Pereboom's Four Case Argument against Compatiblism

I have recently been working my way through some of the arguments in Derk Pereboom’s book Free Will, Agency and Meaning in Life. The book presents the most thorough case for hard incompatibilism of which I am aware. Hard incompatibilism is the view that free will is not compatible with causal determinism, and, what’s more, probably doesn’t even exist. In previous entries, I’ve looked at Pereboom’s critique of non-compatibilist theories of free will. In this post, I want to look at his famous argument against compatibilism.

That argument rests on a series of four thought experiments, each assuming the truth of compatibilism, each involving an agent making a decision, and each provoking an intuitive reaction to that agent’s responsibility. Consequently, the argument is sometimes referred to as the “Four Case” Argument against compatibilism. In what follows, I want to present the argument in as clear a way as I possibly can, and offer some critical reflections.

I’ll proceed in three main steps. I’ll start by reviewing the traditional compatibilist accounts of free will along with Pereboom’s argumentative strategy against those accounts; I’ll then consider the four thought experiments at the heart of Pereboom’s argument; and finally I’ll present some critical reflections.


1. Compatibilism and Manipulation Arguments
Compatibilism is the view that free will exists and is compatible with the truth of determinism. In other words, compatibilists hold that whether a decision is to be judged free (or not) does not depend on whether the decision was causally determined. This raises the obvious question: if causal determinism has no bearing on the matter, what does? The answer is that a certain type of causal sequence is associated with free and responsible decision-making. If the actual causal sequence leading up to a decision fits this type, then we are entitled to say that the decision is free.

So which type of causal sequences do the trick? A variety of accounts have been proposed over the years. Here are four of the more popular ones:

Character-based account: A decision can be said to be “free” if it is caused by, and not out of character for, a particular agent. This is the view traditionally associated with the likes of David Hume. It is probably too simplistic to be useful. Other compatibilist accounts offer more specific conditions.

Second-order desire account: A decision can be said to be free if it is caused by a first-order desire (e.g. I want some chocolate) that is reflexively endorsed by a second-order desire (e.g. I want to want some chocolate). This is the account associated with Harry Frankfurt (and others).

Reasons-responsive account: A decision can be said to be free if it is caused by a decision-making mechanism that is sufficiently responsive to reasons. In other words, if the mechanism had been presented with a different set of reasons-for-action, it would have produced a different decision (in at least some possible worlds). This is the account associated with Fischer and Ravizza, and comes in several different forms (weak, moderate and strong responsiveness).

Moral reasons-sensitivity account: A decision can be said to be free if it is produced by a decision-making mechanism that is capable of grasping and making use of moral reasons for action. This is the account associated with R. Jay Wallace. It is similar to Fischer and Ravizza’s account, but pays particular attention to the role of moral reasons in decision-making.

As you can see, all of these accounts claim that a certain type of causal sequence has the “right stuff” for free will, irrespective of whether the decisions produced are fully determined by those causal sequences.

Pereboom challenges this with a manipulation argument. This is a species of argument that starts with the simple supposition that if a decision by one agent (A) has been manipulated into existence by another agent (B), there is no way in which we can say that this decision has been “freely” made by A. On the contrary, it is produced by something that is beyond A’s control. So if I grab your hand, place a knife in it, and then proceed to use your arm to stab another person, the act of stabbing is clearly not a product of your free will. It is a product of my manipulation. Manipulation arguments then simply add to this starting point the more controversial claim that, on compatibilism, all decisions are effectively manipulated into existence by factors beyond an agent’s control. Consequently, none of them can be said to be free.

To put it more formally, Pereboom adopts the following argument against compatibilism:

• (1) If one agent's decision is manipulated by another agent, then that first agent's action is not freely willed.

• (2) There is no difference between a manipulation by another agent and causation by a causal factor external to the agent.

• (3) On determinism, all of an agent's actions are determined (causally influenced) by at least some factors beyond that agent's control.

• (4) Therefore, on determinism, no agent can be said to freely will their actions (or be morally responsible for them). (from 1, 2 and 3)

• (5) Compatibilism holds that free will and moral responsibility are compatible with determinism.

• (6) Therefore compatibilism must be false. (from 4 and 5)

Most of this argument looks right to me. I’m certainly inclined to agree with premise (1). If I implanted a device in your brain that manipulated your motor cortex in such a way that it allowed me to bring about complex actions through the medium of your body, while at the same time convincing you that you were the one making the decisions, I cannot see how you could be fairly said to have freely willed those decisions. I also think premises (3) and (5) are uncontroversial statements of determinism and compatibilism, and that (4) and (5) are logically valid conclusions from the other premises. That leaves premise (2) as the critical variable. This is the premise that Pereboom’s four case argument is intended to defend.


2. Pereboom’s Four Case Argument
Pereboom’s four case argument works by drawing analogies between four separate hypothetical cases. In each of the four cases, an agent (Professor Plum) decides to kill another person (White) for his own personal advantage. In each of the four cases, Professor Plum’s decision is produced by a mechanism that meets the conditions set down by the different compatibilist accounts outlined above. Despite this, in each of the four cases, the claim made by Pereboom is that Professor Plum is not responsible for his decision. The cases differ in the amount of manipulation involved, starting out with a very clear case of external manipulation by another agent, and moving onto more subtle forms that seem to be akin to standard compatibilist accounts of responsible decision making. The heart of Pereboom’s argument is that these differences in the degree and type of external manipulation are irrelevant to Plum’s lack of responsibility. This then allows him to reach his desired conclusion: there is no difference between manipulation by another agent and causation by factors external to an agent.

To put it more succinctly, Pereboom argues that if there is no responsibility in the first case (Case 1), and there are no relevant differences between Case 1 and all the subsequent ones (Cases 2, 3 and 4), then responsibility (and by proxy free will) is not compatible with determinism. Of course, it is impossible to appreciate this without going through the cases themselves. So here they are (these descriptions are abbreviated from the original):

Case 1: A team of neuroscientists has the ability to manipulate Plum’s neural states at any time by radio-like technology. In this particular case, they do so by pressing a button just before he begins to reason about his situation, which they know will produce in him a neural state that realizes a strongly egoistic reasoning process, which the neuroscientists know will deterministically result in his decision to kill White. Plum would not have killed White had the neuroscientists not intervened, since his reasoning would then not have been sufficiently egoistic to produce this decision. But otherwise Plum’s decision meets the requirements set down by standard compatibilist accounts of free will (i.e. it is consistent with his character, reflexively endorsed by a second-order desires, produced by a mechanism that is sensitive to reasons, moral and prudential).

It seems straightforward enough to say that Plum’s decision is not freely willed in this case. It is the result of external manipulation.

Case 2: Plum is just like an ordinary human being, except that a team of neuroscientists programmed him at the beginning of his life so that his reasoning is often but not always egoistic (as in Case 1), and at times strongly so, with the intended consequence that in his current circumstances he is causally determined to engage in the egoistic reasons-responsive process of deliberation and to have the set of first and second-order desires that result in his decision to kill White. The neural realization of his reasoning process and of his decision is exactly the same as it is in Case 1 (although their causal histories are different).

Again, it seems straightforward enough to say that Plum’s decision is not freely willed in this case. The only difference between this case and Case 1 is that the manipulation took place at an earlier moment in time (during his initial development). But that can’t be a relevant difference. At least not when it comes to assessing free will and responsibility.

Case 3: Plum is an ordinary human being, except that the training practices of his community causally determined the nature of his deliberative reasoning process so that they are frequently but not exclusively rationally egoistic (the resulting nature of his deliberative reasoning processes are exactly as they are in Cases 1 and 2). This training was completed before he developed the ability to prevent or alter these practices. Due to the aspect of his character produced by this training, in his present circumstances he is causally determined to engage in the strongly egoistic reasons-responsive process of deliberation and to have the first and second-order desires that issue in his decision to kill White.

This case is like Case 2. The only difference is that it removes the technological manipulation by neuroscientists and replaces it with cultural and behavioural manipulation. Pereboom’s claim is that the fact that the manipulation is done by some brain implant or programming device, vis-a-vis traditional methods, should play no part in our moral assessment. If technological manipulation undermines free will and responsibility, so too should cultural and behavioural manipulation. Once again Plum is not responsible for his action.

Case 4: Everything that happens in our universe is causally determined by virtue of its past states together with the laws of nature. Plum is an ordinary human being, raised in normal circumstances, and again his reasoning processes are frequently but not exclusively egoistic, and sometimes strongly so (as in Cases 1-3). His decision to kill White issues from his strongly egoistic but reasons-responsive process of deliberation, and he has the specified first and second-order desires. The neural realization of Plum’s reasoning process and decision is exactly as it is in Cases 1-3.

Okay, so this is where things get really interesting. The idea is that Case 4 is like Case 3, only there is no explicit manipulation by another set of agents (neuroscientists or cultural peers). No doubt some environmental manipulation is taking place — we are all, on determinism, products of our historical and contemporary environments — but we don’t know exactly what it is. This is pretty much the view of all causal determinists in the present age. Nevertheless, the differences between Case 3 and Case 4 seem more significant than the differences between the other cases. Bridging the gap between our judgment in Case 3 and Case 4 is crucial to Pereboom’s overall case. How does he do it?

He focuses on two possibly significant differences between the cases. The first is simply our ignorance of the causal history. Could this be a relevant difference? Obviously not. We could be ignorant of a causal history that involved direct neural manipulation of a particular decision, and yet know that manipulation was taking place. Our ignorance of the specific details wouldn’t seem sufficient to hold someone responsible. The second difference is the fact that the manipulation in Case 4 is not being done by a group of other agents. This smells like a more significant difference, but Pereboom argues that, on reflection, it can’t be relevant. To prove his point, he asks us to reimagine Cases 1 and 2 involving the neural manipulation devices. In our reimagining we are to suppose that the manipulation devices are “spontaneously generated”, i.e. not the product of an intelligent designer or anything of the sort. Surely our finding such a device in someone’s brain would entitle us to deem them non-responsible? If so, then the absence of a specific external agent manipulating Plum’s choices in Case 4 cannot be a relevant difference. And if that’s right, then non-responsibility holds across all four cases. Compatibilism is thereby undermined.


3. Does the argument work?
Pereboom’s four case argument is much-discussed, but is it any good? I have some concerns. I think the gap between Cases 3 and 4 is more difficult to bridge than Pereboom lets on. In particular, I’m not a fan of his reliance on thought experiments involving spontaneous neural manipulation devices. For one thing, I find it very difficult to imagine finding such a device in someone’s brain and not being drawn to the conclusion that it was the product of another agent’s direct manipulation. For another thing, and as Steve Maitzen pointed out in the comments section to an earlier post, it seems like thought experiments involving such devices surreptitiously presuppose a difference that Pereboom’s argument is trying to deny. The presence of such a device would be anomalous. That is to say, it would be an atypical cause of someone’s behaviour. It is this atypicality which is likely to drive the intuitive conclusion that Plum is not responsible. But that doesn’t seem to threaten the possibility of responsibility in the more typical cases in which compatibilist conditions are satisfied.

None of this is to suggest that I am not sympathetic to Pereboom’s argument. I am. I too feel that there is something dubious about individualised judgments of moral responsibility in a world in which all decisions can ultimately be traced to causal factors external to the individual decision-maker. But I think there are more persuasive arguments for such scepticism. For instance, I find Bruce Waller’s argument against moral responsibility (which I’ve discussed before) slightly more compelling. Waller’s argument works from a fundamental principle of equality in our treatment of persons and then highlights inequalities in the initial causal conditions of individual behaviour which can have important long-term effects. His suggestion is that those initial inequalities create problems for moral practices like the ascription of responsibility. That said, I accept that Waller’s argument has some doubtful elements too and that the precise implications of free will scepticism for our moral practices is something that requires a longer treatment, especially given that Pereboom himself accepts that most of those practices remain intact in the face of his scepticism.

But those are just my criticisms. There are many others discussed in Pereboom’s book, and he is nothing if not meticulous in his treatment of his critics. I can only give an inkling of what he does here. He starts by noting that there are two basic responses to his argument: the hard line response, which tries to argue that Plum is responsible in all four cases; and the soft line response, which tries to argue that there are differences between the cases which undermine Pereboom’s argument. My comments above are examples of the latter because they suggest that there are important differences between Cases 3 and 4.

An example of a hard line response can be found in the work of Michael McKenna. He says we should examine the four cases by initially being agnostic about the implications of determinism for free will and responsibility. In other words, we should not (prematurely) assume that determinism undermines responsibility. If we do this, then we would simply be doubtful about Plum’s responsibility in Case 4, and we would transfer that doubt over to Cases 3, 2 and 1. The analogy between the four cases consequently works in the opposite direction to that supposed by Pereboom.

Pereboom has a response to this. He thinks McKenna’s agnosticism is too strong. McKenna seems to assume that we should start by being confirmed agnostics about responsibility in Case 4 and then transfer that agnosticism back to the other cases. But really we should be neutral inquirers. That is, we should start off being unsure about the effect of determinism on free will and responsibility. This means we are open to the possibility that determinism undermines free will and responsibility, and that it does not. If we start with this attitude, then McKenna’s desired transference of agnosticism is much less compelling.

I tend to think Pereboom is right about this — we should be neutral inquirers, not confirmed agnostics — but maybe neutrality is not the right stance. As Steve Maitzen also pointed out to me, why should the truth of a metaphysical thesis like determinism have any implications for an ordinary commonsense moral practice like holding others responsible for their actions? I’m not sure exactly what Steve was getting at with this comment, but I’m guessing it is something to do with the fact that our beliefs about ordinary moral practices are more solid than beliefs about recondite and abstruse metaphysical theses like determinism. But what does this mean in practice? Does it mean that no metaphysical claim can unseat our ordinary moral beliefs? Or just that it would take a lot of evidence to unseat such a belief? I can’t answer those questions on his behalf.


Conclusion
This has gone on for too long. It’s time to wrap things up. To briefly summarise, Pereboom’s four case argument tries to undermine compatibilist theories of free will. It does so by suggesting that there is no difference between cases involving manipulation of decision-making by other agents and cases of ordinary compatibilist causation of behaviour. Since no one thinks we exercise freedom in manipulation cases, it follows that we don’t exercise freedom in cases of ordinary compatiblist causation of behaviour either.

There are several responses to Pereboom’s argument. Hard-line responses try to argue that responsibility (or, at least, agnosticism about the effects of determinism on responsibility) holds across all four of Pereboom’s cases. Soft-line responses try to show that there are important differences between the four cases. For my part, I think that there are differences between Cases 3 and 4 that are harder to smooth over than Pereboom lets on. I am also inclined to think that there are more persuasive arguments for free will (and responsibility) scepticism.