My aim in this paper is to defend a libertarian incompatibilist account of free will against four plausible objections. First, I’ll develop the objections; second, I’ll offer a positive argument on behalf of libertarian incompatibilism; finally, I’ll try to show how my positive account handles the objections.
Objection 1: Free Choices Would Be Unmotivated
One celebrated objection to libertarian incompatibilism, tracing its ancestry back at least to Hume, is the claim that if actions were not necessitated by previous states of affairs they would be random and uncaused, and therefore not caused by the agent’s desires, and therefore not motivated or purposeful, and therefore not actions at all – and certainly not actions over which the agent could plausibly be said to have any control, which is what the libertarian incompatibilist was supposed to have wanted. This objection is so familiar that I assume I need not spend more time explaining it.
Objection 2: We Couldn’t Know Whether Anyone’s Choices Were Actually Free A different (and to my mind a subtler) objection to incompatibilism about free will is that it seems to commit us to a kind of agnosticism about whether anybody actually has free will or not. If compatibilism is true, then in order to determine whether someone has acted freely, we need only apply our everyday criteria for responsibility: did the person’s action express her desires? was her rationality unimpaired? was she free from deception or duress? And these are factors which we have some idea how to identify. But if incompatibilism is true, then in order to determine whether someone has acted freely, we must determine not only whether her action meets these everyday criteria, but in addition, we must determine whether her action was or was not entailed by the past history of the universe together with the laws of nature – and that is something for which we have no idea how to test.
Some incompatibilists will happily admit the truth of this point, but deny that it constitutes grounds for criticism. The incompatibility between free will and determinism, these incompatibilists will say, may be a conceptual truth, ascertainable through a priori reasoning; but the question of whether free will exists is an empirical, scientific question – one to be settled in the neurophysiologist’s laboratory rather than in the philosopher’s armchair.
But surely the compatibilist objector is right to think this a very strange reply. Could we really live this sort of incompatibilism? Is it possible to sustain a stable attitude of agnosticism as to whether we and those around us are genuinely free agents? If incompatibilism genuinely involved this sort of deformation of our ordinary conception of ourselves and others, this would be, at the very least, a serious mark against it.
I pause to note that the second objection seems to raise a problem for the first. For the first objection implies that an action’s being done freely requires its being causally determined, which is presumably just as hard a matter to test as its being causally undetermined. So both of these objections can’t be right. Still, either one’s being right would be enough to sink the libertarian incompatibilist’s ship.
Objection 3: Free Choices Wouldn’t Fit into a Scientific Picture of the Universe
The notion that there’s something spooky or supernaturalist about free choices as the libertarian incompatibilist conceives them is a sentiment expressed frequently, but also rather vaguely. Let me try to sharpen it.
Aristotle draws a useful distinction between temporal and natural priority. Even when X and Y are simultaneous in time, X may nevertheless be prior to Y in a nontemporal sense if Y depends on X and not vice versa. Aristotle initially cashes out this notion of dependence in terms of asymmetric entailment: if Y entails X but not vice versa, then X is prior to Y because it can exist without Y while Y cannot exist without it. But then Aristotle adds that even in cases where X and Y entail each other, X still counts as naturally prior to Y if X is the reason or basis or cause of Y.1 Whatever causes or explains something is naturally prior to what it causes or explains. What is naturally prior may be temporally prior as well, but it need not; think of Augustine’s example2 of a foot pressed into a footprint throughout all time.
Now every physicalist accepts the core claim that all facts, including psychological facts, supervene on physical facts. But physicalism requires more than this; for mere supervenience could be accepted by dualists and idealists as well. Suppose, for example, that the physical world is constituted by the dreams of the Great Cosmic Mind. Suppose further that every alteration in the mental states of the Great Cosmic Mind automatically manifests itself in, and as, an alteration in the dream. In that case, the requirements for supervenience would be met: no mental change without a physical change! But that would hardly be a physicalist world. Physicalism requires more than a mere correlation between the mental and the physical; it requires the physical to be in some way the cause or reason or basis of the mental.
The asymmetry to which physicalists are committed need not involve the claim that all explanation is bottom-up or that higher-level properties are epiphenomenal. Think of my shadow, which supervenes on my shape and activity, along with how I’m spatially related to light sources and other objects. My shadow is not causally inert; I can see it and be startled by it, for example. So there can be reciprocal interaction between my shadow and its supervenience base. Nevertheless, there remains an important sense in which what my shadow does is grounded in my spatial relations and not vice versa. The shadow’s causal powers are parasitic on the causal powers of its supervenience base. Likewise for the psychological and the physical. So let’s say that X supervenes robustly on Y just in case X supervenes on Y, and Y is also either identical with or naturally prior to X.
The physicalist holds, then, that all psychological facts supervene robustly on physical facts. Does this view commit her to the more precise claim that, for any given time t, all psychological facts holding at t supervene robustly on physical facts holding at t? Strictly speaking, no, at least if genuinely remembering at t and falsely remembering at t are distinct psychological states, since these might be indistinguishable in terms of physical states holding at t. That, however, is true only because a psychological state holding at t counts as a genuine rather than a false memory only because of facts holding at times prior to t. Let us restrict our attention, then, to hard psychological facts holding at t (where a hard fact is one whose specification makes no essential reference to earlier or later times). One candidate for a hard psychological fact would be apparent memory – a genus of which genuine memory and false memory are species.
Is the physicalist committed to claiming that all hard psychological facts supervene robustly on temporally simultaneous hard physical facts? Perhaps she is, but I won’t argue for that thesis. Instead, I shall argue for a still weaker thesis: any physicalist who rejects backward causation (which is most of them) is committed to claiming that all indeterministic hard psychological facts supervene robustly on temporally simultaneous hard physical facts (where a hard fact is indeterministic just in case no temporally prior hard fact is sufficient for it). Here’s why. An indeterministic hard fact cannot supervene on any hard fact temporally prior to it, for if it did, then it would have a temporally prior hard fact as a sufficient condition, which is ruled out ex hypothesi. It can, of course, supervene on hard facts temporally posterior to it, but such supervenience cannot be robust; for if it is logically grounded in a future fact then it is no longer hard, and if it is causally grounded in a future fact then we have backwards causation.
On the physicalist view, then, any indeterministic psychological event3 must be either a) identical with, or b) naturally posterior to, its temporally simultaneous physical supervenience base. But disjunct (a) is superfluous. Even if psychological events are identical with physical events (which a physicalist of course need not maintain), no psychological event is going to be identical with a simple physical event (say, the motion of a single subatomic particle); that would be enormously implausible. So a psychological event could be identical only with a complex physical event. And if that event is an indeterministic one, at least one of the constituent events on which it supervenes must be indeterministic as well. Now if the whole indeterministic event occurs because its constituents do, then the occurrence of the constituents is naturally prior to the occurrence of the whole.
This is not the only possibility, of course. In quantum physics, Bell’s Nonlocality Theorem appears to show that there are synchronistically linked indeterministic events whose constituents occur because the whole does; the determination is top-down. This is a failure of robust supervenience, but does not contradict physicalism, since the non-robustly supervenient item is not itself nonphysical. Suppose that an indeterministic psychological event were identical with a Bell nonlocal quantum event. In that case, such an event would not be naturally posterior to its supervenience base. But any such neat and tidy correspondences between psychological events and Bell nonlocal quantum events would be amazing. Barring such correspondences, the physicalist is pushed to the view that every indeterministic psychological event has naturally prior though temporally simultaneous hard sufficient conditions.
Now the libertarian believes in free choices that have no temporally prior hard sufficient conditions, while the physicalist believes that any such choices must have naturally prior but temporally simultaneous hard sufficient conditions. (Assume such choices are hard facts.) So stated, there is no conflict between the two positions. An event could quite consistently lack temporally prior hard sufficient conditions while possessing naturally prior but temporally simultaneous hard sufficient conditions. To put it another way, an event could (robustly) supervene on present physical conditions without doing so on past ones.
Nevertheless, it would be unreasonable to hold both positions. Here’s why. The libertarian believes that the absence of temporally prior hard sufficient conditions is a precondition for a choice’s being free. No one who thinks that could reasonably find naturally prior (but temporally simultaneous) hard sufficient conditions unobjectionable. After all, in both cases, one’s choice is settled by factors distinct from and prior to it. If deterministic causes are a threat to freedom, robust supervenience must be so as well. For imagine saying to the libertarian: “The fact that your choices are determined by their supervenience base is no threat to freedom. They’re still your choices; the necessitating explanation goes through your choice, it doesn’t bypass it.” If the libertarian could find such reassurances convincing in the case of robust supervenience, then she should also find them convincing in the case of causal determinism. But, modus tollens. What argument could make robust supervenience palatable without making causal determinism palatable as well? There is no reason to treat temporally prior sufficient conditions and naturally prior sufficient conditions differently; either both are a threat to freedom, or neither is. But the libertarian cannot say that neither is (without ceasing to be a libertarian); so she must say that both are. Determinism is just the view that the future robustly supervenes on the past; the libertarian’s grounds for rejecting this will be equally good (or bad) grounds for rejecting the view that present free choices robustly supervene on anything.
Now those who are suspicious of narrow content may find the notion of hard psychological facts objectionable for the same reason. For example, if “meaning ain’t in the head,” then even an apparent memory of water will count as being that, rather than an apparent memory of twin-water, only if it has the right causal history, and so is not a hard psychological fact after all. So why couldn’t a libertarian physicalist regard psychological events, including free choices, as soft facts, and thus avoid the implication that present choices are determined by naturally prior but temporally simultaneous facts? The answer is that any view that regards present choices as soft facts will presumably have to regard such choices as consisting partly in what is happening now and partly in what has happened in the past. It would be very odd to treat present choices as consisting partly in what has not yet happened (except under non-essential descriptions, such as “Caesar chose to set in motion a series of events which would culminate in the destruction of the Roman republic”). But presumably if my present choice consists in past and present facts, then for the physicalist it robustly supervenes on some conjunction of past hard physical facts and present hard physical facts. If libertarians are committed (by definition) to rejecting robust supervenience on the former, and are likewise committed (by the argument I’ve just given) to rejecting robust supervenience on the latter, it’s hard to see why robust supervenience on the conjunction of the two should fare any better.
It follows, apparently, that the libertarian incompatibilist is committed to rejecting not only determinism but also physicalism. That’s not necessarily a decisive objection to the view, since physicalism remains a controversial position. But one might think it at least adds a heavy weight to the libertarian compatibilist’s already considerable burden of proof.
Objection 4: The Laws of Logic Would Fail
Our first two objections are well established in the literature; the third I haven’t seen stated in precisely the form I state it, but I’ve suggested that it’s implicit in much of the resistance to the libertarian incompatibilist position. This fourth objection, by contrast, is one that few cotemporary philosophers will take seriously; but I think they should.
There is a way of thinking about possibility in ordinary language that is not captured by philosophers’ talk of possible worlds. In this way of thinking, a possible state of affairs is one that is reachable from the actual world, meaning that there must be some available route from the actual world to this possible one. Hence while contemporary philosophers would regard a world in which England won the Battle of Hastings as a possible world, as it involves no contradiction, on the ordinary way of thinking such a scenario is only formerly possible, since any such possibility has since been foreclosed by what actually happened. In other words, for contemporary philosophers a possible world need only be internally consistent, but on the ordinary understanding of possibility, a possible world must also be consistent with the actual world. This, I suggest, is the way of thinking about possibility that Aristotle, Epicurus, and the Stoics were appealing to in their treatment of future contingents. Call this conception of possibility availability.
Now if any available world must be consistent with the actual world, this implies that if it is a fact about the actual world that I will make a certain choice tomorrow, then any future in which I fail to make that choice is unavailable . We are then left with two options. One is to embrace the modus ponens and affirm that there is only one available future; this was of course the Stoic solution. The other is to embrace the modus tollens and insist that there is as yet no fact of the matter as to what choice I will make tomorrow, so that more than one available future is consistent with the actual world; this was of course the Aristotelean and Epicurean solution.4
Given this outlook, the objection to libertarianism – or indeed to any form of indeterminism – is that the indeterminist must choose the second option, and so deny truth-value to predictions of future contingent events, and so deny Bivalence. But the denial of Bivalence, so runs the objection, is absurd; therefore, indeterminism is false.
So what exactly is wrong with denying Bivalence? While many objections have been raised to the position Quine labels “Aristotle’s Fantasy,” I haven’t seen anyone raise the criticism that the position offends against Leibniz’s Law; but it seems to me the most natural objection to raise. According to Leibniz’s Law, A and B are identical only if they have all the same properties. Whether or not sharing all properties is, as Leibniz also thought, sufficient for numerical identity, it is at least necessary. But according to Aristotle’s Fantasy, predictions concerning future contingents are, as yet, neither true nor false. If it is not yet causally settled whether there will be a sea battle tomorrow (to use Aristotle’s example), then there is as yet nothing for a prediction to correspond to or clash with – no truthmaker for either the assertion or the denial – and so any prediction’s status must remain indeterminate until things turn out one way or the other.
But these two claims seem to clash. If, as per Aristotle’s Fantasy, it wasn’t yet true yesterday that I would be writing about free will today, then it seems difficult to resist the inference that I have a property – writing about free will today – that my past self lacks, thus violating Leibniz’s Law. While this objection to indeterminism is one that few contemporary philosophers find compelling, since I accept the availability conception of possibility it’s an objection that I, at least, need to worry about.
Here, then, are our four objections to libertarian incompatibilism: the position is alleged to clash, first with free choices’ being motivated, second with free choices’ being identifiable by ordinary criteria, third with free choices’ having a place in the natural order, and fourth with Leibniz’s Law.
Determinism and Deliberation
Many arguments have been offered for libertarian incompatibilism, but I shall focus on the argument from deliberation – an argument associated most notably with Kant but arguably running back to the Greeks. According to this argument, we cannot coherently engage in deliberation unless we assume that it is not already settled, prior to our choice, what decision we shall make. But if we were to accept determinism, we would on the contrary have to regard this as settled. Since deliberation is an indispensable feature of our lives, we cannot accept determinism without pragmatic incoherence. Indeed, the very act of deliberating about whether to accept determinism would already commit us to rejecting it.
The obvious compatibilist response is to claim that what would render deliberation incoherent is not the assumption that the outcome is already settled but rather any assumption about which way the outcome is settled. In other words, I cannot coherently deliberate between options A and B if I believe it is already settled that I will choose A; but that does not show that I cannot coherently deliberate between options A and B if I believe it is already settled which one I will choose.
The example of Newcomb’s Problem, however, casts doubt on this compatibilist strategy. In Newcomb’s Problem, the case for choosing both boxes seems rationally decisive, and the case for choosing one box only also seems rationally decisive. Attempts to deny the decisiveness of one or the other are motivated by the assumption that they cannot both be decisive. It seems clear to me that they are both decisive, given the conditions laid down in the Newcomb setup. Hence I follow George Schlesinger5 in taking Newcomb’s Problem as a pragmatic reductio of those conditions. What Newcomb’s Problem shows is that no one can coherently take herself to be in a Newcomb’s Problem situation (since if she were, she would have to take herself to have decisive reason for mutually incompatible courses of action, which is pragmatically incoherent).
The two-boxer will point out, correctly, that the opaque box has in it now, and has had in it for some time, whatever money it has in it. The money isn’t hovering in a halfway state of almost-there-ness like Schrödinger’s Cat; it’s either there or not there, and no choice of mine can make such money magically materialise or vanish. Whether the money is there or not is already settled, and I cannot unsettle it. In other words, I must regard the presence or absence of the money as naturally prior to my choice. But on the other hand, the one-boxer will point out, correctly, that as the conditions have been set up, my choosing the opaque box only is both necessary and sufficient for the money’s being in it, so if I want the money I should choose the opaque box only, like a Calvinist performing good works in order to ensure that he is already one of the Elect. If I can control my choice, and the money’s being there or not tracks my choice, then can I not control whether the money is there? I am choosing the one-box strategy in order to make it the case that the money be there. In other words, I must regard the presence or absence of the money as naturally posterior to my choice. To take myself to be in a Newcomb’s Problem, I must take one and the same fact to be both naturally prior and naturally posterior to my choice. That is why one cannot coherently take oneself to be in such a situation.
My argument assumes that when we do X for the sake of Y, we must regard X as a cause of Y and so as naturally prior to Y. Richard Taylor denies this:
[T]here are certain purposeful actions, the ends or goals of which … precede those actions. For example, if it is one’s purpose to flex a certain arm muscle, the only way he can accomplish this, normally, is by moving his arm. The arm is caused to move by the motion of the muscle, but the arm is moved in order to move the muscle, not vice versa. The motion of the muscle does not follow upon the motion of the arm, however … [but] precedes it slightly. Or consider nerve impulses. Part of the cause of the motion of a man’s arm, when he moves it, is certainly a certain nerve impulse from his brain, but it is false that he moves his arm by means of the nerve impulse. It is the other way around. Thus, if a man learns of these nerve impulses – and many go to their graves without ever suspecting they exist – and then has some occasion to produce one of them – perhaps for the purposes of some experiment in physiological psychology – he can do so only by moving his arm. The purposeful action in such a case – namely, the man’s moving his arm – is the means to a certain end – the production of a certain nerve impulse – which actually precedes that purposeful action in time.6 Taylor finds this result “frightfully puzzling,” and well he should. His example strikes me as simply a reductio of his view that basic actions are bits of overt behaviour, rather than volitions. Without that assumption, the paradox dissolves.
But what is it about the conditions of the Newcomb setup that renders pragmatically incoherent the assumption that we are in one? If it is only knowing which option is settled that renders deliberation incoherent, then the incoherence of Newcomb’s Problem is unexplained, since the agent does not know which option is settled. What explains the incoherence, I suggest, is the mere assumption that some option is settled. Whether the money is in the opaque box is already settled, since the predictor has long since decided whether to put in the money or not; but the presence of the money is sufficient (not causally sufficient, but sufficient nonetheless, as the Newcomb story goes) for my choosing the one-box option, and its absence is sufficient for my choosing the two-box option, so it is already settled what option I will choose; and this is the crucial feature that leads to the paradox.
The Newcomb setup as it stands includes three odd features: a) the fact that my choice is already settled, b) the fact that the predictor knows which way it’s settled, and c) the fact that I know that (though not what) the predictor knows. It’s easy to be distracted by (b) and (c), and to take them, rather than (a), as generating the paradox. But (b) and (c) are accidental and eliminable features of the setup; for we can construct a version of Newcomb’s Problem without any predictor at all.
Let us define C as the disjunction of those conditions that might possibly hold at t1 the holding of any of which would be causally sufficient for my -ing at t2. (We can think of choosing just the opaque box rather than both as a possible instance of -ing, and we can think of whatever leads the predictor to predict my one-box strategy and so to place the money in the box as a possible instance of C. But we needn’t; -ing and C can be anything, so long as they stand in the causal relation described.) If causal determinism is true, then the occurrence of C at t1 is causally necessary for my -ing at t2; if not, not.
Now there seem to be very few limits to what human beings might conceivably have preferences about; so I am surely calling for a harmlessly modest assumption if I ask us to assume that I might have preferences rank-ordered as follows.
1. C holds at t1 and I do not at t2
2. C holds at t1 and I at t2
3. C does not hold at t1 and I do not at t2
4. C does not hold at t1 and I at t2
Now if, at t2, it is outside my control whether C holds at t1 – which it must be, since it is settled and past – then the rational thing for me to do, clearly, is not to at t2. For if C does hold at t1, then not -ing at t2 will get me option 1 over option 2; and if C does not hold at t1, then not -ing at t2 will get me option 3 over option 4. However, if I accept determinism as true, then the opposite reasoning also holds; for in that case the only possible options are 2 and 3. If so, then C’s occurrence at t1 is necessary for my -ing at t2, and so my -ing at t2 is sufficient for C’s occurrence at t1; and since I cannot help regarding the choice between -ing and not -ing as up to me, I must regard it as in my power to guarantee, by my present -ing, the holding of C in the past, and so it is rational for me to . (That, incidentally, is why David Lewis is wrong to assimilate Newcomb’s Problem to the Prisoner’s Dilemma.)7
We may phrase the same argument slightly differently: Let D1 be the disjunction of all those possible pasts that, when conjoined with the actual laws of nature, entail my -ing at t; and let D2 be the disjunction of all those possible pasts that, when conjoined with the actual laws of nature, entail my not -ing at t.
Now suppose I believe it is causally determined whether or not I shall at t. In that case, I am committed to assuming that the actual past belongs either to D1 or to D2.
Suppose further that I have, as I might, the following preference ordering:
A. The actual past belongs to D1, and I do not at t
B. The actual past belongs to D1, and I at t
C. The actual past belongs to D2, and I do not at t
D. The actual past belongs to D2, and I at t
Ex hypothesi I can deliberate about whether or not to at t. But since I cannot change the past, it makes no sense for me to deliberate about what the past is to be; I must take it as already settled whether the actual past belongs to D1 or D2. If the actual past belongs to D1, the only options are A and B, so I ought to refrain from -ing at t; if the actual past belongs to D2, the only options are C and D, so once again I ought to refrain from -ing at t. In either case, then, it is rational for me to refrain from -ing at t whatever the past is. But if I also take it to be causally determined whether or not I shall at t, then the only possible options are B and C. Hence the rational thing for me to do is to at t.
The assumption that it is causally determined whether or not I shall at t, conjoined with the assumption that I have a certain preference ordering which I certainly might have, entails both that the rational thing for me to do is to at t and that the rational thing for me to do is to refrain from -ing at t. As a deliberative agent I cannot accept that there is nothing that I rationally ought to do. If the conjunction of two assumptions commits me to a contradiction, I must reject one of the assumptions. Since the relevant preference ordering is one that I can freely adopt at any time – for the purpose of refuting determinism, for example – I cannot coherently reject the assumption that I might have that preference ordering. Instead, then, I must reject the assumption that it is causally determined whether or not I shall at t. In short, I must affirm my own libertarian freedom a priori.
To put the point slightly differently: compatibilists and incompatibilists often disagree as to whether coherent deliberation requires the assumption that alternatives are causally open, or only the assumption that they are epistemically open. But if the epistemic assumption were all that’s required, then we should be able to deliberate coherently while taking ourselves to be in a Newcomb situation. Ergo, modus tollens.
Objection: The fact that we cannot coherently regard our choices as determined does not entail that they are not so; there is no incompatibility between a claim’s being true and its endorsement being pragmatically incoherent. Reply: That is quite true; although determinism is pragmatically incoherent, its truth remains logically possible. But that fact cannot count as a reason to take determinism seriously after all; for to say that the logical possibility of determinism’s truth gives us reason to regard it as a live theoretical option is to say that the logical possibility of determinism’s truth is enough to render the endorsement of determinism coherent. (If a claim cannot be coherently endorsed, then it is not a live theoretical option.) And if the logical possibility of determinism’s truth does render the endorsement of determinism coherent, then the possibility claim and the incoherence claim are not compatible after all; the conclusion of the argument contradicts the premise it starts from. The determinist cannot have it both ways.
That’s my brief for libertarian free will. Now let’s return to the objections.