20th WCP: Actions, Intentions, and Awareness and Causal Deviancy

The intentions with which a person acts are her reasons for acting, and those reasons are given by her antecedent attitudes: beliefs, desires, intentions, choices, decisions, hopes, fears, wishes and so on. Attitudes function as reasons for acting through the combination of pro-attitudes towards certain states of affairs (or propositions) with beliefs that the desired states of affairs are realisable and about how they can best be realised by acting. The causal theory of action claims in addition that the reasons for which an agent acts are what causes her so to act: without the causal relation, it is argued, we have no way of making sense of the distinction between an agent's real reasons for acting and mere rationalisations of her actions. (1) According to the causal theory, therefore, actions are caused by reasons which also rationalise what they cause. The causal theory, as it stands, is famously incomplete, for a combination of attitudes that provides a reason for acting may cause some behaviour that realises it, and yet the behaviour so caused is not an action. In the standard examples of 'internal' (2) or 'basic' (3) causal deviancy an agent either intends to perform an action or has a desire for something he knows can be accomplished by a given action, but the thought of performing the action causes him to be unnerved and this, as luck would have it, causes him to behave accidentally in just the way he had desired or intended. Thus, in Davidson's seminal example, a climber wants to rid himself of the weight and danger of holding another man on a rope and knows that by loosening his hold on the rope he can rid himself of the weight, but the thought of the contemplated act so unnerves him that he lets go unintentionally. (4) Likewise, with Frankfurt's man at a party, who must signal the commencement of a robbery by spilling what is in his glass, but whose nervousness, when the appointed time comes, causes his hands to tremble and to spill his glass unintentionally. (5)

Causation of behaviour by intentional states that rationalise it is not enough for it to count as acting, therefore: the behaviour must be caused, it is thought, in 'the right way'. (6) The principle difference among the various accounts of what is lacking in cases of basic causal deviancy is between those that argue that it is an intention or attitude that is causally sufficient for the occurrence of an action (or that causation of an action by such a state leaves no causal space for deviancy) and those that hold that it consists in the right kind of causal relationship between intentional states and behaviour. The first group encompasses those accounts in which actions are caused by volitions or special categories of intention. (7) The second group brings together those in which intentional actions are characterised as having a counterfactual or differential sensitivity or responsiveness to the content of the initiating intentions or attitudes. (8)

A key motive for the causal theory of action is to say what it is that distinguishes actions, as a class of events, from involuntary behaviour. It is an assumption of both approaches to the problem of basic causal deviancy just mentioned that the key difference between actions and involuntary behaviour consists in the relationship of actions to prior intentions or reasons. In this paper I will challenge this assumption and propose an alternative account of the causation of actions in which the key difference between actions and involuntary behaviour is identified as intention in acting.

In both of our well-known examples of causal deviancy the agents are caused to behave as they do because of nervousness. What causes the nervousness? Indirectly, of course, it is caused by the agents' intentions or beliefs and desires. But the immediate cause of the nervousness in each case is the agent's awareness or contemplation of what he is intending or thinking of doing, which is either caused by or implicit in his awareness of his intentions or beliefs and desires. No doubt alternative examples can be constructed in which nervousness is replaced by over-excitement, neurosis or whatever, but the essential feature of internal causal deviance of the kind envisaged in Davidson's and Frankfurt's examples is that some thought, contemplation or awareness of the imagined or intended action gives rise to unsteadying feelings and these feelings cause the imagined or intended behaviour to be brought about unintentionally. (9) The general pattern of this kind of internally deviant causation, therefore, is that an intention to act, or a desire coupled with a belief about what behaviour will best realise it, causes the agent to think about the intended or recommended behaviour, which in turn causes an unsteadying state such as nervousness, which in turn causes the behaviour to be realised unintentionally. A plausible response to this is that causation in the right way calls for something that cannot cause the agent to think about what she is intending or contemplating and therefore cannot cause the unsteadiness that produces causally deviant behaviour. Assuming that attitudes are intentional states and that intentional states are necessarily states of which we can be occurrently aware, the only way in which an attitude(s) could satisfy this requirement is if its occurrence causes an action immediately, or almost immediately, and therefore leaves no time for it to cause the agent to think about what she is intending or contemplating before she acts.

This is, indeed, what has been claimed for the kinds of intentions and volitions mentioned already, which are thought to be causally sufficient for intentional actions. On Dorothy Mitchell's account, for example, actions are caused almost immediately by concurrent intentions. (10) Since the interval between the occurrence of such intentions and the behaviour they cause is thought to be so short that they are experienced as accompanying the behaviour, there is no time for it to be brought about via a deviant route. Mitchell's account is rejected by John Bishop who argues that there is no reason to accept that a concurrent intention cannot, 'at the very last minute', cause the agent to behave involuntarily just as intended. He also argues that even though a concurrent intention might be experienced as accompanying the intended behaviour, agents might still be able to tell the difference 'by the absence of the normal sense of exercising control over their bodily movements'. (11)

A further reason for scepticism about the claim that actions are always caused by intentions of a kind that cannot fail to result to result in actions is that the claim is ill-motivated. The reason for thinking that the right kind of causation is causation by intentions is that intentions are thought to function as a kind of reason for acting. This is debatable in itself, (12) but in any case it is difficult to see how a state whose occurrence leaves no room for thought or endorsement before action is initiated can be thought of as providing any kind of reason for acting. We think of reasons as proscribing or recommending actions, but if an intention were such that an agent could not fail to act on it, once it has been formed the time would be past for proscription or recommendation. It might be argued that the imagined class of intentions can function as practical judgments on an agent's prior reasons for acting, but the same can be said of actions themselves. A practical judgment made while it is still possible not so to act can be thought of as a reason for acting, but not when the action is actual or inevitable. (13)

If we retain the supposition that the right kind of cause (hereafter RKC) is something that cannot cause the unsteadiness that can lead to causally deviant behaviour (or cannot cause behaviour via unsteadiness), and if intentions are ruled out as candidates, what else might it be? The answer is suggested, I think, by what it is about intentions (and attitudes generally) that makes Davidsonian/Frankfurtian deviant causation possible: that they are states of which we can be aware. My suggestion, then, is that RKC is a state, event or entity whose operation or occurrence an agent cannot be directly and occurrently aware of. The suggestion seems counterintuitive at first blush: how can something the average agent does not intend or know of ground the distinction between intentional actions and unintentional behaviour? My answer would be that it could do so if it were suitably responsive to the agent's attitudes. What this suggests is that RKC is an entity - a mechanism - that is suitably responsive to attitudes and whose operation agents are not directly aware of (which is to say, agents are not aware of it as an operative mechanism whose operation is distinguishable from the behaviour it causes and therefore they have no experience of its operation that is distinguishable from the experience of acting). What suitably responsive means here is that RKC is triggered by the agent's intentions, preferences, reasoning and judgments, and that it responds to them by causing the body to behave in the way indicated by her intentions or her beliefs about what action will realise those preferences or judgments. The difference between intentional acting and the kind of causally deviant behaviour given in our seminal examples, therefore, is that in the latter awareness of what the agent is intending or contemplating causes him to be unnerved and this brings about the contemplated behaviour via a causal route that does not include RKC. There is no possibility that this kind of causally deviant behaviour might be initiated by RKC since the agent is unaware of its operation.

Although causation of actions by RKC, thus stated, gets us over the problem posed by Davidson- and Frankfurt-type examples of internal causal deviancy, there are well-known objections to the causal theory of actions that can also be pressed, mutatis mutandis, against this version of it. In the first place, if the function of RKC is merely that of causal mediator between beliefs, desires, intentions, preferences, reasonings and judgments on the one hand and actions on the other, we face the problem that agents do not always act in accordance with their reasoning and judgments about what is best to do and neither do they always act in accordance with their intentions. Moreover, even if they did, the account is open to a further objection that has been pressed against standard versions of the causal theory, which is that our conception of ourselves as agents is not that we are straightforwardly caused to act by reasons, judgments and intentions, but rather that in acting or in forming intentions we are influenced by reasons, and that we execute intentions by acting. (14) If RKC is responsive to reasons, judgments and intentions, only in the sense that it is causally triggered by, say, intensity of desire, then it would lack the kind of independence from reasons, judgments and intentions that we require for something that can be said to occupy the agent's intermediate role between motives and reasons and intentions and between intentions and actions.

There is no reason, however, why RKC need be responsive solely to strengths of preference, or to judgments based on strengths of preference. Its responsiveness to reasons could be conditioned by the agent's past behaviour: in other words, by training and habit. If this is correct, there would be no reason to accept the Hobbesian claim that agents always act on their strongest desires. If I fail to act as it seems to me that I most desire to do, this may be because RKC has been conditioned by habit to be more responsive to some weaker desire rather than my being mistaken about what I most desire. Thus, the causal role of RKC is not that of simple mediator, being triggered by attitudes to cause actions, but that it responds to reasons according to their respective strengths, not solely or primarily according to strength of desire but also in relation to our existing dispositions and habits, which gives it sufficient independence from reasons for us to be able to say that it independently selects from among reasons. (15) Moreover, if the independence of RKC in relation to reasons is a product of the way it has been conditioned by training and habit, this allows a place in our account for the idea that actions bear some relation to character. Therefore our conception of ourselves as agents, in which we give effect to reasons and execute intentions, can be thought of as being functionally realised by RKC.

Another objection to the idea that RKC is a mechanism whose operation the agent is not aware of concerns the fact that when we act, we do not need to work out or infer that we are acting, or to be told that we are acting. Actions have what Ginet describes as the actish phenomenal quality. (16) But if I am not aware of the operation of the mechanism that causes me to act, how can it be that I am so immediately aware that what it causes is an intentional action rather than unintentional behaviour? We can go some way to dispelling the air of paradox by thinking about the phenomenal character of our agentive awareness of acting. If you make a fist, for example, all you will be aware of is that you make a fist: you will not be aware of being caused to do so, or even of yourself as causing your fingers to clench. When one makes a fist it is not as if one issues a signal or a command to one's fingers to clench, in response to which one's fingers behave as one intends. The content of one's awareness of a basic bodily action such as this isn't easy precisely to describe, but we can say that it is as if one is somehow inside or intimately involved in the behaviour. (17) The description is groping and vague, and it is easier to say what the content of the awareness is not: it is not as if one experiences oneself, one's fist clenching and one's contribution to it as discrete relata. One can examine the clenching of one's fist as if it were something distinct and separate - by thinking of it as if it were an involuntary movement - but that requires an attitude of imaginative distance: it is not how we typically experience an action like making a fist.

There is, however, a further positive addition we can make to the description of agentive awareness of actions, which is that the sense of being somehow inside or involved in intentional behaviour is such that the agent is in control of it. As I contemplate the clenching of my fist, it is not as if I am involved in it as I might be involved in the movement of something that carries me along with it - like that of a train on which I am a passenger. I am, so it seems, in control of what my hand is doing, and that sense of control is indistinguishable from my sense of being inside or involved in what I do. The content of our agentive awareness of basic bodily actions is, I think, one of the philosophical motives for agent-causationist resistance to the event-causal theory of action. (18) Agent-causationists notice that we do not experience our actions in the way that we experience ordinary event-causation, and think that the difference can be captured by the idea of agent-causation. But the content of our agentive awareness of bodily actions is not one that can be captured by any idea of causation, since causation, whatever the metaphysical variety, implies a relation between discrete items and our agentive awareness of basic bodily actions contains no such relation.

If the content of our agentive awareness of basic bodily actions has this form, it is not to be wondered that we are not directly aware of the operation of RKC, for if one experienced one's actions as being caused by RKC, one would experience them as dependent on it. If that were the case, then any sense of control one might have over one's actions would have to be via a sense of control over RKC, which would be incompatible with the direct and intimately involving sense of control that is contained in our agentive awareness of basic actions.

But if actions are always caused by RKC, and if our actions are always characterised by the distinctively agentive awareness I have described, what is the connection between the two? If RKC responds to attitudes by causing the body to behave in the way indicated by the agent's intentions, or by her beliefs about what action will realise her preferences or judgments, it is reasonable to suppose that it does so by means of an action schema that signals the agent's muscles to do whatever is required to realise the preferred action type. Goldman has suggested that the 'feeling of voluntariness' is a product of match up between the action schema that initiates bodily movement and afferent feedback from the muscles. (19) When we act we are not merely aware that we are acting, however, but always that we are acting in a particular way. As Davidson puts it: 'an agent always knows how he moves his body when, in acting intentionally, he moves his body, in the sense that there is some description of the movement under which he knows that he makes it'. (20) Agentive awareness of involvement in actions, I would argue, is an awareness that one is acting (or trying to) in a particular way, where this involves not only immediate awareness of what one is doing but also immediate awareness of the extent to which one is doing it successfully. The sense of control, on this account, has two components: an awareness of what it is that one is doing - one's intention in acting - and also that one is doing it (e.g. that one's muscles are behaving as they should), the first component deriving from the action schema and the second from the match up between afferent feedback and action schema.

An important upshot of the account just given, in which an agent's intention in acting consists in her awareness of what she is doing and is fixed by the action schema that initiates bodily movement, is that there is no necessary relationship between an agent's intention in acting and any prior intention that might have caused it. Suppose, for example, that you become aware of impending cramp in your left knee. To head this off you decide to do some step-up step-down exercises at the bottom of the stairs. As you make your way to the staircase you begin thinking over some unrelated problem and are so engrossed in it that before you know it you have climbed halfway up the stairs. It was never your intention to do that and you had no reason to, but it would be wrong to say that your steps were involuntary. It would not have been as if your limbs suddenly developed a will of their own or that you lost control of them in any way. Neither would your behaviour have been irrational, as such, since there is an explanation in terms of absentmindedness and what one routinely does (and routinely has reason to do) after one climbs the first step of a staircase. Although you would have had no prior intention to do what you did and you would not have been acting with an intention, therefore, you would have acted intentionally. If you had not been acting intentionally - if you had not had an intention in acting (climbing the stairs) - your movements would not have been coordinated in the appropriate stair-climbing way.

Having arrived at this elaborated account of RKC, I can now respond to another objection to it, which is that while it might, in virtue of our not being directly aware of the operation of RKC, rule out deviant causation in its Davidsonian and Frankfurtian forms, cases of deviant causation of behaviour can be imagined in which agentive awareness of intentions or contemplated actions plays no essential part, involving what Bishop has described as heteromesial causal chains. (21) The deviant route might, for example, be brought about via the intervention of a neuroscientist, who, on detecting the presence of an intention or belief-desire pair, and wishing to ensure that the intended action actually takes place, guarantees its doing so by some form of neural manipulation. (22) Nothing about RKC as specified rules out the possibility of external behavioural manipulation via a route that includes RKC. In that case, it may be argued, causation by RKC is insufficient to distinguish acting from involuntary behaviour, since behaviour brought about by external manipulation cannot be thought of as acting.

In reply, let us suppose, first, that a neuroscientist intervener causes RKC to be triggered by a belief-desire pair but that if the intervener had not intervened, RKC would not have been thus triggered, because its conditioning by training and habit had not been such as to make it responsive to such a belief-desire pair. In that case the intervener's actions would have (partially) disabled RKC by interfering with its pattern of responsiveness to attitudes. This is no mere technical objection, since the effect of the intervener's actions would be such as to insulate RKC and the behaviour it causes from the agent's prior training and habits - including, of course, her past actions, choices and decisions - and therefore from her character. RKC would still have responded, after a fashion, to the agent's attitudes. Moreover, the behaviour it caused might still feel to her like acting, according to the account set out above, because, for all we know, the intervener could ensure the selection by RKC of an action schema that issues the appropriate motor signals, which results in an appropriate match between afferent feedback and action schema. For those reasons, it would be inappropriate to class the resulting behaviour along with that of blood circulation, spasms and so forth. And if it still felt like acting, it would also differ from the kinds of causally deviant behaviour set out in Davidson's and Frankfurt's examples. Nevertheless, in view of the impairment of RKC and the consequent insulation of the behaviour from the agent's character, the behaviour might best be described as quasi-acting. (23)

If we suppose instead that the intervener is able somehow to take account of the RKC's established responsiveness pattern, and his intervention is such that the behaviour he causes via RKC is exactly what would have been caused without his intervention, then we can no longer say that RKC, and the behaviour it causes, is insulated from the agent's character, since although the agent's past training and habits do not directly affect RKC's responsiveness to her belief and desire, they continue to have the same effect, via the deviant route that includes the actions of the intervener, as they would have had if the intervener had left things as they were.

One response to this kind of scenario is just to tough it out and argue that since the intervener's actions make no difference to the outcome, he has simply turned himself into a (rather pointless) kind of prosthetic device. As John Bishop has argued, it is possible to imagine cases of prosthetic additions to the central nervous system aimed at restoring an agent's ability to act. The causation of actions by such a prosthetic would certainly deviate from a normal person's action-causing mechanism, but this would not intuitively be a case of causal-deviance in the sense that concerns us. It is possible to imagine that the neuroscientist's intervention might be aimed at giving effect to the agent's wants and intentions in just the same way as the imagined prosthetic. Why, then, should we regard the results of his intervention any differently than we would those of the prosthetic? If we imagine, instead, that the agent was not disabled prior to the intervention, but the intervention makes no difference to what would have happened otherwise, why should we regard that outcome any differently?

One reason is suggested by Bishop, who argues that since it is up to the intervener whether to give effect to the agent's intentions or block them, the agent lacks control over the outcome. The kind of control Bishop has in mind differs from that involved in the agent's sense of control mentioned earlier. That sense of control is an awareness of what one is doing together with one's awareness that one is doing it successfully, and one would normally be thought to have such control if one's awareness is veridical. Both the kind of control involved and the awareness of it could be satisfied in cases involving the kind of intervention we are imagining. One way in which an agent subjected to such intervention might be thought not to lack control is that, since the outcome is in the hands of another, she could not rely on it regularly to produce the intended or desired behaviour. But it is possible to imagine other cases in which behavioural outcomes are unreliable, or even unlikely, that should not be described as deviant. Consider the following example from Christopher Peacocke:

There could be a substance in the brain that is, for a given person, only fortuitously present, and which stops neural messages becoming scrambled: if this substance were absent, it could be reliably the case that some bodily movement were made, though there would in general then be no matching with the intention. So if the substance were not reliably present, a conditional reliability requirement would not be fulfilled. This would not prevent intentionality of the putative agent's bodily movements under appropriate descriptions when the substance did happen to be present. (24)

Peacocke presents a different reason for regarding behaviour brought about by a neurophysiologist as deviant. Following David Pears, he argues that examples such as this put a great strain on the concept of the agent as originator of actions. Peacocke's own response to the problem is to suggest a requirement that 'the chain from intention to bodily movement should not run through the intentions of another person'. (25) This kind of requirement is rejected by Bishop who argues that it is possible to imagine another kind of action-enabling prosthetic aid, with components lodged outside the agent's skull, 'which one day breaks down but is briefly repaired by having a second agent intentionally hold the broken wires together until they can be resoldered'. (26) Such an arrangement, rather than preempting the agent's control, as in the case of the neurophysiologist intervener, would enable it and therefore, again, we would not intuitively describe the case as deviant.

Bishop's own solution to the problem draws on Irving Thalberg's suggestion that the causal theory should require 'continuous' or 'sustained' causation in which an action is not only initiated by an agent's intention but also continuously regulated by it. (27) Bishop argues that such sustained causation in action can be explicated (in line with what I have suggested about the sense of control in acting) in terms of a control system (or servosystem) that regulates sustained actions by adjusting its outputs according to afferent feedback. (28) In the case of the neurophysiologist intervener, according to Bishop, sustained monitoring and guidance of bodily movements in conformity with the agent's intentions can only be carried out by the neurophysiologist because of the 'disconnection' by the neurophysiologist of the agent's efferent nervous system. As Bishop notes the monitoring and regulation of the agent's bodily movements by the neurophysiologist could itself be described as a control system of the appropriate kind. Nevertheless, such a 'system' could not count as 'realising the agent's controlled regulation of his or her bodily movements [because] the feedback information about orientation and muscular states does not get carried back to the agent's central processing system'. (29) Bishop therefore argues that the causal theory of action should include a clause to the effect that if the causal mechanism from intention to bodily behaviour involves feedback 'then the feedback signal is routed back to [the agent's] central mental processes if to anyone's'. (30)

As Bishop notes, afferent feedback of the kind required for continuous monitoring and regulation of an agent's movements in conformity with his or her intentions is not possible for actions that happen too quickly for feedback monitoring to take place (intentional blinking for example). (31) According to our account of RKC, however, although afferent feedback may have no role to play in the successful production of instantaneous actions, it would be required for the sense of control we have in respect of basic actions and, as I have suggested, this sense of control is intrinsic to, or identical with, the agentive awareness or 'actish quality' we experience in respect of our actions. In respect of instantaneous actions, given that afferent feedback has no role to play in the successful execution of intentional behaviour,it might be argued that the sense of control is, in a certain sense, epiphenomenal. (32) Where an instantaneous action forms part of an action sequence, (33) however, it may well be that the sense of control is essential. And in respect of sustained actions it is difficult to imagine how effective monitoring and regulation could be accomplished in the absence of that sense of control. As I have indicated already, however, the sense of control that is involved in our agentive awareness of actions is different from the that which Bishop takes to be lacking in cases of pre-emptive neurophysiological intervention where realisation of the agent's intentions can only be effected through the intentions of the intervener. It appears, therefore, that the kind of control involved in Bishop's solution to the deviancy problem posed by preemptive neurophysiological intervention is of a different kind from that which was supposed to be lacking in such cases. The idea that the kind of control that is thought to be missing from this kind of case can be explicated in terms of unreliability has already been rejected. If Peacocke is right, it is bound up with the idea that agents are the originators of their actions. But however the idea of origination is understood it seems unlikely that it can be satisfied by a specification of the kind of control involved in sustained actions.

It is possible to imagine that the neurophysiologist only ever intervenes to initiate actions (rather than to block them) and that where the successful accomplishment of intended behaviour requires monitoring and guidance involving afferent feedback, this all takes place in the normal way without any subsequent intervention. In that case Bishop's proposed clause would be satisfied. Should we regard the products of such interventions as intentional actions? One reason for thinking that we should would be that in that case such agents would have a choice about whether to persevere with the action that the intervener has initiated. Perhaps, in that case, the kind of control required for sustained actions would also guarantee the kind of control that Bishop thinks of as missing in cases of preemptive intervention. I am doubtful about whether an intervener case could not be contrived that satisfies Bishop's clause but in which the intended behaviour is brought about deviantly. It is not necessary to construct such a case, however, since it is possible to imagine preemptive intervention in respect of instantaneous actions. Thus, the intervener might detect an intention to blink and ensure that the appropriate efferent impulse is triggered. In Bishop's words, the intervener's involvement would have been such as to block the agent's exercise of direct control over his or her bodily movements and yet the clause requiring feedback signals to be routed back to the agent's central mental processes would be satisfied because the intended behaviour does not require such feedback.

I have suggested that the most plausible reason for regarding cases of preemptive heteromesy (even where the intervention makes no difference to behavioural outcomes) as deviant is provided by Peacocke: that such cases are not compatible with the idea that actions somehow originate from the agent. Accordingly, Peacocke's proposed requirement that the chain from intention to bodily movement should not run through the intentions of another person seems intuitively to be the most plausible response, were it not for the fact that, seemingly, it would also render the kind of second-agent prosthetic assistance described by Bishop as deviant. In that case, we should consider whether there is a way of qualifying Peacocke's solution so that it does not count second-agent prosthetic assistance as deviant. One way of approaching this is to ask what it is about Bishop's example of second-agent prosthetic assistance that makes us reluctant to regard it as involving causal deviance. According to Bishop we should not think of it as deviant because 'the sensitivity of the causal links between intention and behaviour will be preserved since it is quite irrelevant whether the wires are held or soldered together'. (34) In what sense irrelevant? Certainly not in the sense that the second-agent's intentions would be irrelevant to the outcome, since she might, on moral grounds perhaps, refuse to hold the wires together in order to forestall some action. It is irrelevant in the sense that although the second-agent's intention to hold the wires together is a necessary condition for the causal route from the agent's intention to her action, the causal route does not pass through the second agent's intentions: it passes through the wires. In that case, the second agent's actions and intentions are no more threatening to the status of the outcome as acting intentionally than many other actions we can think of that require the actions of others. What is different about the kind of preemptive intervention we see in the case of the neurophysiologist intervener is that the causal route from intentions to action necessarily does pass through the intentions of the intervener.

Suppose then that we imagine that the intervener creates a circuit breaking switch somewhere along the route from intention to action. Suppose in addition that the intervener sometimes uses the switch to censor actions. How should we regard behaviour that the intervener has permitted? I can see no reason for regarding such behaviour as causally deviant. We do not, after all, regard someone who is physically restrained from acting in some particular way as not being able to act in other ways that are permitted by the restraints or as not acting when the restraints are removed.

There is therefore no need to qualify Peacocke's stipulation (35) that the causal chain from intention to bodily movement should not run through the intentions of another person, since Bishop's counterexample, although it involves dependency on the intentions of another, does not involve a route through those intentions. Moreover, if the right kind of causal route begins with RKC (or that the right route from intentions and attitudes is via RKC), as I have specified it, rather than, as Peacocke suggests, with intentions, another possible objection to his stipulation - that it is ad hoc - can also be set aside. Peacocke's general solution to the problem of causal deviancy is to specify a kind of differential/functional responsiveness of intentional action to prior intention: clearly the notion of agent-origination that Peacocke's stipulation is meant to satisfy draws on rather different intuitive considerations from his general solution. But if, in the first place, RKC is defined as suitably responsive, in the way I have described, to the agent's attitudes, then any route between attitudes and the operation of RKC that went through the intentions of a second agent would rule out that sensitivity. And if the right kind of causal route from attitudes to actions is via RKC and actions are appropriately sensitive to RKC, then since RKC is a post-attitudinal causal mechanism, and one whose operation the agent is not directly aware of, there is nothing ad-hoc in the stipulation that the causal route from RKC to actions should not go through intentions. It would be incompatible with the hypothesised characteristics of RKC not only that the causal route from RKC to intentional behaviour should run through the intentions of a second person, but even that it go through an intention of the agent herself.

While the stipulation that the causal route from attitudes (via RKC) to actions should not go through the intentions of a second agent (and that the route from RKC to actions should not go through anyone's intentions) is not ad hoc, however, it is conceivable, and has already been granted, that RKC could be manipulated and also that the normal route from RKC to actions could be replaced by a deviant route involving some form of external interference. The possibilities of post-RKC causal deviance will be excluded, provided that the relationship between intentional behaviour and RKC is appropriately sensitive and that it does not go through intentions (or through attitudes more generally) and the characteristic responsiveness of RKC to the agent's attitudes excludes the possibility of causal deviance through manipulation of RKC. Moreover, whatever the details of a correct theoretical account of the sensitivity requirement, it should be satisfied by the characteristic responsiveness of RKC to attitudes and by the match-up mechanism described above between afferent feedback and action schema, which, I have suggested, is the basis of the sense of agentive involvement and control in intentional behaviour. At any rate, I can think of no deviant counterexamples that would be true of causation by an unmanipulated RKC together with afferent feedback match-up, while not running through the intentions or attitudes of a second agent in the route from attitudes to behaviour.

However, the key characteristic of RKC, by means of which it avoids Davidsonian and Frankfurtian types of causal deviance, was that the agent is not directly aware of its operation and therefore that its operation cannot lead to the nervousness (or whatever other states might cause unintentional behaviour) that is responsible for those types of deviance. The question therefore arises of whether, if this characteristic only enables us to avoid those forms of causal deviance and there are other strategies that might avoid all forms (such as Peacocke's or some other version of the sensitivity requirement), RKC is redundant. In other words, if all cases of causal deviancy can be avoided by means of some sensitivity strategy, which excludes fortuitous causation of intended behaviour by nervousness, freak acts of nature (36) and so on, and which can be specified without reference to lack of direct agent awareness of the operation of a specific mechanism in the causal route from attitudes to intentional actions, why need we bother with such a mechanism?

In reply, I want to show that there are important differences between Davidson's and Frankfurt's examples of causal deviancy and intervener or post-RKC examples, which show up an important misconception about intentional action in the sensitivity strategy and, more generally, in all attempts to characterise the nature of intentional action in terms of a relationship between behaviour and antecedent attitudes. One obvious feature of Davidson's and Frankfurt's examples of causal deviancy that distinguishes them from those involving neurological manipulation, freak accidents with nerve-endings and so on, is that what happens in them, although we would consider it remarkable were we to encounter or hear of such things, is plausible enough in relation to what we know from experience about nervousness and unintentional behaviour. This is related to a further difference between the two kinds of case, which is that the Davidson and Frankfurt examples do not specify some feature of the behaviour-causing route - like neurological manipulation - that renders the route deviant: rather we know from experience that nervousness sometimes causes unintentional behaviour, we can see that unintentional behaviour caused by nervousness might realise whatever attitudes brought about the nervousness and we know (or we imagine) that an agent in the situation of Davidson's climber or Frankfurt's robber would know immediately that their nervous behaviour was not acting or that they would not experience it as acting. If we are right in thinking that the agents in the two examples would know that in letting go of the rope or in spilling the glass they were not acting intentionally (and also that if instead, and despite his nervousness, the climber let go of the rope, or the robber spilled his glass, intentionally, he would have known, just as immediately, that he was acting intentionally), we face a question about how they would know this. Since the nervous behaviour in the examples is caused by attitudes that it realises, neither agent could infer the unintentional character of their behaviour from the bare relationship between it and their antecedent attitudes. Nor, for that matter, could they infer it from their nervousness, since, as Adam Morton has pointed out, it is possible for actions to be partly caused by nervousness. (37) Therefore, they could only know it by means of some other feature of the way the behaviour is caused or by some intrinsic feature it lacks that is present in acting intentionally, or both, and it is implausible to suppose that this feature is some accidental or secondary property of acting intentionally. I have suggested that agents know that they are acting (as well as how they are acting) through their awareness of the action schema initiated by RCP and of the match-up between it and afferent feedback from the muscles, that this is the basis of our agentive awareness of what we are doing and of whether we are doing it successfully and that it is constitutive of our sense of control in acting. I now want to argue that the key feature of acting intentionally is the intention with which one acts, that this is given by the action schema initiated by RKC and that it is this, rather than any relationship - however specified - between behaviour and prior attitudes, that is the mark of acting intentionally.

If action counted as intentional in virtue of being caused by antecedent rationalising attitudes, then the example of unintended stair-climbing mentioned above could not count as acting intentionally, since the stair-climbing was not caused by any prior intention or pro-attitude that would have rationalised it. Causation by antecedent rationalising attitudes provides the intentions with which one acts, rather than one's intention in acting and it is one's intention in acting that determines that one is acting intentionally. The point can be usefully illustrated in relation to a scenario described by Christopher Peacocke:

As Freud's servant I form on Monday the intention to break the master's vase on Wednesday when I pick it up. Call this intention i. But I also know that I ... will become extremely nervous once the vase has actually been picked up. I know too ... that an inevitable consequence of such nervousness is a loosening of the grip and a smashing of the vase. So I know that intervening and operative nervousness will not prevent attainment of my goal. On Monday I decide also, since I am bad at remembering how my more devious plans are meant to work, that I must continue to have an ordinary intention to break my master's vase throughout Tuesday and Wednesday for it to become operative. (I also loose the knowledge about the intervening nervousness and its effects.) I am successful in continuing to have the intention of breaking the vase in the normal way up to the relevant time on Wednesday, when I pick up the vase and then drop it out of nervousness. Now suppose intentions are 'individuated' not only by their content, but also by when they are possessed (and perhaps by more too); and let us call my intention of Wednesday to break the vase in the normal way 'i'' ... Was the event e of my releasing my grip intentional or not? When we reflect both upon what actually happened, and my quite reliable plan of Monday, we can no more give a non-relative answer to this question than we can to the question 'Is New York north?' when there is no tacit fixing in some way or other of some place relative to which the question is being asked. We ought to say that e is intentional with respect to intention i but not with respect to intention i'. (38)

Peacocke's question about whether Freud's servant's releasing his grip was intentional or not is ambiguous. It could mean either 'did he intend that his grip would release?' or 'was his releasing his grip an intentional action?'. Obviously, he did intend on Monday that he would release his grip on the vase on Wednesday, whether (as we might say) deliberately or through nervousness, and therefore the answer to the first question is yes. But releasing his grip was, categorically (i.e. however described and however relativised), not an intentional action. Freud's servant could intend that he would release his grip, even though releasing his grip, when it happened, was not an intentional action, because agents can intend things that are not actions provided those things can be brought about or made more likely by actions. Thus, Freud's servant could intend to release his grip, even through nervousness, because he was capable of an action (picking up the vase) that could lead to his releasing his grip and because he knew that it could. Not only, therefore, can events that are not actions be intentional, in a specific sense (and under specific descriptions), but they have that status only in virtue of being brought about or made more likely by intentional actions proper. What it is that makes actions intentional, and which distinguishes them from involuntary behaviour, is not to be found in their relations to prior intentions or to other antecedent attitudes. (39)

It would be mistaken to conclude from this that the relationship between actions and prior attitudes is a secondary or inessential feature of the status of actions as actions. Prior attitudes, as noted at the beginning of this essay, provide an agent's reasons for acting, and the reasons for which an agent acts are a proper part of at least one description of her action. (40) For this reason, an account of RKC that is restricted to what it causes, without reference to what causes it to operate - what it is responsive to - would be incomplete. If actions are events for which there can be reasons and if actions are caused by RKC, RKC must (in line with Davidson's argument that the notion of acting for a reason can only make sense as a causal relation) be causally responsive to prior attitudes. (41) But the intentionality of actions is what makes it possible that actions be carried out for reasons and therefore we must look to the characteristics of RKC, as set out above, and the relationship between it and the behaviour it causes for an explanation of that property, rather than to prior attitudes. RKC, as a mechanism with the characteristics set out above, therefore, cannot be eliminated either from a resolution of causal deviancy or more generally from an adequate account of the intentionality of actions.

Perhaps some will shudder at the thought that what distinguishes actions from unintentional behaviour is a mere mechanism, to say nothing of the idea that it is a mechanism whose operation we are not directly aware of. But after all, thus characterised, RKC is nothing other than the Will. If our actions are always caused by our wills, and if our wills are suitably responsive to what we want, decide and intend, and if causation by the Will, as I have described it, can account for our agentive awareness of actions, is there anything in the account that is inconsistent with the belief that our actions are our own doing?

Notes

(1) Davidson 1980a, pp. 9-12.

(2) Davidson 1980b, p. 79.

(3) Bishop 1989, p. 140.

(4) Davidson 1980b, p. 79.

(5) Frankfurt 1978, p. 157.

(6) Davidson 1980b, p. 78.

(7) See Bishop 1989, pp. 137-42, for a discussion of this type of approach.

(8) Ibid., pp. 148-75.

(9) Examples of deviantly caused behaviour can be constructed in which the deviant route does not begin with thinking about, or awareness of, intended or contemplated actions: these are considered later in this paper.

(10) Mitchell 1982.

(11) Bishop 1989, pp. 138-9. See also pp. 138-42 for a discussion and rejection of related strategies for resolving the problem of basic deviancy.

(12) 'Because I intended to' seems a queer kind of response to the question 'why did you do that?', and would certainly fall short of explaining or justifying what one did.

(13) Another reason for rejecting the claim that actions are always caused by intentions is that intentions cannot be defined other than in relation to actions and therefore an analysis or theory of action that defines actions as always being caused by intentions would be circular (Magill 1997, pp. 129-34).

(14) The point has been well made by David Velleman:

Various roles that are actually played by the agent himself in the history of a full-blooded action are not played by anything in the [standard causal] story, or are played by psychological elements whose participation is not equivalent to his. In a full-blooded action, an intention is formed by the agent himself, not by his reasons for acting. Reasons affect his intention by influencing him to form it, but they thus affect his intention by affecting him first. And the agent then moves his limbs in execution of his intention; his intention doesn't move his limbs by itself. The agent thus has at least two roles to play: he forms an intention under the influence of reasons for acting, and he produces behaviour pursuant to that intention (Velleman 1992, p. 462).

(15) See Magill 1997, p. 143, where I give a fuller account of the way in which the reasons-responsiveness of RKC/the Will is conditioned by prior training and habituation.

It might be worth adding that it is certainly a conceptual requirement on causation by RKC being the distinguishing factor between actions and involuntary behaviour that RKC be responsive to what the agent wants and believes, but this requirement can be satisfied if RKC is triggered by a perception - hearing the phone ring, say - in respect of which the agent has no occurrent desires, but only what are sometimes described as standing desires, and if hearing the phone ring counts as a belief, just in the broad sense that hearing and seeing is believing.

(16) Ginet 1990, p. 11-15.

(17) Magill 1997, pp. 122-9, 134-7.

(18) See, for example, O'connor 1995.

(19) Goldman 1976, pp. 81-2.

(20) Davidson 1980c, p. 51.

(21) Heteromesial chains, according to Bishop, are those 'in which the agent exhibits the behaviour required for his or her intended action only by dint of the intentional action of a second agent' (Bishop 1989, p. 156-7).

(22) See Peacocke 1979, pp. 87-8 and Bishop 1989, pp. 158-72.

(23) Alternatively the intervener might know of a way to alter RKC's pattern of responsiveness to attitudes. The effect of the intervention in that case would amount to brainwashing. Certainly we would be unlikely to hold an agent who had been subjected to such a procedure responsible for her actions, as with any case of brainwashing, but I can see no reason for denying that behaviour caused by such a manipulated RKC would be acting.

(24) Peacocke1979, p. 87.

(25) Ibid., p. 88.

(26) Bishop 1989, p. 159.

(27) Ibid., p. 167; Thalberg 1984, p. 257.

(28) Bishop attributes feedback-directed action-regulation to a sub-system rather than intentional regulation by the agent, since the latter would involve the agent doing something and that would introduce circularity into the account (Bishop, pp. 167-9).

(29) Bishop 1989, p. 170.

(30) Ibid., p. 172.

(31) Ibid., p. 171. See also Churchland 1986, pp. 430-1, Magill 1997, pp. 118. Instantaneous actions present a problem for Frankfurt's (1978) account of actions, in which guiding mechanisms, sensitive to negative feedback, are held to be what distinguishes actions from involuntary behaviour.

(32) It does, however, signal the realisation of behaviour initiated by RKC.

(33) Such as the discrete finger movements of an accomplished violinist (Churchland Ibid.).

(34) Bishop 1989, p. 159.

(35) Although, as I will argue, the right causal route cannot, as Peacocke supposes, be characterised as necessarily beginning with intentions.

(36) See, for example, David Pears's gunman whose trigger finger, unknown to him, has a severed nerve. When the gunman's brain issues an impulse signalling his finger to pull the trigger, the impulse attracts lightning that generates another impulse in the severed part of the finger, thus causing the intended movement deviantly (discussed in Peacocke 1979, p. 93).

(37) Morton 1975.

(38) Ibid., p. 73.

(39) It is worth noting that versions of the sensitivity strategy, such as Peacocke's, that make no essential reference to a match-up between actions and the action-directed contents of prior intentions or attitudes do not exclude all cases of deviancy. According to Peacocke's 'differential' version of the strategy 'x's being f is a non-redundant part of the explanation of y's being y, and according to the principles of explanation (laws) invoked in this explanation, there are functions ... specified in these laws such that y's being y is fixed by these functions from x's being f' (Ibid., p. 66). But if an agent's action causing mechanism(s) were fixed in such a way that behaviour caused by any member of some particular class of intentions is always exactly the opposite of what is intended, such intentions would count, according to Peacocke's definition, as differentially explaining the behaviour they caused. (Peacocke goes on to set out an amended, 'strong', form of differential explanation, but it does not escape the objection.)

(40) Moreover, as I remarked earlier (n. 15), it is a conceptual requirement of acting that actions are in some way responsive to what the agent wants.

(41) While RKC must be causally responsive to prior attitudes, however, I do not think that it is triggered exclusively by prior attitudes (see note 15).

References

Bishop, J.ohn, 1989, Natural Agency: An Essay on the Causal Theory of Action. Cambridge University Press.

Davidson, D.onald, 1980, Essays on Actions and Events. Oxford, Clarendon.

- 1980a, 'Actions, Reasons and Causes', in Davidson 1980.

- 1980b, 'Freedom to Act', in Davidson 1980.

- 1980c, 'Agency', in Davidson 1980.

Frankfurt, Harry, 1978, 'The Problem of Action', American Philosophical Quarterly, 15, 2, 157-62.

Donagan, A., 1994, Philosophical Papers, Vol. 1. J. E. Malpas ed. London, University of Chicago Press.

Ginet, C.arl, 1990, On Action. Cambridge University Press.

Goldman, Alvin., I., 1976, 'The Volitional Theory Revisited' in M. Brand and D. Walton, D., eds, Action Theory. Dordrecht, Reidel.

Grice, H.P., 1971, 'Intention and uncertainty', Proceedings of the British Academy.

Magill, Kevin., 1997, Freedom and Experience: Self-Determination without Illusions. London, Macmillan.

Mitchell, Dorothy, 1982, 'Deviant Causal Chains', American Philosophical Quarterly, 19, 351-3.

Morton, Adam., 1975, 'Because He Thought He Had Insulted Him', Journal of Philosophy, 72, 5-15.

O'Connor, Timothy., 1995, ed., Agents, Causes and Events. Oxford University Press.

Peacocke, Christopher, 1979, Holistic Explanation: Action, Space, Interpretation. Oxford, Clarendon.

Thalberg, Irving, 1984, 'Do Our Intentions Cause Our Intentional ActionsSellars, W., 1968, 'Volitions Reaffirmed', in M. Brand and D. Walton eds., Action Theory. Dordrecht, Holland.

?', American Philosophical Quarterly, 21, 249-60.