1. Introduction

Much of the debate about free will is driven by competing notions of what free will is, and I will spend most of this post discussing various such notions. So it is important to establish a theory-neutral definition of what the words ‘free will’ denote. Following the Stanford Encyclopaedia of Philosophy we shall define:

Free Will: The unique ability of agents to exercise control over their conduct in the manner necessary for moral responsibility.

The question that we shall discuss is whether it is possible to have free will under determinism. However, it is important to note that determinism is not the only threat to free will. If all of our actions were truly random we have a similar charge: I am not in control of, nor responsible for, a random event. Thus the advocate for free will can’t just refute determinism, e.g. by appealing to quantum randomness, to defend free will. Nevertheless, determinism has been the most important challenge to the claim that we have free will. The type of determinism we shall consider here will be physical determinism, which we shall define:

Physical Determinism: There is only one future physically possible from the present. Given perfect knowledge of the present and of physical laws, one could infer perfect knowledge of the future.

In this post I will generally assume that persons do indeed have free will, and focus mostly on the question of determinism and free will. This may seem a somewhat large assumption, but it is my experience that generally the only reason people doubt the existence of free will is due to their belief in determinism. As I will be defending compatibilism, I will simultaneously be addressing this concern.

2. The ‘Garden Path’ model of free will

Perhaps the most obvious way of thinking about free will, about the control we have over our actions, is that it is akin to walking down a garden path that forks in many directions. We are free when we can choose which path to walk down. Determinism is a problem, as it implies that there is in fact only one path that we can walk down.

2.1. Variations on alternate possibilities

We can formalise the above intuition into the classic argument for incompatibilist (a.k.a. libertarian) free will:

(1) Principle of Alternate Possibilities (PAP): An agent is responsible for an action only if said agent could have done otherwise.

(2) If determinism is true, no agent can ever do otherwise.

(3) Therefore, if determinism is true then no agent has free will.

But is (2) true? We must clarify exactly what ‘could have done otherwise’ means. There are different degrees of what one cannot do: one cannot make 2 + 2 equal 5, one cannot fly unaided, nor can one under determinism pick up a pencil if that deviates from the causal chain. (2) uses this final notion, and this notion is quite strong. Nothing and no one can make 2 + 2 equal 5, and no human can fly unaided, but no such general conditions prevent the lifting of a pencil: it is prevented by the specific physical causes present at that time. Perhaps then there is a sense of ‘doing otherwise’ that is weaker, weak enough to be possible on determinism, that still preserves the moral responsibility of the agent.

A classical tactic—found for example in Hume’s Enquiry Concerning Human Understanding—is to define this ability to do otherwise counterfactually. That is, we say that I was able to do otherwise if, had I desired to do otherwise, I would have done otherwise. This is distinguished from cases where I am coerced, where I would act the same whatever my desires. And what is freedom but the ability to act as one wishes without coercion?

However this analysis is too simplistic. Suppose Robert is a severe form of psychopath, and is physiologically incapable of desiring to put others’ needs before is own. It is obvious that Robert lacks the ability to do anything other than selfish actions. Yet the above analysis would say that Robert was able to act otherwise, as he would have if he desired to. But of course to say ‘if he desired to’ is to miss the point. Robert can’t desire to, anymore than he can fly. So the classical compatibilist must revise their understanding of ‘ability to do otherwise’ to account for such cases. In the free will chapter of his book Think, Simon Blackburn explores various potential revisions. For example Blackburn defends the revision that one must have been able to desire otherwise were they under the influence of appropriate reasons. No reason can compel Robert to be empathetic, so he lacks this ability.

2.2. The Consequence Argument

We shall not dwell on these various revisions, as in his paper “The Incompatibility of Free Will and Determinism” Peter van Inwagen raises a major issue for any project of reconciling ‘doing otherwise’ with determinism. His Consequence Argument is that if persons have an ability to do otherwise, it follows that persons have some extraordinary abilities. The argument can be stated as follows:

(4) If I have an ability to make P true, and P entails Q, then I have an ability to make Q true.

(5) If determinism is true, then for me to act otherwise entails
that either the past state of the universe was different or the laws of nature were broken.

(6) Therefore if determinism is true and I have an ability to act otherwise, then I have the ability to change the past state of the universe or to break the laws of nature.

As no person would seem to possess the ability described in (6), this argument is a powerful challenge to compatibilism.

In response David Lewis, in his paper “Are We Free to Break the Laws?”, argues that (6) is not as incredible as it seems. He distinguishes between law-breaking actions, actions that cause the violation of a physical law, with actions for which a law must have been broken for them to occur. The former we surely cannot do, but with the latter it is less clear. Suppose it was predetermined that I would not raise my hand just now. Then truly, for me to have raised my hand a physical law must have been violated at some point in the past. But this violation long precedes the raising of my hand, so to raise my hand I need not cause a violation to occur. But it can be objected that even so the compatibilist is still committed to a bold claim. Does it really make sense for us to be able to act in ways that require miracles to have occurred? Van Inwagen contends that such a conclusion, though not incoherent, raises the dialectical price of compatibilism.

2.3. Objecting to the PAP

Fortunately, the compatibilist need not just object to (2) in the above argument. Harry Frankfurt, in his landmark paper “Alternate possibilities and moral responsibility”, powerfully argued against the PAP. He devised scenarios now termed ‘Frankfurt cases’ in which an agent is unable to do otherwise—even in a weak sense—and yet is still responsible for their actions.

A classic example of a Frankfurt case is if Jones were to have a chip implanted in his brain that, if he ever tries to vote Democrat (or third-party), will activate and force him to vote Republican instead. However in fact Jones is a life-long Republican, and so votes for them without the chip ever activating. It seems then that Jones is responsible for voting Republican, even though he could not have voted otherwise because of the chip. We maybe don’t even need sci-fi brain chips here; we can imagine that Jones is threatened with bodily harm if he votes Democrat but still votes Republican for his own reasons. Despite the coercion Jones experiences, his responsibility is undiminished as he is coerced to do what he already wishes to do.

An important reply to this argument is the so-called ‘flicker of freedom’ reply. This says that Jones does in fact have an alternate possibility: he could have tried to not vote Republican, triggering the chip. Advocates of this reply argue that that Jones voting Republican on his own and Jones voting Republican because of the chip are distinct actions, and therefore because Jones had the choice between these actions the Frankfurt case is not a counterexample to the PAP.

In response to this, John Martin Fischer objects that these flickers of freedom do not constitute a robust alternative possibility for Jones that grounds our intuition that he is responsible for his voting Republican. This is because the alternate scenarios involve Jones voting un-freely, and Fischer argues that alternative scenarios in which you lack responsibility can’t be what ground you having responsibility in the actual scenario.

Furthermore, in her paper “Alternative Possibilities and Moral Responsibility: The Flicker of Freedom”, Eleonore Stump argues that this notion of acting-on-one’s-own being a distinct sort of action is confused, and allows for Frankfurt cases with no flickers of freedom. That is, if voting-Republican-on-one’s-own is a concrete action O that Jones takes, then there is nothing to stop us forming a brain chip which will compel Jones to O if he shows any indication of not voting Republican.  Therefore, either the chip is not involved and Jones votes Republican on his own, or is the chip activates and Jones performs O. Either way, Jones performs O. And if we try to run a flicker of freedom reply, we find we are distinguishing voting-Republican-on-one’s-own and voting-Republican-on-one’s-own-on-one’s-own. And it is hard to see how on Earth those could be different. The mistake then is that the flicker of freedom reply conflate the action Jones performs with the context of that action. Doing an action on your own is not a distinct action to doing it not on your own, so these are not alternate possibilities available to Jones.

3. The ‘Source’ model of free will

If Frankfurt is right, then the Garden Path model paints a misleading view of control. Control isn’t about having multiple paths to choose from, it is about your actions coming from you and no other source. This is the Source model of free will.

There are two notions of what sourcehood might mean. The libertarian argues that the agent must be the ultimate cause of their actions: some key part of the action’s causes must originate solely within the agent. Since determinism asserts that the causes of any event originate solely in the initial state of the universe, on this notion of sourcehood determinism precludes free will. Hence, the compatibilist argues that sourcehood requires only that the agent is the relevant cause of the action; they feature prominently enough in the causal chain leading up to their actions to sensibly describe them as their actions.

3.1. Frankfurt, Fischer and Ravizza: Variations on relevant causes

There are many different accounts compatibilists have given as to what makes the agent the relevant cause of their actions. I will discuss just two here.

Harry Frankfurt argued that it is crucial to distinguish between first-order desires, what we desire to do, and second-order desires, what we desire to desire. For example, an addict may have a first-order desire to smoke, but a second-order desire to no longer desire to smoke. Freedom of action is the state of acting as we desire, but this is not enough for free will if we do not also desire to have the desires upon which we act. The addict does as they will, but they do not will to will so. Freedom of will, for Frankfurt, is the state of both our first-order and second-order desires¹ according with our action. By desiring to have the desires that cause us to act, we take ownership of the causes of our action. We are thus its relevant cause.

An important objection to Frankfurt’s analysis is that the focus on second-order desires is a bit arbitrary. Suppose we possess second-order desires that are not our own: we have, say, been hypnotised to want to change our ways. Why should we suppose we are any freer for having second-order desires beyond our control than for having first-order desires beyond our control?

A perhaps less arbitrary account is the reasons-responsiveness view championed by Fischer and Ravizza. This argues that an agent is responsible for an act if they acted in response to reasons. The basic idea here is that an agent responds to reasons if, had they been affected by reasons to act otherwise, the agent would have acted otherwise. However this is a little too simple, as the example of Jones in §2.3 illustrates. Jones votes Republican in response to reasons, but even if in fact he had been presented with compelling reasons to vote Democrat, the chip would still have forced him to vote Republican. The crucial point is that the chip plays no role in the actual sequence of causes that led to his vote, whilst his loyalty to the party did. Thus to quote the SEP:

According to Fischer, an agent, and the mechanism of her action, can be entirely determined in the actual sequence of events in which she acts. Yet the actual manner in which her mechanism responds to reasons could be appropriately sensitive to reasons such that, if different reasons were to bear upon it, it would respond differently, and the agent whose mechanism it is would act differently than she does act.

This I think fares better than Frankfurt’s account, since there is a very clear link between responsibility and your reasons for action being the relevant cause of your action. However both accounts are susceptible to another line of attack, to which we now turn.

3.2. Pereboom: Manipulation and the need for ultimate causes

It is generally accepted that if someone manipulates you into performing an act, then you aren’t responsible for that act. If I hypnotise you to become a murderer, then the murders are my fault not yours. This leads to an important incompatibilist argument, the Manipulation Argument, which has the general form:

(7) Any agent manipulated to act via a manipulation strategy S is not free, despite them satisfying compatibilist-friendly conditions (e.g. reasons-responsiveness, second-order volitions etc.)

(8) Manipulation by S is not relevantly different to physical determination

(9) Therefore, any physically determined agent is not free

Recently, Derk Pereboom has articulated a particularly powerful version of this argument called his Four-Case Argument. Pereboom asks us to imagine a case where Professor Plum murders White, under a series of manipulation cases gradually culminating in determinism.

Case 1: Imagine a case where a team of neuroscientists can remotely control Plum’s mind. Shortly before Plum starts to deliberate over whether to murder White, they push a button that creates in Plum’s mind a strongly egoistic reasoning process that they know will cause him to choose to kill White. Had they not intervened Plum wouldn’t have killed White, but his decision still results from a reasons-responsive process and he possesses the appropriate second-order desires. But clearly, Plum was not free.

Case 2: Now imagine that, instead of manipulating Plum in the moment before deliberation, the neuroscientists manipulate him at birth so that he would be disposed to often but not always reason egoistically. This has the effect that in his present situation he reasons egoistically as in Case 1 and kills White. But surely when the manipulation occurs is morally irrelevant, so Plum is no freer here than he was in Case 1.

Case 3: Next imagine that, instead of a team of neuroscientists manipulating Plum’s brain to be often egoistic, it is instead the cultural forces of the community that he grows up in. The effect however is the same so that he possesses the same disposition he did in Case 2, and furthermore the effects of these forces are complete before Plum has the ability to resist them. But surely whether the manipulation is performed technologically versus culturally is morally irrelevant, so Plum is no freer than in Case 2.

Case 4: Finally suppose that the generic causal forces of the universe, rather than some group of agents, caused Plum to be egoistic. Is this the same as Cases 1-3? Pereboom argues it is, since suppose that in Cases 1 and 2 the manipulation devices are “spontaneously generated” and not the result of any agent. Nevertheless, it seems that their presence would undermine Plum’s responsibility. Thus that Plum is manipulated by agents can’t be relevant, so Plum is not free in Case 4. Therefore compatibilism is false.

Pereboom’s argument illustrates the importance of Plum being the ultimate source of his decision to kill White. It would seem to indicate that if factors beyond Plum control Plum’s decision, then this is not relevantly different from direct coercion.

3.3. The incoherence of libertarian sourcehood?

However, an important worry for a libertarian account is if free actions are not determined by past states of the world, then what exactly are they caused by? And how do I relate to whatever it is that causes my actions? As we have previously discussed, it does little good to say that past states of the world indeterministically cause my choice, as that seems to leave me no more room for control than deterministic causation.

In particular, we have what is called the Luck Objection: if my free action is caused indeterministically, there must be a another world—exactly like this one up until the moment of choice—in which I choose differently. But if this other world is exactly the same, at what point do I influence which action I choose? Isn’t it just luck which way I end up acting?

In answer to this, many libertarians have accounted for free actions in terms of a concept called agent causation. To explain this, we shall have to consider the question of what entities act as causes: substances or events. A substance is at it were a ‘self-standing’ entity, such as a dog or lump of gold, and are the bearers of properties. Contrasted with these are events, which I shall take to be states of affairs at some particular moment in time, involving various substances and their properties. Typically, metaphysicians considered only the latter to be causes: the ball flew through the air because of the event of it being hit by a bat. However, recently the view that some substances can directly act as causes has begun to be rehabilitated among metaphysicians.

The motivation is a move away from a ‘billiard ball’ understanding of causation, in which objects can cause changes only in virtue of themselves being acted upon. For example the cue ball is able to move the red only because it itself was hit by the cue. However, this view is perhaps at odds with contemporary physics: what acts on an atom to cause it to spontaneously decay? Hence the view that certain substances possess active powers—abilities to cause changes without being acted upon—has become more popular. Agent causation is the idea that agents are such a substance, and that agents possess an active power to make decisions (via deliberation on reasons) without prior events determining those decisions.

In his paper “Is our conception of agent-causation coherent?” Derk Pereboom discusses various objections to the coherence of agent causation. He first notes that variations on the ‘luck’ objection seem not to succeed against this view, for in this case which of the various indeterminisic outcomes happens is decided not by luck but by this special power of the agent. However, the more pertinent objection is whether it is coherent for agent causation to be irreducible to event causation. Most of the time when we speak of substances as causes, what we say can be equally spoken of in event-causal terms. If I say that fire has the power to produce smoke, this is the same as saying that the events of incomplete combustion in burning wood produce the smoke. It is another, much more mysterious thing, to speak of substance causation that is not also event causation. Yet this is crucial, or else the luck objection resurfaces. Pereboom concludes that there is some prima facie reason to think that such causation is coherent, but that the issue is not decided. Certainly to me, it seems that this power we are ascribing to agents is very peculiar. It sounds like we are saying that agents are able to make undetermined decisions because they have the power to make undetermined decisions—which is not a satisfying explanation.

3.4. My compatibilism: Sensitivity to changes

Let us take stock of where we stand. Thanks to Frankfurt, compatibilism is freed from its responsibility to give an account of how we could have done otherwise on determinism. However, it must still explain in what sense we are in control of our actions. Such an account of sourcehood must account for why Plum is not free in Pereboom’s Case 1 and yet is free by Case 4. Nevertheless, there is hope for the compatibilist as libertarian sourcehood is also problematic.

In general, a source compatibilist will say that: S was free with respect to their action A if S was the relevant cause of A (or perhaps if S’s Reason is the relevant cause). The rub is what relevant causes are. The idea behind the compatibilism I subscribe to is that S is the relevant cause of A if whether or not S chooses to A is more sensitive to changes² in S than to anything else. That is, in order to make S not A the easiest way to do this is to change facts about S. Consider a manipulation case where S is controlled by another agent M. Then in such cases you can make all sorts of changes to S without affecting what S does, since S’s actions are far more sensitive to changes in M. To change S’s actions without changing M you must undo M’s control, whilst it is much easier just to make M not wish to control S.

To answer Pereboom, we must unpack this more precisely. After all, if the ‘butterfly effect’ is true then a tiny change in the seconds after the Big Bang might make S (and indeed the whole Earth) never exist, yet this sort of sensitivity seems irrelevant to whether S was free. The solution, I think, is to require that whatever changes we consider it remains the case that S has the same choice to A or not A, even if S themselves need not be the same. So for example, it might require only a small change to the history of the Trump family for Donald Trump to not have a fortune and so not run for president. But the choice to run or not run with a fortune is very different to the choice to run or not run without one. So this sensitivity does not mean that Trump was not free to make that choice.

So, how does this view respond to Pereboom? Well, in Cases 1 and 2 we can agree Plum was not free. To make Plum not kill White it is enough just to change the neuroscientists to not choose to influence Plum, and this change is easier as it doesn’t have to oppose anyone else’s control. In Case 3 we notice there is a relevant difference between cultural and technological manipulation: in the former the manipulating factors are more distributed than in the latter. Ergo it requires a more complicated web of changes to undo their influence. I think Case 3 is a lot closer than Pereboom thinks it is, but we can still argue that the changes to Plum’s community’s view of child rearing are smaller than those required to undo their influence on Plum.

But in Case 4 everything is blown wide open. The prior physical causes—running deep into Earth’s history and prehistory—are far more distributed than in Cases 1-3. Most small changes to them will either have no effect at all (the other causes take up the slack) or have changes so dramatic as to make Plum’s choice never occur. To change them in such a way as to steer Plum to have the same choice to kill White and yet make a different decision will be very complicated, and it will be much simpler to make changes directly to Plum. So Plum is the relevant cause in Case 4.

———-

[1] Strictly speaking, I should say second-order volitions here. Frankfurt distinguishes between desiring to have a desire but not desiring to act on that desire, e.g. a psychologist desiring to have a craving for narcotics to understand a patient but not desiring to actually indulge that craving, from desiring to have an effective desire. Only the latter he terms volitions.

[2] For those wondering what I mean by ‘changes’ here—and especially what I mean by the size of such changes—I mean here to evoke Lewis-ian divergence miracles. Lewis discusses issues relevant to our present case in his paper “Counterfactual Dependence and Time’s Arrow”.