Time and the Compatibilist

A Friday provocation...

Compatibilists claim that moral responsibility is compatible with determinism. Even if facts about the universe at t, in conjunction with the laws of nature, entail that an agent acts in a certain way at t1, that agent may be morally responsible for acting in that way. Now suppose that time travel to the past is one day a practical possibility. In that case, it might be possible to go back in time to a point prior to t1 and punish (or reward) an action, even though it has not yet occurred. Compatibilists seem committed to holding that this kind of punishment would be just. But this is extremely implausible, so implausible that an account of moral responsibility that entails it cannot be accepted.

Compatibilist Replies and Responses

Incoherence

The compatibilist might argue that the scenario is incoherent. Suppose the punishment meted out by our time traveller (call him Tim) represents a novel intervention in the history of the malefactor, in the sense that prior to the traveller’s observation of the act at t1, no such punishment had occurred. In that case, an event (the malefactor’s punishment) both would and not have occurred at some time prior to t1. Familiar paradoxes raise their head at this point, but here I want to draw your attention to a less familiar problem. If Tim is able to intervene in the history of the malefactor (call her Marie), to make it the case that she is punished for an action before it takes places, even though before Tim observed Marie performing the wrongful act the punishment had not taken place, then determinism as defined above would not hold. It would not be true that Marie is causally necessitated by the state of the world and the laws at nature at t to perform her wrongful act at t1. Not only might she possess the ability to act in such a way that, were she to act in that way, the laws of nature or past states of the world at t1 would have had to be different (Lewis 1981), but at least one person – Tim – could act in such a way as to bring it about that past states of the world at t1 actually are different. But if this is the case, then determinism is false. There is no world in which determinism is true and backwards punishment is possible.

Response

The plausibility of this compatibilist reply depends crucially on the kind of punishment envisaged and on the metaphysics of time. On an eternalist view of time, an intervention of the kind envisaged in the reply seems to be impossible. On this view, time is fixed and we cannot change the past even if it proves possible to observe it. But kinds of punishment that do not alter the past seem possible, and other views of time may be true. First, backwards punishment may be possible on the eternalist view. It would have to be the case that the punishment, though caused by time-travelling Tim, always preceded the wrongful action. There is nothing incoherent about this idea (it needn’t involve closed causal loops, for instance), but it does raise some complications. If the punishment now precedes the crime, how can we be sure that it isn’t causally necessary for the commission of the crime? Perhaps Marie, embittered by what she perceives as her unjust treatment, turns to a life of crime in protest. In that case, her punishment does seem unjustified, and the counterexample to compatibilism fails. However, there seem to be ways in which this problem can be avoided. Perhaps the events which constitute punishing Marie take place prior to t1, but do not have any impact upon her actions at t1 since she does not learn of her punishment until after that time (perhaps her bank account is emptied by Tim).

Even if this problem were to prove insurmountable, an eternalist view of time might prove false. Suppose the many-worlds interpretation of quantum mechanics is true, but that macroscopic determinism governs the multiverse. In that case, it would be possible for Tim to observe Marie’s wrongful action at t1, and travel back in time to punish her, secure in the knowledge that his punishment would not intervene in the causal chain leading to her action. His punishment would, to be sure, impact causally upon the Marie punished, but the Marie punished and the Marie who performed the original wrongful act would not exist in the same universe. Nevertheless, since macroscopic determinism is true, Tim could be confident that his punishment is just: the Marie punished would have performed the wrongful act without his intervention. On the Strawsonian view, which underlies many compatibilist theories of moral responsibility, the reactive attitudes are a justified response to the good or ill will of agents (Strawson 1962). Tim knows all he needs to know about the quality of Marie’s will to justify punishing her.

Alternative Possibilities

The compatibilist might insist that she is not committed to the justifiability of backwards punishment, since she holds that alternative possibilities (understood in a distinctively compatibilist manner) are a necessary condition of moral responsibility. Since Marie is punished before her wrongful act, she is punished before she chooses between alternative possibilities; she is therefore not morally responsible for the act at the time of the punishment and the punishment is unjust.

Response

Marie will have alternative possibilities, in the compatibilist sense, at the time of her action, even if she undergoes backwards punishment. If it is ever true that agents are responsible for X-ing in a deterministic world because, if they had wanted to Y instead of X-ing, they would have Y-ed, then Marie is responsible for the wrongful act. But since we know that Marie will freely (in the relevant sense) perform the act, her punishment is just. In any case, many compatibilist have now abandoned the thought that alternative possibilities are a necessary condition of moral responsibility (Fischer and Ravizza 1998; Dennet 1984; Frankfurt 1969).

Biting the Bullet

The compatibilist might insist that backwards punishment is just. We balk at the thought, she might argue, because though the world in which we live is or might be deterministic, the future is unknown. Perhaps none of us have alternative possibilities metaphysically open to us; nevertheless the future is epistemically open. Because we do not know the future, we cannot punish people for what they will do: we might wrong, and if we are, we act unjustly. But in the time travel scenario, the future is known, and backwards punishment is just. The gut reaction that it is not is simply the product of the illegitimate perserverance of intuitions which are a response to epistemic openness.

Response

Suppose the wrongful act performed by Marie is very wrong – on the order of mass murder, for instance – and that the only time at which we can punish her is significantly prior to t1. In that case, the view that it is just to punish her is highly counterintuitive. Of course, worries about personal identity might arise, but they seem no more pressing than the similar kind of worries which arise with ordinary (forwards) punishment. Suppose that the only way to prevent Marie from committing her crime is by killing her prior to her even contemplating the crime. Compatibilists need not think that this is just (they need not subscribe to a consequentialist account of punishment, for one thing), but they are committed, it seems, to holding that it is less unjust than killing an innocent person at random. And that seems highly counterintuitive.

Since we know that Marie will commit the wrongful act at t1, there seems no reason not to punish her at some time prior to t1 (assuming that punishment is ever justified). Once epistemic worries are set aside, as they are in the time travel scenario, it does not seem to matter, from the point of view of justice, whether punishment precedes or follows the crime. And that is a hard bullet for the compatibilist to bite.

Comments

[Neal Tognazzini]

Though not a compatibilist myself, here's how I think the compatibilist ought to respond.

Bite the bullet. But only in a restricted sense. The only way punishment would be justified is if it's true that Marie will perform the wrong action in the future. Then you can punish her before she performs the action. But since your punishment is also part of the deterministic world, your punishment cannot include anything that will "change" the facts about the world -- that is, it can't be the case that as a result of your punishment, Marie doesn't actually perform the wrong action in the future. So, you can't punish Marie by, say, taking her out of existence before she performs her action - then we are no longer considering a world in which it is true that Marie will perform a morally bad action.

So, if you're a time traveller in a deterministic world, you can punish a crime before it occurs, but you can't *prevent* the crime from occurring.

Given that the crime will in fact occur, I don't see why this bullet is so hard for the compatibilist to bite after all.

[Neil]

Neal,

I addressed the point about changing the world. In a single track deterministic world, I think the compatibilist *can't* prevent the crime taking place (as Lewis argues, you can't change the past). But in a branching world, you can. Your response seems to address the single track world only. I agree the bullet is more palatable here. But what about the branching world? If I can prevent WWII from happening (on one branch) by assassinating Hitler when he was a baby, then clearly I should on consequentialist grounds. The compatibilist who is also a deontologist (and compatibilists typically are) must also say that assassinating Hitler when he was a child in such a case is just (or at very least, less unjust than killing an innocent person who would cause a similar disaster through no fault of her own). That seems a harder bullet to bite.

[tnadelhoffer]

I am curious whether Neil is correct about compatibilists and deontology. I suspect the opposite. Could the esteemed readers of GFP--especially those afflicted with compatibilism (heehee)--please weigh in on this one.

[Neal]

Maybe you could say a bit more about the multi-worlds story. I'm not sure I understand it fully yet, but here is an initial (and perhaps confused) thought.

Is it compatible with the fact that macroscopic determinism governs the multiverse that in one world a time traveller kills Marie before she commits the crime and in another world no time traveller exists before Marie commits the crime? And if the worlds can be different in this sense, then what allows the time traveller to be so sure that the world he has time travelled to is one in which Marie would commit the crime without his intervention?

I was under the impression that if the many-worlds interpretation is true, but macroscopic determinism still reigned, there could not be two distinct universes that are macroscopically different. But aren't they, on the scenario you have described?

[Neil]

Thomas,

I must admit that my impression is that compatibilists tend to deontology is just that - an impression. It is fostered, in part, by the observation (again, not systematic) that hard determinists tend to support punishment on consequentialist grounds. I think there is a stronger case to be made for the libertarian/deontologist link. But once again, my impression is that compatibilists tend to be at least weak deontologists, by which I mean only that deontology plays a role in their moral thinking. If you want to know under what conditions blame is justified, and you don't mention utility in setting out those conditions, you seem ipso facto to be relying upon deontological considerations.

Neal,

Here's how I thought the many worlds/macroscopic determinism story worked. Tim observes Marie commit the crime and travels back in time to punish her, where the punishment is incompatible with her committing the crime. Since you can't change the past, his very act of punishing her splits the time line, into two worlds, in only one of which she commits the crime. This story seems to be consistent with my reading but I have to confess that my reading is shallow and the science is basically a mystery to me.

Neil

[Mark Smeltzer]

Here's the scenario, or at least a relevantly similar scenario, boiled down to three facts:

F1. At time T2 agent X knows that (1) agent Y commits evil action A at T1, and (2) that Y was responsible for A.
F2. At some point T3 agent X is transported through time to T0 at which X apprehends Y, thus preventing Y from performing A at T2.
F3. T0 < T1 < T2 < T3

Since agent X has gone back through time, it is certain that the truth of F1 is no longer guaranteed, and it is certainly false if X intervenes and apprehends Y before T2, or otherwise prevents Y from committing A at T2. If my beliefs regarding the nature of justified punishment are true (c.f. my response to Tom Clark at http://gfp.typepad.com/the_garden_of_forking_pat/2004/11/jaccuse.html#c2999432), then knowledge of an agent's character is a critical component of justified punishment. Since the conditions of the case dictate that F1 becomes falsified in the alternate timeline, X has no relevant knowledge regarding Y's character in the alternate timeline -- the knowledge he possessed in the original timeline of Y's character in relation to A at T2 died with that timeline.

Therefore, punishing Y in the alternate timeline for an action Y never had a chance to commit in the alternate timeline would be unwarranted on *these* grounds. In the alternate timeline the beliefs X has regarding Y's character may constitute a sort of pseudo-knowledge, and are in fact true, but that is insufficient to fulfill the epistemic requirements for punishing Y in the alternate timeline. This has to do with the way X came to know F1 in the original timeline: X's time travel invalidates that which previously grounded the knowledge of F1 -- namely the existence of the actual states of affairs that made F1 true in the original timeline.

I do not think this analysis applies to the idea of a being with perfect foreknowledge punishing an agent for an action it has not yet committed because that being's knowledge would not be contingent upon the existence of the underlying states of affairs.

While others may have to "bite the bullet" here, I do not, and I am sure that other compatibilist views that emphasize the epistemic requirements for justified punishment could adjudicate this scenario in a similar way.

[Neil]

Mark,

Your reply depends, firstly, on what the metaphysics of time turns out to be. It doesn't follow, from the fact that time travel could be possible, that events are not determined, even events observed by the time traveler (see Lewis's well known paper). F1 might still be guaranteed true. That's one scenario; on that scenario (as you recognize) punishment that pre-empts the crime is not possible. But *some* kind of punishment might be.

A more interesting case arises if pre-emptive punishment *is* possible. You seem to think that if that's right, punishment wouldn't be just, because punishment is a response to knowledge of character states. But if there are any such things as robust character states, they are dispositions. If you phi in circumstances C, that tells me something about your character. I know how you would act in C. C needn't be an actual state of affairs. Tim, it seems, has knowledge of Marie's character that is as well justified as any knowledge of character can be (he has seen how she acts in C). So his punishment is justified.

[Mark Smeltzer]

Neil,

You may be right about the metaphysical dependencies, but even if time lines are parallel and thus co-exist, I don't see how knowledge gained from one time line could apply in this normative sense to another time line.

If parallel time lines do not exist, then Tim doesn't know Marie's character because the truth conditions for his beliefs (though produced in the right way in the original time line) fail due to the alterations caused in the time line. In this circumstance, we can presume that he has true beliefs about Marie's character, but that they do not constitute knowledge.

If time lines are parallel, then Tim doesn't know Marie's character because his beliefs regarding Marie's character were produced by events in the original time line, not by events in the alternate time line. (And furthermore, his beliefs about Marie’s character may be false in the alternate time line.)

Again, I am not asserting that punishing an agent with compatibilist freedom for an action it has not yet committed is metaphysically impossible. The case could be modified to provide assurance that Tim does in fact know about Marie’s character in the alternate time line, thus justifying punishment in the alternate time line, but I frankly have no idea how that could be accomplished.

[Neil]

I just don't see why Tim doesn't have knowledge about Marie's character. The world is macroscopically deterministic, and Tim's punishment splits the time line. Tim knows how Marie would have acted if he had not intervened; he therefore knows about the dispositions underlying her behavior and that just is to know about her character. You say that he has true beliefs about her character, but I don't see why he doesn't know that he has true beleifs about her character.

[Kip Werking]

I just want to add that I thought this was a brilliant post. The idea of time travel reminded me of one of my own ideas, which I had called Time Travel compatibilism. Now, if you're willing to venture even farther into the realm between sense and non-sense (perhaps where all good philosophy must be done), consider this:

Many incompatibilists (libertarians or skeptics) generally require that an agent satisfy certain conditions about some regress in time. In particular, they would require the agent to be in control of the initial conditions (IC) of themselves. The problem is that these IC usually exist well before the agent is even born (perhaps as early as the Big Bang), and so the agent cannot be held responsible for them.

But one day I thought: what if the person could travel back in time? What if the agent could travel back in time to the Big Bang, and change the parameters of the universe to create their life? For a moment there, there seems to be some hope for the libertarian with a Time Machine.

However, a moment's reflection shows that even if the agent's original IC were consistent with what the agent wanted them to be--if she had the power to alter her IC but chose not to--she would not have free will. Her freedom does not survive the sort of objections that Galen Strawson makes. In particular, although we might view time as a loop, at some point it must have started, and at that point the agent just had whatever IC she was given--not that she chose. On this view, whether time goes in a circle would not give her freedom of the will because she will be stuck with whatever IC she originally had--just as we all are. Realizing that not even Time Travel Compatibilism would rescue libertarianism showed me just how "hard" Galen Strawson's position is.

This is Galen Strawson's fundamental point: that algorithms do not write themselves. It is logically impossible to do so. Algorithms can modify themselves--self-modifying programs are an exotic form of computer programming--but this does not satisfy the libertarian's hunger. How a self-modifying program changes itself is purely a function of its initial conditions--which are beyond its control. The libertarian tries to chase her tail (or lift herself up by her bootstraps, to use Nietzsche's phrase) but is doomed to failure by logic.

Of course, some algorithms are more sophisticated than others. Some manipulated agents are more responsible than others. Even if God designed and set up every moment of our lives, persons vary in their mental capacities from ordinary persons, to kleptomaniacs, to chimpanzees. The fact that manipulated persons--who cannot choose their own initial conditions in Galen Strawon's sense--can nevertheless admit of such variation seems to motivate compatibilists. Compatibilists look at the differences between an ordinary, reasonable person and a kleptomaniac and say "Look! Even if we are all the puppets of some distant God, we are not rocks! We have mental equipment which allows us to take responsibility for our actions, punishment deters us, and so on."

Galen Strawson's position makes no mention of this important compatibilist distinction. Likewise, most compatibilists do not seem to appreciate Strawson's fundamental insight. They are using the same words (free will, moral responsibility), but talking about different things. This is what I think Fischer meant by his comment (if I recall correctly) that moral responsibility "doesn't involve the cards you are dealt, but how you play them."

One final comment: The thrust of Neal's post seems to be "in a deterministic universe, choices are made before the agent appears to make them." A Laplacian demon would be able to predict our behavior before we behave. But this conflicts with our sense of choice. There seems to be a metaphysical element to our understanding of choice: we are not just translating previous states of the universe into new ones, rather, we are injecting new information into the universe, information that did not exist before we chose. Neal makes the clever point that if you are going to hold people responsible for their choices, and if the choice has been made since the beginning of the universe (before the person was born), why not do so before they "appear to choose?" Only holding them accountable after they "appear to choose" for information which has existed in the universe since the Big Bang would seem to be arbitrary. Punishing the person after they "choose" would be just as unjust as punishing them before. This clever insight is, I think, correct.

[David Hunter]

I will echo Kip and note that this is a very interesting discussion, Neil. In response to Kip, there are ways to respond to Strawson's worries here is one, which I don't personally endorse but is at least logically possible. Suppose we believe dualism of a fundamental kind and that we exist pre-birth. Suppose the realm we live in then is libertarian, though this world is deterministic. Suppose also that we are excellent predictors of lives in the deterministic world, and we select which life we choose to be born into. Thus we can be held responsible for our actions, even though this is an entirely deterministic world. This hold since although we now have no choice about what we do, we did have choice about which initial conditions would apply with knowledge of the consequences.

[Neil]

Thanks for the kind words, Kip and David. Kip, I think you've put your finger on one very real, but usually underplayed, difference between compatibilist and incompatibilist intuitions. Incompatibilists think that how you modify yourself is a function of your intial state, whereas compatibilists think that if you modify yourself to some extent, by endorsing and rejecting the hand you're dealt, you become autonomous. Before I became an analytic philosopher, I was on the dark side: I wrote my dissertation on Sartre and Foucault. It's interesting to see exactly this dialectic play itself out in their work. They both begin with the intuition that if self-alteration is a function of the initial state, then there is no freedom (Sartre rejects the antecedent, since he rejects determinism, Foucault accepts both sides). They both gradually become compatibilists in the course of their work. Sartre writes something like 'freedom consists in being able to make something of what your society has made you'. This is all more naturally discussed under the heading of autonomy, but as John has written, the autonomy debate and the free will debate are essentially the same.

[Paul Torek]

Neil, in your Dec 10 response you consider the justice or injustice of

killing an innocent person who would cause a similar disaster through no fault of her own.

Now, consider the rights of self-defense and defense of others. It is mostly agreed that one has the right to kill an "innocent aggressor". That is, if someone threatens your life through no fault of their own - they are drowning after a shipwreck, understandably panic-stricken, and climbing on top of you so that you can't swim, for example - you have the right to use force against them. Lethal force, if necessary.

Similarly, one would have the right to defend others against an innocent aggressor, especially if there were many others whose lives were at stake.

Even if Marie were an innocent aggressor, therefore, one would have the right to enter an alternate time-line and use force to prevent the massive carnage. Provided, of course, that Mark's worries can be answered, and that you really do know that without intervention Marie would create this disaster.

Bite the bullet? I'd say it tastes more like candy.

[Neil]

That's a nice response, Paul. My worry is this: reactive attitudes seem to be appropriate with respect to Marie (resentment, for instance) that are not appropriate with regard to an innocent aggressor. We may be entitled to preempt Marie's actions, using lethal force if necessary, but we're not entitled to blame her, it seems. But compatibilism cannot explain why we're not entitled to blame her.

[Paul Torek]

I'd say we are entitled to blame her, if she has developed the character that makes her blameworthy. We can blame the Marie in our own time-line for being a mass murderer, and the Marie in the alternate time-line for being such an evil person as to necessitate our time-travel.

Your point seems to involve the principle that we can blame a person only for an event that has actually happened. But, why can't the compatibilist endorse this very principle?