FISCHER ON ALTERNATIVE POSSIBILITIES AND RESPONSIBILITY

David Mackie

. . . Black inserts a mechanism into Jones’ brain which enables Black to monitor and control Jones’ activities. Jones, meanwhile, knows nothing of this. Black exercises this control through a sophisticated computer which he has programmed so that, among other things, it monitors Jones’ voting behavior. If Jones were to show any inclination to vote for Bush, then the computer, through the mechanism in Jones’ brain, intervenes to assure that he actually decides to vote for Clinton and does so vote. But if Jones decides on his own to vote for Clinton, the computer does nothing but continue to monitor—without affecting—the goings-on in Jones’ head.[1]

This case, presented by John Martin Fischer, is an example of a kind of case devised originally by Harry Frankfurt and designed to refute an intuitively attractive thesis about responsibility—the thesis that alternative possibilities are necessary for moral responsibility.[2] Following others, I shall call this thesis the Principle of Alternative Possibilities, or PAP. PAP, of course, is of interest because, if true, it supports the incompatibility of causal determinism and responsibility; for causal determinism would rule out the existence of the alternative possibilities that (according to PAP) there must be, if attributions of moral responsibility are to be correct.

Though the standard argument against PAP that is based on these Frankfurt-type cases is fairly obviously flawed, Fischer has recently presented an ingenious new argument, based on the Frankfurt-type cases, which he thinks does present a genuine threat to PAP. Fischer’s argument is among the more interesting recent contributions to the continuing debate about determinism and responsibility. It is this new argument that I want to examine in this paper. First, however, it will be helpful to consider briefly the standard version of the argument.

The standard version of the argument against PAP that is based on cases of this kind runs as follows. Suppose that Jones decides to vote for Clinton on his own, just as he might have done even if Black had not inserted the mechanism into his head. Here, we are inclined to accept that Jones could be held responsible for his act of voting—he is, intuitively, an appropriate candidate for attributions of moral responsibility. But, in these circumstances, Jones could not have done otherwise than he actually did. For if he had not decided on his own to vote for Clinton, but had been inclined, instead, to vote for Bush, the computer would have intervened to ensure that he did, after all, vote for Clinton. Hence, according to the standard argument, alternative possibilities are not necessary for moral responsibility.

This Standard Argument runs, in effect, as follows:

(1) In the Jones and Black case as described, Jones is responsible for acting as he does.

But

(2) Jones has no alternative to acting as he does.

Therefore

(3) PAP is incorrect.

As Fischer is aware, however, there is a simple and obvious reply to this standard argument. The reply can be stated as follows:[3]

Initial impressions notwithstanding, the Frankfurt-type cases do involve alternative possibilities. In other words, premiss (2) in the argument is straightforwardly false. All such cases must involve both an actual sequence and a counterfactual one, and these sequences will be genuine alternatives to one another. In our case of Jones and Black, the actual sequence is the one in which Jones decides on his own to vote for Clinton and does so; the counterfactual sequence is the one in which he sets out (whatever, precisely, such ‘setting out’ involves) to vote for Bush, but the computer intervenes to ensure that he votes for Clinton. Defenders of PAP can simply point out that, in imagining the case, we assume that Jones does have the power to choose between these two sequences—we assume that, though he actually decides to vote for Clinton on his own, he could have set out to vote for Bush. Admittedly, such a choice would also (thanks to the computer’s intervention) have led to Jones’s voting for Clinton. But that should not blind us to the fact that the sequences are distinct, and that, in imagining the case, we imagine that Jones had the freedom, or ability, to do otherwise than he actually did. (Similarly, the fact that both of the roads leading from the fork lead to Rome does not mean that they are not alternative routes. If these are the only two roads I can take, then (assuming that I have to follow one of them) I shall end up in Rome either way; but that doesn’t mean that I don’t have the choice which road to take.)

Now, since Jones ends up voting for Clinton in both the actual sequence and the counterfactual sequence, we do need to be a little careful when we specify what Jones can be held responsible for. Since we are to imagine that these are the only two sequences available to Jones, we should not, and intuitively we do not, hold Jones responsible for voting for Clinton, or for his vote’s going to Clinton; for that was inevitable—Jones could not have prevented this result.[4] Jones is responsible, rather, for taking the course of action that he does take; he is responsible for acting as he does—for voting for Clinton on his own. Similarly, in the forking roads case, it seems wrong to hold me responsible for ending up in Rome; but I may nonetheless be responsible for taking the inland route, rather than the coastal route.

The crucial point is that defenders of PAP can correctly point out that the Jones and Black case does involve an alternative possibility. Hence the standard argument does not succeed in its attempt to provide a counterexample to PAP—a case of responsibility in the absence of an alternative possibility. Since this is so, and as Fischer rightly concludes, the standard argument that is based on the Frankfurt-type cases fails to establish the falsity of PAP.

I turn now to consideration of Fischer’s new argument against PAP. Fischer’s suggestion is that, although the Frankfurt-type cases involve alternative possibilities, these possibilities are ‘not sufficiently robust’ to ground the relevant attributions of responsibility. Though there is an alternative possibility in the Jones and Black case, it is, according to Fischer, ‘highly implausible to suppose that it is in virtue of the existence of such an alternative possibility that Jones is morally responsible for what he does.’ According to Fischer, it is not enough for the defender of PAP to show that these cases involve alternative possibilities; ‘what also needs to be shown is that these alternative possibilities play a certain role in the appropriate understanding of the cases. That is, it needs to be shown that these alternative possibilities ground our attributions of moral responsibility.’[5] But that, he says, cannot be shown; for the alternative possibilities in these cases are not sufficiently robust. What makes them insufficiently robust is that they are possibilities in which the agent ‘does not act freely’.

Briefly think about the basic picture of control that underlies [PAP] . . . Here the future is a garden of forking paths. At various points in life, it is envisaged that there are various paths that branch into the future, and one can determine which of these genuinely open pathways becomes the actual path of the future. The existence of various genuinely open pathways is alleged to be crucial to the idea that one has control of the relevant kind. But if this is so, I suggest that it would be very puzzling and unnatural to suppose that it is the existence of various alternative pathways along which one does not act freely that shows that one has control of the kind in question. . . . And notice that this is precisely the situation in the Frankfurt-type cases. In particular, note that . . . it is . . . evident that Jones would not be freely voting for Clinton in the alternative sequence.[6]

Fischer’s central idea is that, although the Frankfurt-type cases are not straightforward counterexamples to PAP, they can be used to challenge PAP in an indirect way. For if the alternative possibilities in these cases are insufficiently robust to ground the attributions of responsibility in the actual sequences, they are not doing the work that defenders of PAP must claim that they do. If so, that casts doubt on the view that these alternative possibilities are necessary for responsibility at all.

This new argument may be recast as follows:

(1) In the Jones and Black case as described, Jones is responsible for acting as he does.

(2) In the counterfactual sequence, in which Jones sets out to vote for Bush but Black’s device intervenes, Jones’s action is unfree.

(3) If action in a counterfactual sequence is unfree, then the existence of this alternative sequence cannot be part of the reason why the agent is responsible for what (s)he actually does.

Therefore

(4) The existence of this counterfactual sequence is irrelevant to Jones’s being morally responsible for acting as he does.

Hence, plausibly,

(5) PAP is incorrect.

III

Clearly, the interesting new premisses here are (2) and (3). We can begin by considering premiss (2). Fischer’s new argument depends on the description of the Jones’s action in the alternative sequence as ‘unfree’. Accordingly, to assess his suggestion, we need to understand what this means. Curiously, Fischer himself nowhere explains in any detail what the term is supposed to mean in this context. Particularly in a work on free will, such cavalier use of this term is, to say the least, unfortunate. Still, it is fairly clear what the point must be: Jones’s action in the alternative sequence is presumably unfree in that it is Black’s computer that controls Jones’s behaviour here. Once the computer intervenes, Jones’s choices and actions are unfree in that it is not up to him what he does: the computer controls what he does; it causes him to decide to vote for Clinton, and ensures that he does so vote.

We can agree, then, that Jones’s action in the counterfactual sequence is, in this sense of the term, unfree. Hence premiss (2) in Fischer’s argument, understood in this way, is unobjectionable. Premiss (3), however, is much more dubious: Jones’s action in the counterfactual sequence is unfree. But it is very doubtful whether this has the intuitive significance that Fischer claims it has. For it remains entirely unclear why this fact should render it implausible that the existence of this alternative sequence can ground the ascription of responsibility in the actual sequence. What Jones does in the counterfactual sequence is caused and controlled by Black’s computer. But that does not alter the fact that, in the case as we have imagined it, Jones has the ability, or freedom, to choose which of the two sequences occurs: he is imagined to have the ability to choose whether to vote for Clinton on his own, or, alternatively, to set out to vote for Bush. If he had set out to vote for Bush, his intention would have been promptly thwarted by the computer’s intervention; and everything he did after this intervention would have been unfree, in the sense in which we are now using that term. But that presents no intuitive threat to the view that it is the existence of this alternative sequence that grounds the attribution of responsibility in the actual sequence. The status—free or unfree—of what Jones does after the computer has intervened is intuitively quite unimportant here: what is relevant is simply whether Jones has the initial ability, or freedom, to choose which of the two sequences occurs. And this freedom, in the case as we have imagined it, is clearly still present.

Fischer is right that, if PAP is correct, there must be some feature of the Jones and Black case that renders the counterfactual sequence capable of grounding the attribution of responsibility in the actual sequence. The counterfactual sequence must be, to use Fischer’s terminology, sufficiently ‘robust’. To see why this is so, we need only consider the obvious truth that the mere presence of some alternative, counterfactual, possibility is not sufficient for responsibility in the actual case. Causal determinists, for example, believe that earlier states of the world, together with the laws of nature, entail later states of the world. Although I cannot change either the way the world was in the past, or the laws of nature, these determinists may believe that it is nonetheless true that I could have done otherwise than I actually did. For they may hold that I could have done otherwise if the past, or the laws of nature, had been different. Still, even if there is in this sense an alternative possibility, it is completely implausible to suppose that the existence of this kind of alternative possibility renders me responsible for what I actually did. It is insufficiently robust to ground such an attribution of responsibility. Again: England could have beaten Australia in the 1997 Ashes series (they didn’t). I may believe that that statement is true, on the ground that England could have beaten Australia if England’s players had been more naturally talented cricketers than they are. If so, I believe that there is an alternative, counterfactual, possibility in which England did win the 1997 Ashes series. But I need not believe that the existence of this alternative possibility means that England are responsible for losing. (As we might say: they might have been more talented; but it may not be their fault that they are as bad as they are.) That is enough to show that the existence of any old alternative possibility is not enough. It has to be an alternative possibility of a certain kind: it must be, to use Fischer’s term, robust.

Fischer is right, then, that alternative possibilities must have some property of robustness, if they are to ground ascriptions of responsibility for what an agent actually does. Moreover, he is right in thinking that robustness depends on the agent’s having a certain kind of freedom. But he locates this freedom incorrectly, and in so doing misrepresents our intuitions, when he assumes without argument that the relevant freedom would have to be freedom (in the sense specified) of action in the counterfactual sequence. Fischer assumes that robustness, for defenders of PAP, would have to be an intrinsic property of the action that takes place in the counterfactual sequence: he assumes that it would have to be a matter of such action’s being ‘free’, where its being ‘free’ entails, at least, that it is not caused and controlled in the way that Jones’s action is controlled by Black’s computer. But this view is implausible. Robustness is far more plausibly to be identified with a relational property of the counterfactual sequence itself: a counterfactual sequence is robust only if the agent had the ability, or freedom, to choose the occurrence of that sequence, rather than the actual one.

After all, in the examples given above, the lack of robustness is clearly not due to the fact that the agents in the counterfactual sequences do not act freely (in the sense specified). Rather, what is wrong is simply that these agents do not have the ability, or freedom, to choose which of the two possible sequences is actual: I do not have the freedom to choose the laws of nature or to alter the past; England’s cricketers do not have the freedom to choose to be more naturally talented than they actually are. It is, then, far more plausible to suppose that it is this kind of freedom that is relevant to the robustness of an alternative possibility. Indeed, it is hard to see how freedom of action in the counterfactual sequence could be relevant: after all, how could a freedom that is never exercised conceivably be relevant to whether an agent is responsible for what he actually does? Far more plausibly, what matters here is the freedom that is exercised—the freedom, which precedes the counterfactual sequence itself, to choose which of the two possible sequences occurs. This view of robustness provides the most plausible explanation of the truth of PAP: alternative possibilities are necessary for responsibility not because of any intrinsic property of what happens in these counterfactual sequences, but simply because the existence of such sequences is essential to the agent’s having a choice to make: an agent cannot have the relevant freedom to choose unless there are alternatives for him to choose between. And, on this more plausible view, the alternative possibilities in the Frankfurt-type cases are robust. Defenders of PAP, then, can reply to Fischer’s argument by rejecting premiss (3) in the formulation given above.

Once we recognise that robustness does not depend on there being any free (in the sense specified) action in the counterfactual sequence itself (after, that is, this sequence has been initiated), but depends, rather, on the agent’s having the freedom, or ability, to choose whether that sequence occurs, we can also see the weakness in an appeal that Fischer makes to an epistemological analogy in support of his view. Fischer writes:

. . . certain accounts of knowledge imply that an agent knows that p only if he can distinguish a class of situations in which p obtains from a contrasting class in which p does not obtain. On this approach, knowledge requires a certain kind of discriminatory capacity; this model is clearly analogous with the view that moral responsibility requires regulative control [i.e., the kind of control that implies the presence of alternative possibilities]. More specifically, on this approach to knowledge, an agent knows that p only if there exists a set of alternatives to the actual world in which the agent’s beliefs line up with states of the world in the right way. What would be highly implausible would be to suppose that what transforms some case of lack of knowledge into a case of knowledge would be the existence of a range of alternative scenarios in which the agent gets it wrong! . . . [A]rguably, it is not much more plausible to suggest that it is in virtue of a set of alternative possibilities in which Jones does not act freely that he actually can be held morally responsible for his behavior.[7]

Fischer’s use of this analogy would be unobjectionable if it were the case that responsibility in the actual sequence depended on the presence of free (in the sense specified) action in the counterfactual sequence. That would be true, if robustness required freedom of action in the counterfactual sequence itself—after that sequence is under way. If this were the case, then freedom of action in the counterfactual sequence would indeed play a role in responsibility that was analogous to the role that truth of belief in the alternative scenarios plays in knowledge (according to this kind of account of knowledge). But that account of robustness, as I have already argued, is implausible and counterintuitive. Rather, what is relevant is whether the agent has the freedom to decide which sequence becomes actual. The exercise of this freedom clearly precedes the chosen sequence. Since this is so, the relevance of alternative possibilities to moral responsibility is not analogous to the relevance of alternative scenarios to knowledge, on the kind of account of knowledge to which Fischer refers. Since the cases are in this way non-analogous, Fischer’s appeal to this epistemological example does nothing to support his argument against PAP.

Though more thoroughly worked out, the response to Fischer’s argument against PAP that I have given here is similar in essence to an objection that Fischer himself briefly considers, but rejects. Accordingly, I need finally to examine this objection and Fischer’s reply to it.

An opponent, Fischer writes,

. . . may respond that nevertheless there is at least the following thing in the alternative sequence which is freely done: Jones begins to initiate the process of making a choice to vote for Bush. (Of course, this process is then cut off before it can be completed.) So we seem to have isolated at least something in the alternative sequence which can plausibly be thought to be freely done and which thus may be able to ground the ascriptions of responsibility.[8]

Against this reply, Fischer appeals to a variant of the Jones and Black case in which what triggers the intervention of the computer is an involuntary sign evinced by Jones which precedes any initiating action of making a choice, or forming an intention, or setting out in any other way, to vote for Bush.

Suppose we . . . consider [a] version of the Jones and Black case in which Black can be alerted to Jones’ future inclination to vote for Bush by the presence of some involuntary sign, such as a blush or a twitch or even a complex neurophysiological pattern. So if Jones were (say) to blush red, then Black would intervene prior to Jones’ doing anything freely and ensure that Jones indeed votes for Clinton. Here the “triggering event” (i.e., what would trigger the intervention of Black) is not any sort of initiating action, and thus cannot be said to be freely done. Again, precisely as above, this sort of triggering event appears to be not sufficiently robust to ground responsibility ascriptions.[9]

The argument based on this variant of the Jones and Black case can be summarised as follows:

(1) In this variant of the case, Jones is responsible for acting as he does.

(2) What would activate Black’s device is not any decision or action of Jones’s, but an involuntary blush or twitch.

Therefore

(3) Jones here lacks the ability to choose which sequence occurs.

Therefore

(4) The existence of the counterfactual sequence is irrelevant to Jones’s being morally responsible for what he actually does.

Hence, plausibly,

(5) PAP is incorrect.

Though ingenious, this reply is unsuccessful. To start with, the inference from (2) to (3) here is dubious. It can’t be assumed that because a blush is involuntary, Jones has no ability to control whether it occurs. Involuntary blushes and twitches might be under an agent’s control. For example: by deliberately reflecting on some recent embarrassing experience, I can cause myself to blush involuntarily; by crossing my legs in a certain fashion, I can ensure that some involuntary twitching occurs. Moreover, I can (it seems) choose whether to reflect, or cross my legs, in these ways. If this is how things stand with Jones, however, then the case is clearly of no use to Fischer. For if Jones has the ability to choose whether the course of events in which the relevant sign is evinced takes place, then he plainly does have the ability to choose which of the two sequences—actual or counterfactual—occurs. But that ability, or freedom, as I have already argued, is precisely the kind of freedom that can ground the attribution of responsibility that we make.

For a case which has some hope of serving Fischer’s purposes, we need to suppose not merely that the triggering event is involuntary, but that Jones has no ability to decide whether it occurs. We need to stipulate that premiss (3) in the argument is true. We must suppose that the triggering sign would appear when, but only when, Jones was about to set out to vote for Bush, but before he did anything. And we must suppose that the evincing of this sign is outside Jones’s control in this sense: he must literally not have the ability, or freedom, to decide whether or not the course of events in which the sign is evinced takes place. If this is how things stand, however, then it clearly follows that Jones literally has no ability, or freedom, to choose which of the two sequences occurs. But, importantly, where this is so, we will not, after all, be inclined to hold him responsible for what he actually does: we will not regard him, in the actual sequence, as an appropriate candidate for attributions of responsibility. That is, when the case is set up in this way, I see no reason why we should continue to accept premiss (1). In other words, if we revise the case in the way that Fischer requires, we do indeed end up with a case in which, though there are two sequences, the alternative possibility is insufficiently robust to ground ascriptions of responsibility. But that presents no threat to PAP; for in revising the case in this way we have converted it into a case in which it is no longer natural to think that Jones is an appropriate candidate for attributions of moral responsibility. What Fischer needs, of course, is a case in which it is intuitively plausible to hold Jones responsible for what he does in the actual sequence, but in which the counterfactual sequence is insufficiently robust to ground that attribution of responsibility. But, despite his efforts, he has simply failed to present any such case.