Cohen on the Impossibility of Responsibility

In a recent article in Philosophical Studies ("Openness, Accidentality, and Responsibility", Feb. 2006, pages 581-597), Daniel Cohen puts forth a novel argument for the conclusion that moral responsibility is impossible because the necessary conditions for morally responsible agency cannot all be satisfied at the same time.  But I'm not convinced.  And here's why.

First, let me lay out his argument.  The argument runs as follows:

1)      If S is responsible for doing A, then S’s doing A is not accidental.

2)      If S’s doing A is not accidental, then S’s character determines that she do A.

3)      If S is responsible for doing A, then it is open to S not to do A.

4)      If it is open to S not to do A, then either S’s character underdetermines her doing A, or it is open to S that her character be such that she not do A.

5)      If it is open to S that her character be such that she not do A, then S is not constituted by her character.

6)      S is constituted by her character.

7)      If S is responsible for doing A, then her character both determines and underdetermines her doing A.

Since we are led into a contradiction, this argument constitutes a reductio on moral responsibility.  Cohen presents the argument near the beginning of his paper, and then spends the rest of the paper defending each of the premises.  Here's the basic idea.

Premise (1) seems relatively uncontroversial to me, so we'll take that as given.  Premise (2) is where the action comes in.  Cohen lays out three possible ways that he thinks an action may be accidental, and then argues that the accidentality in each is due to underdetermination.  He concludes that in order for an action to be not accidental, it must be determined by the agent's character.  The three ways an action may be accidental are:

a)      where S’s belief/desire set underdetermines her action

b)      where S’s process of reasoning underdetermines her belief/desire set

c)      where S’s psychological dispositions underdetermine her processes of reasoning

(I guess I should also mention that Cohen is here working with the following conception of what happens when someone acts: "As manifested in the occurrent environment, S's psychological dispositions (or her 'character', as I shall say) cause S to engage in a process of reasoning which results in her having a particular belief/desire set.  The strongest of S's desires then cause S to act as she believes is necessary in order to satisfy that desire.")

As an example of accident (a), Cohen gives someone's being tied to a chair, as an example of accident (b), Cohen gives addiction, and as an example of accident (c), Cohen gives brainwashing.  In each of these cases, a connection that is crucial to the metaphysics of agency is undermined.  But -- and here's the important point -- Cohen thinks that in each case, "simple indeterminacy is sufficient to undermine [the] connection".  All of his cases to support this claim involve some sort of random process in the agent's brain that makes it indeterminate whether the agent's belief/desire set will result in that particular action, or makes it indeterminate whether the agent's process of reasoning will result in that particular belief/desire set, or makes it indeterminate whether the agent's psychological dispositions will result in that particular process of reasoning.  In each case, Cohen thinks that S's doing A is accidental, and concludes that what is required for non-accidentality is determination by character.

But here's the problem.  If I'm reading Cohen correctly, he appears to be arguing as follows: "If there were a random process in S's brain at any of junctures (a), (b), or (c), the result at those junctures would be accidental, and hence S's doing A would be accidental.  Therefore, in order to avoid accidentality, we need the result at each juncture to be determined by the agent's character."  But it seems to me that this reasoning leaves out an implicit (and false) premise, namely: "If the result at each juncture were not determined by the agent's character, then the result would be due to mere randomness."  But isn't this to confuse indeterminism with randomness?  That is, just because a random process in the agent's brain would seem to make the result at each juncture accidental doesn't mean that we need determinism to fix the problem.  We might also be able to use an indeterministic process that is (somehow) not purely random.  I agree that this is one of the tough challenges for the libertarian -- to explain how indeterminism can nevertheless not be mere randomness -- but it's certainly a challenge that many libertarians are willing to take head-on, and so it seems at best unfair of Cohen to gloss over this point.  There might be a way to get rid of randomness without adding determinism.

Moving on.  Premise (3) would be rejected by anyone who denies PAP.  (Cohen considers the Frankfurt examples, and gives a response to them, but I won't get into that.  The type of Frankfurt-example he considers isn't a "prior-sign" example, and I'm inclined to think that the strategy he uses against it wouldn't carry over to the prior-sign cases, but that's just a hunch.)  I take it that this is not trivial -- that Cohen's argument relies on PAP is a weakness that other arguments for the impossibility of responsibility, like Strawson's, do not share.

The rest of the argument is pretty straightforward.  If you think an agent has alternative possibilities, and you accept Cohen's picture of what happens when someone acts (above), then the alternative possibilities could be placed at one of the three junctures mentioned above (a, b, or c).  But if it's placed at any of these junctures, we're back to the underdetermination problem argued for in connection with premise (2).  The alternative possibilites could also be placed, according to Cohen, prior to the psychological dispositions.  That is, "the agent might have been differently disposed to reason than she is actually disposed".  But if the alternative possibilites are there, then you run into an incoherence if you try to spell out how that's a case of openness for the agent, because the agent just is, in an important way, her character.

So, I find the argument interesting, and I recommend it as reading for all, but I utlimately reject Cohen's conclusion because I think both premise (2) and premise (3) are false.  (In fact, I'm also inclined to reject his picture of what happens when someone acts, and so reject his argument since it relies on that picture, but that's for a different post.)

Comments

[Ben Golden]

Neal, I agree with your assessment. The only point I'd disagree on is that Cohen's argument is novel. Philosophers have been trying to argue against free will for eons--I believe the first attempts was by Gorgias in his Encomium of Helen--but their arguments tend to rely upon the implicit assumption of their conclusion.

Your point about indeterminism vs. randomness is quite valid. Suppose for instance, that a spiritual entity, which doesn't exist in our universe, is able to cause an effect in our universe. We have no way of perceiving the entity but we can perceive the effect of its actions. From our frame of reference, we may conclude that the effects are random, since we can perceive no cause. Yet we've assumed that there is a cause and that the effects are nonrandom.

[Daniel Cohen]

Hi Neal,

Thanks for your perceptive and careful criticism. I’m not entirely confident as to how best respond, but here is an initial response.

You characterise a key argument as involving something like the following inference:

(1) If a psychological juncture at (a), (b) or (c) is indeterminstic, then an agent’s behaviour will be accidental.
(2) Therefore, in order for a juncture to be non-accidental, it must be deterministic.

Your argument, I take it, is that (1) is question-begging. We mustn’t simply assume that indeterminism always makes a process accidental. At least conceivably, some indeterministic processes might still be non-accidental, so the inference to (2) doesn’t follow.

In reply, I would state that I don’t simply assume (1). I try to argue for it, more or less as follows:

(3) Compulsion, addiction, and brainwashing are all intuitive instances of accidentality.
(4) A clear, unified characterisation of the accidentality in all three cases is as follows: in all cases, an agent’s behaviour isn’t determined by her character.
(5) Therefore, whenever an agent’s behaviour is not determined by her character, that behaviour is accidental.

It seems that you now have two options

(a) You may wish to challenge step (4) of this argument. That is, you may resist the characterisation of the accidentality in the three cases in terms of lack of determination by one’s character. That is, you may argue that there is some alternative characterisation of accidentality, one that doesn’t imply lack of determination by one’s character.

To mount this argument, I suggest, the onus is on you to offer an alternative characterisation of accidentality.

(b) You may, alternatively, wish to resist the inference between (5) and (1) (that is, you may argue that (3, 4, 5) is not a good argument for (1). To do this you, you would have to argue that causal indeterminism doesn’t necessarily imply that an agent’s behaviour isn’t determined by her character.

Again, I would suggest that the onus is on you to explain why indeterminism is consistent with determination by one’s character. I consider, and reject, one such proposal, in the paper, by Robert Kane. I would be very interested to hear if you have any alternative proposals.

Finally, please note that I have no particular investment in the claim that indeterminism implies accidentality. Indeterminism only arises dialectically, as follows:

My argument is that any case where the non-accidentality requirement is satisfied will not satisfy the openness requirement. Libertarians may propose indeterminism as a source of openness. I can respond to this claim in two ways. I can either argue that indeterminism violates the non-accidentality requirement (as we have seen). However, if some indeterministic process could be described that didn’t violate non-accidentality, I could argue, rather, that the indeterminism described isn’t sufficient to characterise the openness requirement. (I raise this, rather obliquely, in footnote 3 of the paper.)

[Paul Torek]

Thanks for the links. I find the most interesting premise to be 4. Little seems to be said in the paper specifically to defend it.

As an alternative, I urge that it is open to S not to do A if the aspect of character that determines her doing not-A is specifically that she comprehended the reasons in favor of A and found them wanting. It is also open to S not to do A if the reason she does not-A is that she rationally arrived at a policy which licenses ignoring actions of a class that includes A (and nothing warrants abandoning the policy). And I suppose one could go on in a similar vein.

I would further urge that "it is open to S that her character be such that she not do A" is satisfiable in much the same ways. Not that it is necessary to go to this meta-level of self-control to understand the basics at the level of immediate action.

[Kip Werking]

Some initial and rambling comments on Cohen’s argument (I haven’t read the whole paper yet):

1. I think there are plenty of ways to break the (partial) cognitive illusion that people are (ever) morally responsible. Cohen’s seems to me to be one such way. But I wish that those who deny the existence of moral responsibility would make a greater effort towards unification. Pereboom’s four case argument (together with arguments against libertarianism), Strawson’s Basic Argument, Double’s arguments for the impossibility of moral responsibility (because objective moral facts do not exist) and, later, his argument for free will subjectivism, and now Cohen’s argument… I would like to see philosophers explore what unites and mutually motivates these arguments. For example, it seems to me that something like TNR motivates most or all of them.

2. Cohen’s argument, and critique of Kane’s view, seems very similar to Double’s argument (that it is logically impossible for one to satisfy, simultaneously, all three of Kane’s criteria for a free action). See Double’s The Non-Reality of Free Will. Furthermore, his argument against Kane’s business woman example seems similar to the luck objection, about which there is a giant literature.

3. It seems to me that Cohen sometimes conflates determinism with something like “the relevant processes in the brain are deterministic” and indeterminism with the opposite of that. Mark and I talked about this issue in the thread Mapping the Terrain.

4. The ultimate problem, I think, is that people are not causa sui. The problem is not that we lack alternative possibilities. There are two ways to respond to this problem. One way (i) is to say: “I take ownership of my character. In the past, I magically was causa sui, and the genetics and environment played no part in creating my character, only I (who did not yet exist) created my character.” This response has not attracted much attention, because it is obviously non-sensical. But if one could, magically, do this, then I don’t think this would entail AP (although it might entail KPAP, see below at item 5). The other strategy (ii) is to say: “true enough, I have not been causa sui and I did not form my own character. But my character does not determine my fate; I can just as well radically revise and shun my character now, as if it never existed. I can begin, this instant, to be causa sui.” And again, if one could magically do this, I am not sure that alternative possibilities are required. This is the crucial issue (and I may talk about this at Inland): does being causa sui entail ever having alternative possibilities? And the answer may be no. Does being a married bachelor require alternative possibilities?

The problem, I think, is that Cohen’s argument, because of its premise (3), only attacks strategy (ii). If he wanted to attack strategy (i) as well, I suspect he would need to say the equivalent of Strawson’s Basic Argument.

5. Neal seems correct when he says: “that Cohen's argument relies on PAP is a weakness that other arguments for the impossibility of responsibility, like Strawson's, do not share.” Cohen is relying on his premise 3: “If S is responsible for doing A, then it is open to S not to do A.” But it is not clear that this premise is true. A more accurate statement might be “If S is responsible for doing A, then it was once open to S not to do A.” This is so because one might reasonably hold an agent responsible, at least according to typical compatibilist standards, for actions which the agent locked itself into performing (it was once open to the agent to do either, but at the time it was not, as in the Luther example). And the most accurate might be “If S is responsible for doing A, then S once thought that it was open to S not to do A, whether S was correct in this belief or not” (because it is the agent’s internal belief about its own freedom, and not its actual freedom, which is relevant to moral responsibility; Frankfurt examples try to show this but it is not clear to me that they succeed, nevertheless this remains intuitively obvious to me). Let’s call this KPAP, for Kip’s Principle of Alternative Possibilities.

Fitting KPAP into premise 4, one gets: “If S once thought that it was open to S not to do A, then either S’s character underdetermines her doing A, or it is open to S that her character be such that she not do A.” But this change seems to break the premise. The problem seems to begin with premise 3.

[Kip Werking]

After looking at Cohen's argument again, I see that he has built the distinction I draw (between stragies i and ii) right into his premise 4. So my point 4 (above) is mistaken.

[James Gibson]

"“If S is responsible for doing A, then S once thought that it was open to S not to do A, whether S was correct in this belief or not” (because it is the agent’s internal belief about its own freedom, and not its actual freedom, which is relevant to moral responsibility; Frankfurt examples try to show this but it is not clear to me that they succeed, nevertheless this remains intuitively obvious to me). Let’s call this KPAP, for Kip’s Principle of Alternative Possibilities."

Just a quick heads up. This comes pretty close to what Tomis Kapitan tries to lay out in his paper, Doxastic Freedom, and some of his other work (the references tucked too far away in a drawer to find at the moment). So, 'KPAP' is suitable; but KapitanPAP historically antedates KipPAP.

[Kip Werking]

Thanks, James. I didn't want to imply that KPAP was original to me. I would have expected someone else to say something similar. I just didn't know who.

It's interesting to note that, at one point, Cohen says something which may show that KapitanPAP, and not PAP, is the relevant standard:

"Implicitly, people can only ever be responsible if there are alternatives there for them to
reject – these alternatives must be open to agents whom we evaluate for their conduct."

This statement emphasizes that the agents are considering, and rejecting, alternatives. Their doing this is consistent with both determinism, and with their being mistaken about these alternatives being real. Again, it seems to me that it is this introspective decision making which is relevant to moral responsibility, and not alternative possibilities in the real or metaphysical sense.

[Bruce Galbreath]

Re: KapitanPAP, can S's thinking that it was open to S to not do A be the crucial condition necessary for S's being morally responsible for doing A? Suppose I knock a vase off a table, breaking it. The rest of the world might agree that what happened was an accident and not hold me morally responsible. But if I believed that (somehow) I could have avoided breaking the vase, I very well might sincerely believe in my moral responsibility for breaking it. But my belief would be false (on the assumption that, desipite my belief, I could not have avoided breaking it). What KapitanPAP establishes is that if I believe it was open to me to not have done A then (other things being equal) I will believe that I am morally responsible for A. It does not establish that I am in fact genuinely morally responsible, at least so far as I can see.

[Kip Werking]

Bruce,

KapitanPAP is of the form:

MR->once believed

You suggest that it can be true that an agent "once believed" and yet not be MR. But I am just claiming that "once believed" is a necessary, but not sufficient (as your example shows quite well), condition for MR.

Indeed, over the last few days, I've come to think that not even a belief in AP is always necessary for compatibilist-type responsibility. Consider the agent who, somehow, believes he has no APs, of either the compatibilist (conditional) or incompatibilist (metaphysical) kind. Perhaps he has discovered that he is an agent in a Frankfurt example. Suppose that this agent just doesn't care: the path he feels is available is the only path he would want, and he happily owns it. It seems that such an agent could be morally responsible, according to typical compatibilist standards. Frankfurt examples try to show this, but it is not clear to me that they succeed.

Recently, I've been moving in this direction because I favor an internalist conception of moral responsibility. I suspect that defending an internalist conception of MR will both (i) show that MR does not require AP and (ii) show that the compatibilist cannot soften the bite of the manipulation argument by appealing to either agent-relative notions of MR or distinguishing between MR and blameworthiness.