because the causal status of negative events is rela- tively unclear. But, of course, on his own view, acts may generate other acts by other than causally linked factors. Also, if negative acts can generate or be generated by other acts (including other negative acts-in particular, omissions), then we would seem to have allowed for a fantastic proliferation of acts. Would the act of not not not turning the head be the same act as (if the critical terms are not synonymous though coextensive) that of not turn- ing the head (or would it be a token of a distinct type)? Another difficulty confronting omissions is simply this, that, contrary to Goldman’s view, if omissions were acts, then it would be difficult

Page 10: Alvin I. Goldman, A Theory of Human Action

REVIEWS 357 to hold that “at the bottom of any (complete) act-tree will be one or more basic act-tokens” (p. 46), for unintentional omissions cannot be basic-not being intentional. Again, if causal accounts may be offered for acts, then it would seem that acts actually performed cannot be indefinitely many or infinite in number. On the other hand, it seems curious to hold that simply by providing a description (even far-fetched descriptions) of what may be said to be additionally “generated” by a given act-as, say, by aug- mentation-we have increased the number of acts actually generated or have increased the number of acts that are to be causally accounted for. There seems to be no way of avoiding these anomalies, on Goldman’s account.

The rest of the general theory of action is provided in Chapter 3. First of all, Goldman there offers an account of intentional action. Roughly, an act is intentional if S’s “action-plan”, “in a certain characteristic way, causes S’s doing” the act (p. 57). An action-plan is “the combination of an agent’s action-wants and his projected act-tree”, that is, a combination of a predominant desire to do some act and a set of beliefs of varying certitude to the effect that doing a certain basic act would either generate or be on the same level as the act desired (p. 56). The central difficulty with this maneuver is the analogue of Grice’s, with respect to the causal theory of perception-which Goldman actually cites, in order to imitate (p. 63). Neither Grice nor Goldman squarely face the need to provide a model of causal explanation for perception or intentional action: they wrongly assume that “a complete delineation of the relevant causal process’’ may be provided “only through specialized research”-which philosophers cannot be expected to supply (p. 63). Goldman even acknowledges that “what. . . this ‘characteristic’ mode of causation” is is a question, “I confess”, to which “I do not have a fully detailed answer” (p. 62). But, of course, i t is (as Strawson, for one, has pointed out) questionbegging to present the thesis this way : we want to know, precisely, what the analysis of the concept would be when formu- lated exclusively in causal terms (that is, in terms of causal ex- planation), not merely that there are causal factors that bear on perception and intentional action.

The failure to supply the analysis wanted affects as well the definition of a “basic act-token”, since being intentional is a necessary condition of being such a token (p. 63). In principle, being unable to specify the required analysis entails lacking a precise criterion for marking off basic actions. Roughly, an act-

Page 11: Alvin I. Goldman, A Theory of Human Action

358 REVIEWS

token is a basic act-token if and only if it is of the type that, in standard conditions, S could exemplify if S wanted to and doing so would not depend on S’s knowledge of what other act-tokens would thereby be generated nor on S’s knowledge of what so doing would cause-and those conditions obtain (p. 67). A closer reading shows that Goldman simply wishes to accommodate an intuitive notion of basic actions--restricted essentially to bodily movements (p. 68); but there are difficulties. For one thing, doodling with one’s finger is characteristically unintentional, seemingly as basic an action as any bodily movement we might perform, and yet it is ruled out. Secondly, Goldman himself ad- mits that perhaps “there are some basic act-types which do in- volve level-generational knowledge” (raising one’s hand slowly, for example); but such he thinks depend on clear basic act-types (raising one’s hand, for example), so he restricts eligible knowledge to augmentation generation alone (p. 66). But then, if S knows that uttering certain sounds generates speech, will speaking be denied its status as a basic act-type? And if it is not, then level-generational knowledge cannot, with or without re- strictions, serve as a decisive distinction. Again, members of a given society may press their eyeballs only to get themselves to see double (may not have analyzed their behavior “further”)- they may, shall we say, only prouble. If they do not know how to generate seeing double thus by pressing on their eyeballs, is it then true that, in “proubling”, they are performing a basic action? If so, then what counts as a basic action seems to depend on ignorance; and if not, then knowledge cannot serve as the critical condition it is alleged to be (cf. pp. 65-66). Furthermore, in trying to distinguish between raising one’s hand and having one’s hand merely rise, Goldman appeals to the (admittedly quarrelsome) thesis that in order for the act to be a basic act-token, “his raising his hand must be intentional” (p. 71). But this is irrelevant for distinguishing merely between raising the hand and the hand’s rising and, even with respect to basic acts (as the case of finger- doodling makes clear), being intentionally performed (no matter how analyzed) cannot be an entirely satisfactory solution.

Other difficulties include the following. If unintentional acts are admitted as bona fide acts, then it ought to be a contingent thesis whether they are always generated by some basic act, but unin- tentional commissions like finger-doodling, acts legally specified in accord with strict liabilitity, and unintentional omissions serve as clear counterinstances; in such cases, basic act-tokens will not

Page 12: Alvin I. Goldman, A Theory of Human Action

REVIEWS 359 appear on the relevant act-trees. Again, if the basic acts are bodily movements, then omissions and mental acts will count as counter- instances. Again, for any alleged basic act, it is always, in prin- ciple, possible to provide an act-type in accord with which that act may be said to be generated. For instance, if clenching the fist is a putative basic action, then (omitting letters in ‘clenching’ to signify selected, more basic actions) “cleching” is the action of closing the fingers at least nearly to a clenching position, by which (ex hzjpothesi) S knows that he can generate clenching; “clening”, the action of closing the fingers at least half-way to- ward the clenched position, by which (ex hzjpothesi) S knows that he can generate clenching. Clenching may then be treated, per causal generation, as the act of causing oneself to clench by cleching or clening; or, perhaps more happily per augmentation generation, as cleching or clening “superabundently”. That these are utterly artificial distinctions count against the oddity of Gold- man’s criteria and the peculiarity of counting acts as act-tokens instantiating unique act-properties. Another serious difficulty (which, curiously, Goldman does not consider) lies in conceding that wants are really want-tokens. For, if want-tokens be ad- mitted, then want-token A may cause act-token A’ (where ‘A’ and ‘A” signify a relevant, however slight, non-concordance between want and act); hence, where A’ might otherwise be fairly regarded as a basic act-token, either it is not (on the definition given) or else it is (and the definition fails) or else we cannot decide (be- cause the meaning of Goldman’s condition “in a certain character- istic way” has not been satisfactorily detailed).

The rest of the chapter concerns aspects of causality. For one thing, Goldman reminds us that “there is a causal connection be- tween wanting to do a basic act A and the actual performance of a token of type A”; he seems to mean that there must be a uni- versal law connecting the two, whether or not (as with singular causal statements) we actually know the requisite laws (p. 72). But he misses the force of Strawson’s objection (which he cites) to the effect that actions cannot as such be effectively correlated with specific physical movements and hence, universal causal laws cannot relevantly be provided (pp. 75-76). Strawson’s view apart, the fact that every physical movement is caused in accord with some universal causal law is insufficient to establish that such laws may be provided for actions as well: Goldman reason- ably insists that knowledge of such laws is not necessary to justify construing wants and beliefs as causing acts (p. 72), but he

Page 13: Alvin I. Goldman, A Theory of Human Action

360 REVIEWS

somehow supposes that the concession bears on providing causal explanations of what may not be subsumable under universal causal laws. He moves on to support Davidsoti’s well-known thesis that “reason-explanations are a species of causal explana- tions” (p. 77). But, like Davidson himself, he fails to compare the properties of both sorts of explanation and settles the issue by supposing that “reason-explanations” are explanations given for actions performed by an agent for the reasons he has-where, ex hypothesi, to have a reason in acting entails that the reason “had” plays a causal role in producing the action (pp. 79-80)! Finally, he offers an entirely fair criticism of the notion of agent-causation (associated with the views of Chisholm and Richard Taylor) (pp.

I shall say very little about the remainder of the book-which is indeed provocative and worth the most careful scrutiny. My intention, here, is to give a fully rounded account only of Gold- man’s theory of action, and that is substantially confined to the first three chapters. Chapter 4 introduces Goldman’s theory of wants, which, very interestingly, elaborates a distinction between occurrent and standing (dispositional) wants (and occurrent and standing beliefs) substantially borrowed from William Alston (pp. 86-87). Roughly, the emphasis here is on showing that mental events cause acts (p. 91)-by showing that only occurrent wants and the like play the causal role (p. 88), that wants are not mental acts (p. 93), and that wants (and beliefs) are not “language- bound” (p. 94). Goldman tries his hand at interpreting the Aristotelian notion of practical reasoning (“practical inference”), introducing optatives to convey the content of a want (pp. 99- 109). His intent is to defend a causal theory of inference (which really is benign enough-in the sense that causes operate in in- fr-tlce : Goldman does not, however, provide an explanatory model of inference that could be called causal); but he seems to regard “cognitive” and “practical inference” as alternative forms of inference related to belief and want, respectively) (p. loo), without sufficiently emphasizing that both must embody certain formal or logical relationships if we are to distinguish between inferences causally effected (believing or wanting P because of other beliefs and wants) and mere causal sequences that are not inferential at all. Put this way, there need be no fundamental formal differences between the two sorts of reasoning. There is also a promising sketch of what I should call a model of minimal rationality (not characterized so by Goldman), of the inter-

8 1-84).

Page 14: Alvin I. Goldman, A Theory of Human Action

REVIEWS 361

connections among the concepts of acts, wants, and beliefs-in virtue of which Goldman neatly dispatches Melden’s well-known error (and related errors) regarding the causation of actions (pp. 109-11 1) : here, he usefully concludes that “the concept of an occurrent want is the concept of a mental event that tends to result in behavior”; that it is, therefore, “a logical truth that wants tend to cause action”-which he contrasts usefully with the Wittgensteinian position (pp. 112, 114-1 15).

Chapter 5 attempts to show, chiefly in the context of a survey of views among empirical psychologists, that the usual accounts (even that of B. F. Skinner) are by no means “incompatible with commonsense explanations which rely on . . . the causal role of wants and beliefs” (p. 129). But, in fact, in reinterpreting Skinner’s material (quite fairly), Goldman reminds us of his definition of ‘action’, to the effect that “no event is an ‘act’ unless it is caused by wants and beliefs (at least by wants)” (p. 153). He has not quite appreciated the conceptual question facing the causal theory of action (or, of intentional action), as has already been noted. He does-incidentally opposing Normal Malcolm’s thesis about the exclusiveness of neurophysiological explanations and explanations in terms of desires and intentions (pp. 157-162b-return to the question, but in a way that confirms the challenge already posed. For he says that “purposive behavior, it may be expected, is a function of events occurring in the central nervous system alone, independent of the autonomic nervous system and the sympathe- tic nervous system” (p. 166; italics added); but he later qualifies this, given particular counterinstances, holding that, in the counterinstance (of fidgeting resulting from one’s desire to attract the speaker’s attention), “the route of the causation has involved the sympathetic nervous system in an essential way; it has not been oriented exclusively within the central nervous system” (p. 167; italics added), which simply shows (by the use of decidedly qualified terms) that the analogues of Gettier-like counterinstances presuppose an explanatory theory that has yet to be supplied. And, in any case, Goldman provides no grounds for entertaining a reductive theory or for eliminating the inten- sionality of inferential, justificatory, and rationalizing contexts.

Chapter 6 offers an ingenious defense of determinism against the attacks of an “anti-predictionist”, that is, one who holds that “determinism entails predictability” and that, given “some suit- able sense of ‘possibility of prediction’ . . . closely related to, if not entailed by, determinism, . . . it is impossible for acts to be

Page 15: Alvin I. Goldman, A Theory of Human Action

362 REVIEWS

predicted” (p. 171). Goldman shows both premises to be false. With all of this I have no quarrel. But the fact remains that the argument is not brought to bear on the central issue of determin- ism with respect to acts, in particular, with respect to voluntary acts. The issue is fundamental. For, Goldman admits that “to say that an event is determined is to say something that implies the existence of universal laws pertaining to that event’’ (p. 170; cf. p. 172); but he nowhere shows that there could be universal laws governing human acts, in particular, voluntary acts. The problem has two sides. For one thing, the concept of an act is-to use Ryle’s phrase-polymorphous, in that acts of most kinds may be embodied in different physical movements and, for given kinds of acts, there are no kinds of movements that are clearly necessary for the performance of the kind of act in question. To concede that there are universal laws that govern purely physical move- ments, therefore, does not a t all show that there are universal laws that govern acts as weI1. For another, a great many kinds of acts-particularly, those that we call voluntary-count as be- havior that is performed in accord with conventions, customs, institutions, and rules. But all of these are subject to violation, revision, change. “Laws” governing behavior that conforms with given institutions and rules (say, through habituation) cannot, for that reason alone, be universal in form; and that there are uni- versal laws governing human conformity with any set of institu- tions or rules-especially when such conformity (by way of performing acts) is to be specified in accord with polymorphous concepts-seems at least doubtful. Goldman’s thesis about deter- minism requires that universal laws of nature be formulable in principle for human actions. He himself says that he has to omit any discussion of how a scientific predictor intent on writing a “book of life” about an agent (possibly in advance of that agent’s life) might gather the relevant deterministic data, “because I do not know all the laws which the predictor would have at his dis- posal” (p. 185). But the truth is that Goldman does not know of any deterministic laws governing human actions. Presumably, there could not be any such laws unless a t least some form of physicalist reduction were defensible-and that has not been shown and is not at all obvious.

Finally, Goldman seems to have missed the force of Richard Taylor’s charge that determinism is incompatible with it being in one’s power to do A and to forego doing A (p. 195): he seems to have supposed that if doing A and foregoing doing A are “logic- ally compossible” with one’s wants, deliberations, and the like,

Page 16: Alvin I. Goldman, A Theory of Human Action

REVIEWS 363 the alternative sets of one’s wants, deliberations, and A and of one’s wants, deliberations, and foregoing doing A are “causally compossible” sets (p. 196). But this is ruled out by his own notion of causal compossibility, for he holds (reasonably enough) that, “assuming determinism, every non-actual event whatever is causally incompossible with some set of actual prior events” (p. 180). The matter is puzzling because Chapter 7 is devoted to an elaboration of the sense in which “the fact that an act was causally necessitated does not entail that [an agent] was not able to do otherwise” (p. 197). Goldman succeeds in providing such a sense (pp. 201-2051, but it is a sense that cannot be squared with the deterministic position of the preceding chapter. In fact, according to Goldman’s final formulation, “S’s being able, a t t, to do A, by t’ does not depend on S’s actually performing the sequence of acts from A1 to An-,. S is able, at t, to do A,, by t’ just in case he is able to do A, at t , and i f he were to perform A, at t l , this would, together with the state of the rest of the world, cause him to be able to perform A, at t,, etc. Thus, even if S fails to perform A, a t tl (or fails to perform any of the other acts A, at tl), it is still true that he is able, at t , to perform A, at t’ ” (p. 205). But though this sense of ‘able’ is- viable and even of interest, it is flatly incompatible with the sense of “causally in- compossible” already introduced. In that sense, if I am caused to be disinclined to do A, then I am unable to do A, even if-in Goldman’s sense-I am able to do A but am disinclined to do A. The issue has some interest because, in this chapter, Goldman wishes to show that his theory of ability fits usefully in the con- text of excuses, defenses, and evidence of constraint with respect to moral, legal, and related disputes. He fails, however, since he all but misses the central issue-whether there is any sense in speaking of responsibility (and of distributing blame and the like) in a context in which determinism is granted full sway. It is clear that, on Goldman’s definition of ‘ability’, i t is quite possible to blame or defend someone for what he has done even if what he has done is causally necessitated (p. 214); but Goldman simply never comes to grips with the question whether this is a thesis that meets in a fair way the objection to speaking of responsi- bility where determinism prevails. Isn’t it just true that responsi- bility presupposes that, a t critical junctures, A and foregoing A are, alternatively, causally compossible with S’s condition and circumstances?

The Epilogue collects the major theses, but I see no reason to

Page 17: Alvin I. Goldman, A Theory of Human Action

3 64 REVIEWS

qualify the foregoing criticisms. Also, there are other funda- mental philosophical puzzles that Goldman’s book raises-most particularly, the scope of intensional discourse-that would need a fresh start.

TEMPLE UNIVERSITY JOSEPH MARCOLIS