Al Mele's paper: a Garden Discussion

Comments on Al Mele, "Manipulation, Compatibilism, and Moral Responsibility"
Tim O'Connor

Prof. Mele provides us with an admirably clear presentation and analysis of several related arguments for the incompatibility of determinism with freedom and moral responsibility. Mele is undecided on the truth of incompatibilism, whereas I am an incompatibilist who has always found it difficult to feel any temptation towards compatibilism. Despite our differences, I find much to agree with in Mele's critical discussion of certain incompatibilist arguments. I will organize my remarks around four issues. I ask you to forebear my taking the liberty in this context of sometimes stating things in a quick, loose, and sloppy fashion. As one instance of such looseness, I will slide back and forth between talking of freedom and moral responsibility.


1. Mele's historical condition on responsibility

Mele argues that everyone (regardless of one's views on incompatibilism) should recognize an historical condition on freedom and moral responsibility for one's action at a given time. Imagine an agent who involuntarily undergoes a motivational overhaul such that he now performs action A as an expression of values that were "very recently produced in a way that bypassed his capacities for control over his mental life by value engineering." Mele argues that if we flesh out the situation in certain ways (see p.8 for a full and careful statement of the scenario Mele has in mind), then independent of our views concerning compatibilism, and even if all typical compatibilist conditions on freedom are meet, we have good reason to regard the action as nonautonomous, as not having been freely and responsibly performed. And he goes on to draw the conclusion that a plausible compatibilism will recognize an historical dimension to freedom, requiring that the values a free agent expresses in action must have been developed in a way that does not undercut ordinary means of exercising control over one's mental life.

I agree with all of this. And if this is correct, then an argument for incompatibilism that assume that there is no relevant difference between just-prior-to-action manipulation scenarios and non-manipulative deterministic scenarios should be unpersuasive. Still, there might be reason to poke at certain details in Mele's discussion here. Notice that the sufficient condition he provides (p.8) for an agent's being unresponsible for A-ing owing to recent manipulation (NFM) involves his being unable to do otherwise on a compatibilist reading. This is a stronger form of inability than an incompatibilist one (as compatibilist ability is less demanding than incompatibilist ability). Why restrict NFM in this way? Granted, NFM lays down a sufficient condition, not a necessary one, but it still seems a gratuitous weakening of the condition. Suppose manipulation that leaves me strongly desiring to A (contrary to what I would have desired previously) and with no countervailing desire, and so I unhesitatingly A. Surely the compatibilist can agree no less than the incompatibilist that in this case, even if I am able to do otherwise in a compatibilist sense (but not in the plain incompatibilist sense, I act unfreely.

O.k., now suppose that metaphysical freedom in fact requires indeterminism and that the way my action unfolds post-manipulation corresponds to the indeterminist account that best captures the notion of freedom, at least apart from a plausible historical condition on freedom. (I suggest you consult a volume entitled PERSONS AND CAUSES for helpful clues on what this might involve.) On these assumptions, it's tempting to say that while the agent's freedom is diminished, it is not lacking altogether.  (Note on p.7 Mele's passing use of the qualifier, "unfree -- to a significant extent.") For possible significance of this last point, I turn to issue 2.


2. Mele's criticism of Pereboom's manipulation argument to the best explanation for incompatibilism

Pereboom defends a manipulation argument that does not depend on a "no difference" premise, but instead on a "best explanation" premise. The best explanation of the intuition that an agent is not responsible in certain manipulation cases is that "his action results from a deterministic causal process that traces back to factors beyond his control" (quoted on p.19). Mele suggest that we can test this judgment by considering cases of manipulation that are nondeterministic. He then suggests that we imagine cases where the manipulation leaves open exactly two causally possible outcomes: either the A-ing of the original scenario or the incapacitation of the agent. Suppose the agent in facts A-s. Mele notes that we judge the agent not be responsible here, even though the action was not determined. The conclusion drawn is that the no-responsibility intuition in the deterministic scenarios is more likely driven by the presence of manipulation, and not by the deterministic nature of the scenario.

But I wonder whether Mele gives the right kind of indeterministic test-case scenario. Why not consider a case where it is causally open to the agent to undertake two or more different courses of action, with the manipulation consisting in the fact that the motivational package driving the consideration of these options constitutes a complete change-over from what was the case previously. In this scenario, we might be inclined to credit the agent with some (albeit highly diminished) degree of responsibility for the choice he makes. And if this is correct, this might lend support to a more modest conclusion than Pereboom's, viz., that the best explanation of our intuition in his deterministic manipulation scenarios is that it is to some degree undergirded by the presence of both determinism and manipulation.


3. Mele's discussion of an 'original design' argument for incompatibilism

Mele thinks a better incompatibilist argument is of an 'original design' sort. We imagine the value engineering to occur not fairly immediately before the target action, but at the time of the creation of the agent. Imagine Diana in a deterministic universe creating a zygote in a precise way in order to bring about the mature individual Ernie's A-ing thirty years later. Here Diana employs impressive computational powers in relation to her equally impressive knowledge of the laws of nature and the state of the universe just prior to her creating the zygote, so as to deduce what it will take to achive the desired result. First premise: Ernie is not responsible for his action, owing to the value engineering. Now compare this case to an otherwise identical situation involving an agent Bernie's A-ing, where Bernie comes about in the usual way, absent any designer intending any of his actions. The second premise of the argument is that there is no freedom-relevant difference between Ernie and Bernie. From these two premises, the appropriate conclusion is drawn.

Mele is uncertain whether this is a good argument, with his uncertainty centered on the first premise. He suggest that the ideal adjudicators between the conflicting intuitions had by compatibilists and incompatibilists are "reflective agnostics" (such as…himself!). And his hunch is that what will determine the judgments on this premise for such arbiters is the degree to which they have doubts about the compatibility of freedom and moral responsibility with any form of indeterminist agency. The thought is that most reflective agnostics will assume that freedom and responsibility are possible, so if they are not possible under (significant) indeterminism, then they must be under determinism, but then the argument in question must be unsound, and premise 1 is the one to go.

I agree that an antecedent judgment about the prospects for indeterministic free agency will skew the judgments of many reflective agnostics about Mele's premise 1. Here I'll simply note that I am not persuaded that it is reasonable to allow the former judgment to drive the latter one.

Mele further suggests (p.28 and n.22) that verdicts about premise 1 may vary depending on what event it is that Ernie is designed to produce. (Compare cases involving Ernie killing his aunt with ones in which he donates money to charity.) If Mele is right about this, and premise 2 is quite strong, then I suggest that the fact that judgments may waver depending on what kind of event our designer intends may reflect doubts about compatibilism in the face of its consequences. Well, that claim is very quick, but to keep things short, I will simply toss it out and move on.


4. The bearing of Humeanism wrt laws of nature on incompatibilism

Mele invites us to think hard about whether Humeanism about laws of nature and causation should affect the strength of the compatibilist position (29-30). He contends that Humeanism would undermine the thought experiment concerning the origin of Ernie's zygote. Here I think he falls into error, owing to his focusing on only certain of the relevant consequences of Humeanism for freedom. He notes that, since on the Humean view, laws of nature are determined by the totality of (allegedly non-causal) local matters of fact, what the laws of nature are may be partly owing to the way Ernie acts. The laws are not logically prior truths that, together with Diana's action, guarantee his action. All the facts that are fixed by events prior to Ernie's actions do not ensure what he will do, since future facts, including Ernie's action, are needed to fix what the laws of nature are. (If this sounds crazy, that's largely because it is. And I don't have space to do the spelling out that makes it just a little less crazy-sounding.)

So, all that would seem good for Ernie and his freedom. But consider this: if Humeanism is correct, it's not merely the case that the past history (including Diana's designing intentions and action) leading up to Ernie's action at time t leave it open what the laws are, and so also whether Ernie will A at t. It is also open at time t itself (the time of the action) and for a long time thereafter whether, properly speaking, Ernie has even so much as acted at t (let alone acted freely)! Whether the events involving Ernie constitute an action depends on whether there are appropriate causal connections at play. If the future were to unfold in accordance with very different fundamental patterns than those that held sway hitherto (or with no interesting patterns at all) -- and nothing up to and including what occurs at t will ensure that they won't -- then the event that might at the time have looked like an action will turn out not to have been any kind of an action at all.

What conclusion should we draw from this strange, even grotesque state of affairs, on which whether you or me (never mind Ernie) have ever performed so much as a single action is still an open matter, and will remain open long after our deaths? I suggest this: if all this is so, then however the radically contingent, 'loose and separate' events of the future fall out, whether validating certain events as actions or not, then for any one of them, it surely wasn't up to me just then -- something I made to be the case at that very time -- that I did what I did (if indeed I did anything at all). The very dependency of the facts of agency on the entire world history, in the Humean scheme of things, itself supplies grave grounds for doubt about the possibility of freedom and responsibility in anything but the most pathetically deflationary sense. My compatibilist friends, maintain some minimal standards, please! Whatever exactly freedom and responsibility consist in, surely they can't be grounded in so radically external a way as that.

Comments

[Neil]

Just a few words on the historical condition. First, it seems to me that Tim's criticism of NFM is unfair, because it misses the dialectical role of NFM in the paper. If Al's intention were to provide the historical condition governing attributions of responsibility, than NGM would be open to criticism, for the reasons Tim gives, and for others as well. It would be absurdly weak. Consider condition (5), A is the first overt action he performs on the basis of his new values. Why restrict it to the first alone, we might ask? But Al is not trying to engage here with other people who (like him, Tim and me) think there is a historical condition. He is giving a deliberately weak condition so that oppoonents of any historical condition aren't given any gifts. Their counterexamples and arguments need to be stronger: because the condition is so weak it is relatively easy to satisfy.

I said that Tim accepts the historical condition. But I'm not certain just what kind of condition he accepts. The way the dialectic goes, if I'm right, he should accept *at least* NFM. But I'm not certain he does, because in his revised four-case argument, in which manipulation installs the motivational package leading to action, he says 'we might be inclined to credit the agent with some (albeit highly diminished) degree of responsibility'. NFM is designed to rule that out (and that is - incidentally - the right result). So Al can't go this route. (BTW, those who are do not see the force of the historical conditions might like to consider cases in which an agent performs a morally good action, rather than a bad one. My experience is that people are more easily persuaded that the formerly bad agent who has a new motivational structure installed by brainwashing is not praiseworthy for their first actions than they are for the analogous cases of blameworthiness).

Finally, I think Tim is right to worry that though Al might be right in thinking that people's antecedent judgments about whether moral responsibility exist will influence their judgments about whether responsibility is compatible with determinism, the influence may not be rational. We know people are strongly wedded to the idea that moral responsibility exists. A lot of the debate is about whether what we know about the causal structure of the universe and about the nature of actions and reasons is compatible with that belief. It risks question begging or circularity to allow our antecendent commitments to then affect our judgments about the nature of reasons and actions.

[Al Mele]

Thanks, Tim, for getting this started. And, thanks Neil, for following up. I'll exercise self-control, sit back for a few days, and see what rolls in.

[Tim O'Connor]

Neil's point is well-taken that since Al's purpose in advancing NFM is merely to make plausible that there is some historical condition on freedom and responsibility (F&MR), a minimalist statement of the condition is appropriate. I wanted to advance the suggestion that the impact of manipulation on F&MR might not be quite the same in deterministic and indeterministic scenarios, but it was misplaced as a criticism of Al's NFM principle.

Neil is also correct that my remarks on Al's minimalist historical condition left it unclear just what sort of condition I want to endorse. I judge it plausible that freedom and moral responsibility come in degrees (though in a deterministic world, the degree is necessarily zero) and that value engineering manipulation diminishes but does not altogther negate the degree of freedom/MR.

Given these views, I see no way to make very precise claims concerning manipulation in indeterministic scenarios, apart from limit cases such as Al's where the only options left open are to choose A or be incapacitated. So, a general principle would have to be something like: F&MR for an action vary inversely to the significance of the motivational changes induced through manipulative value engineering, the manipulation's temporal proximity to the action, and the intensity of desires or values that are newly introduced.

[Neal]

I think this is more of a request for clarification than anything else.

How does Humeanism about the laws of nature entail that whether a particular event involving me is an action is an objectionably extrinsic matter? (I take it that this is the objection.)

Is the idea that Humeanism about laws entails Humeanism about causation, and Humeanism about causation makes facts about what causes what rely on objectionably extrinsic matters? If that's the argument, I'm not sure why we should accept either entailment.

Regarding the first entailment -- that Humeanism about laws entails Humeanism about causation. I take it that Humeanism about laws merely states that the laws of nature are a matter of regularities, rather than a matter of necessary connections. Couldn't you think that laws are a matter of regularities even while you think that there is an irreducible causal connection between particular events? Unless we suppose that causation itself must be governed by the laws (and don't singularists about causation deny this?), then it seems one could be a Humean about laws without being a Humean about causation.

Second, about the claim that Humeanism about causation entails that whether an event counts as an action depend on objectionably extrinsic matters. To be sure, I can see how, say, a counterfactual theory of causation makes causation partly an extrinsic matter. But I can't yet see why it's extrinsicness is objectionable when it comes to thinking about actions.

[Kip Werking]

Some scattered remarks:

1. I loved Mele’s paper. I’m just finishing Autonomous Agents, which I’ve been reading off and on for over a year (better late than never), and Al’s paper compliments that book well, helps me better appreciate Cuyper’s critique, and makes me hungry to read Free Will and Luck. Autonomous Agents has also given a greater appreciation for action theory, which I usually find quite dull.

2. I think Hume’s problem about causation/correlation is entirely orthogonal to the problems in the free will debate (at least, those problems that interest me).

3. Anyone who thinks freedom and moral responsibility, in the philosophical sense traditionally at stake in this dispute, admits of degrees, needs to explain, I think, why these things were traditionally regarded as all or nothing. As Gary Watson wrote “Bernard Williams also notes that, on the traditional conception, free will is supposed to be an all or nothing affair.” This is a point in favor of those who use all or nothing concepts, such as Galen Strawson, Pereboom, and (some) agent causation theorists. Free will, on these views, is something like a soul. You either have it or you don’t.

4. Mele’s treatment of Pereboom’s four case argument does not strike me as very charitable—and this may explain why Pereboom seems to have retreated from his earlier way of articulating the argument. Part of the problem seems to be that Pereboom drew his book along the unfortunate but traditional lines of compatibilism and incompatibilism, and still insists that some form of agent causation would be coherent.

The question is how to read the phrase:

“The best explanation for the intuition that Plum is not morally responsible in the first three cases is that his action results from a deterministic causal process that traces back to factors beyond his control.”

Here, Mele emphasizes (underlines) “[t]he best explanation”. I think he also emphasizes, but does underline, something that was not intended to be essential to the best explanation: that the process is “deterministic.” Without this tacit emphasis, one could charitably read Pereboom as saying that the best explanation is that the action results from a process, deterministic or indeterministic (but which happens to be deterministic here), that traces back beyond Plum’s control—in other words, that Plum is just (in Smilansky’s phrase) “the unfolding of the given.”

A charitable reader could pursue this possibility by considering the four-case argument in the spirit of Pereboom’s entire book Living Without Free Will and asking: “if Pereboom thinks that determinism is the best explanation for why Plum is not morally responsible, does he think that indeterminism could somehow help?” The answer is that, although Pereboom does think that some sort of agent causation might help, he also finds more ordinary indeterminism to be just as threatening as determinism. And so a charitable reader might read the emphasis out of “deterministic” in Pereboom’s original sentence.

5. Not only does Mele criticize Pereboom’s “best explanation”, he also distinguishes it from his Zygote Argument. I love Mele’s Zygote Argument, and I’m glad his sympathy for it makes him nervous about compatibilism. But the differences between it and Pereboom’s argument are not so great.

Mele justifies this distinction by appealing to the difference between Pereboom’s “best explanation” principle and his own “no-difference” principle. But Pereboom’s four case argument *also* has a no difference principle: there is no relevant difference between the cases and so Plum is not morally responsible in any of them. Pereboom even writes, in Determinism Al Dente, “If the compatibilist wishes to argue that Mr. Green is morally responsible under these circumstances, he must point out a morally relevant feature present in Case 3 but not in the first two cases, and such a difference is difficult to detect” and the same point applies to Case 4. So the essence of Pereboom’s argument relies upon just the sort of “no difference” principle Mele says it lacks.

6. Indeed, when Mele implies that the Zygote Argument is original to him, I feel like Watson reviewing Van Inwagen’s Essay on Free Will, where he noted that the Consequent Argument is not so original at all but rather has a long and distinguished history. Similarly, the Zygote Argument is the same, in its essentials, to Pereboom’s four case argument (as I argued above) and I think the idea goes all the way back to the Hobbes and Bramhall debate.

7. The last point I would make is a plea for consistency. It will not due, I think, to remove the Zygote Argument from the rest of one’s view, as some sort of strange anomaly that can be dealt with later. The Zygote Argument works, at least partially, and compatibilist views need to be revised or abandoned accordingly.

Consider the following. If the Zygote Argument works, *how* does it work? It works because it involves the manipulation of things that Mele otherwise argued did not need to be under the agent’s control. As Mele eloquently argues, some agents can autonomously possess sheddable brainwashed-in pro-attitudes but no agent can autonomously possess unsheddable brainwashed-in pro-attitudes. But the critic immediately wonders (and I think Cuypers was trying to make this point): why does Mele limit himself to only considering those unsheddable pro-attitudes as threatening which conflict with prior existing pro-attitudes? Why, in other words, is it no threat for human autonomy that our original/initial evaluative perspective necessarily bypasses any capacities for rational control (which presupposes an existing evaluative framework; you can’t evaluate anything if you have no sense of what is good or bad, right or wrong), but it *is* a threat that some later unsheddable pro-attitude could conflict with this original set? As I noted in my post on Cuyper’s critique, David Zimmerman calls this “the puzzle of naturalized self-creation in real time.”

As I wrote in another thread:
“Suppose that X is, like myself and yourself, born with certain largely species-typical pro-attitudes or values, like preferences for family members over strangers, and preferences for eating food over starving to death. Now suppose that X is subject to a radical curing-like intervention which, instead of making X safe for society (and more similar to the rest of us), reverses these species typical pro-attitudes or values, making him prefer strangers to family members and prefer starving to death over eating food (and less similar to the rest of us).
It is a strange example, to be sure. But you may see my point by considering this question: what autonomy has this curing-like intervention taken away from X? Sure, he did not freely choose to prefer strangers over family members, and he did not freely choose to prefer starving to death over eating food. But he *also* never chose to prefer family members of strangers, or eating food to starving to death. He was *thrown into this world* with those original preferences, as part of his genetic makeup, and if he is now a slave to the new preferences, he would also have been just as much a slave to the old preferences. In at least one important sense, he seems just as autonomous, with respect to these preferences, as he has always been: not very.”
It is *this* asymmetry that allows Mele to defend a compatibilist view on autonomy while also finding the Zygote Argument troubling. In fact, if one was going to be consistent, it makes no big difference whether one’s evaluative perspective bypasses one’s faculties for rational control at the beginning of one’s creation, or later in life, overwriting whatever original perspective one had. They are both equally out of the agent’s control, and s/he is equally non-autonomous with respect to both of them.

If this is right, then I think it is inconsistent to both (i) leave this exception for initial pro-attitudes to bypass one’s faculties for rational control and (ii) find the Zygote Argument troubling. The Zygote Argument works because there is something erroneous about leaving that exception open. There is something deep down, in human nature, however irrational it is, that still thirsts for rational control not just over later pro-attitudes given original ones, but over *all* pro-attitudes—and this amounts to being causa sui.

8. Mele writes that he might find the Zygote Argument less persuasive if he was less optimistic about libertarianism. But although one can try to fashion a libertarianism that provides as much control as compatibilism, it cannot provide more. Gary Watson eloquently makes this point in his Soft Libertarianism and Hard Compatibilism, which considered the Zygote Argument in its essentials back in 1999. Of course, Mele notes that such a libertarianism might avoid the Consequence Argument, and gives agents multiple futures, but neither of these would enhance their autonomy.

Mele wrote in Autonomous Agents that his modest libertarianism “provides for an agent’s having more than one physically possible future in a way that turns, essentially, on what goes on in him” but this is not the sort of power that libertarians want. The indeterminism may take place inside “the agent”, or at least hir brain, but that agent does not have control over how-it-will-turn-out (as Dennett notes, it is the essence of indeterminism for nothing to control how it will turn out). Yet this is the sort of control that libertarians want and need. In summary, perhaps Mele should, in the face of the Zygote Argument, instead place his hope, not on libertarianism, but, like Watson, on a “hard” or revisionist compatibilism.

Finally, I am wondering, like Fischer wondered in his review of Mele’s Autonomous Agents, why Mele is not more sympathetic to the denial of autonomy, or at least agnostic with respect to it. And even if I doubt that the Zygote Argument is as original as Mele suggests, I am glad he brought it to our attention. Like him, I hope it gets the attention it deserves.

[Tim]

Neal,

Right, Humeanism about laws doesn't entail Humeanism about causation. Remember, we're doing things in shorthand here! I was taking it as given that most Humeans about the first are also Humeans about the second. Al specifically has in mind a Lewisian view. And on that view, what causes what depends on what the laws of nature are, since these are significant determinants of the truth of counterfactuals, which in turn help to determine the truth of causal claims.

I take it to be objectionable -- indeed, utterly bizarre -- that whether or not I am now performing (and so whether I am now freely performing) the action of typing on my keyboard, or performing any action at all, depends on what has happened in the far distant past and what will happen in the far distant future. Humeans may persaude themselves that there are some theoretical benefits acrruing to their position overall, but surely they cannot plausibly contend that this consequence is not a mark against the view.

[Eddy Nahmias]

I hope to offer more complete remarks later (once I'm finished with my paper for Manuel's conference), but I am hoping for more clarification from Tim (or Al) about Tim's reductio of the Humean compatibilist view. First, isn't a Humean able to say that we can make statements about the laws (and do science and make predictions using our understanding of the laws, etc.) with great confidence, just not absolute certainty. Perhaps we have enough evidence to say with great confidence that certain regularities we think are laws of physics will "remain" laws even at the end of time, and that also likely holds for whatever regularities are required to say that agents perform actions. (Perhaps the complexity of some human deliberative actions--dare I say, free actions--is such that we cannot be highly confident about predictions regarding, or laws governing, such behaviors, but that's OK--that's why we might be inclined to say they are free even if it turns out the laws governing them are deterministic.)

Also, isn't the libertarian in an analogous situation to the one Tim suggests the Humean compatibilist is in. The libertarian has to say that, assuming determinism is possible, we do not currently know whether any of us has free will or is morally responsible or whether any human has ever performed a single free action. Indeed, even if we find out indeterminism is true, that wouldn't be enough to secure our freedom for a libertarian, since we would also have to find out the indeterminism happens in the right place (presumably, somewhere in the brain), or the right emergent properties have to exist (for an agent causationist like Tim) or the right agent-causal relations have to exist (for all agent causationists), etc. Perhaps a "metaphysical flip-flopper" like van Inwagen can avoid this implication, but most libertarians, I suspect, are not flip-floppers and their position makes our free will dependent on what physicists eventually tell us about the laws governing our universe (unless, as a recent post argued, that's not something physicists could tell us).

Well, now that I'm rambling, and just in case I don't get back to it, here's a very brief response to Mele: I believe the compatiblist should require historical conditions and should reject the second premise of Manipulation arguments (including Mele's Zygote argument)--that's the no-difference principle. The difference is that manipulation cuts off possibilities that determinism does not. Since a manipulator has ends of his own he is seeking to ensure and is using the manipulated agent to bring about (and he is powerful enough to make her do it), then were circumstances different such that the agent would have good reasons not to act to secure those ends (and if she is reasons-responsive such that she might act on those reasons), the manipulator would force her not to act on those reasons (by changing her mental states or directly intevening). Determinism does not intervene in her reasons-responsive capacities in this way. So, despite attempts to present manipulation as working through reasons-responsive capacities, the right counterfactuals are not in place to say the agent has those capacities.

This is brief and underdeveloped (and may be more difficult to apply to the zygote argument than other manipulation arguments). For the best defense of this view that I've seen, see Gideon Yaffe's "Indoctrination, Coercion, and Freedom of Will" (PPR).

[Neil]

Eddy,

Suppose that a Laplacean demon wanted to manipulate an agent into stealing from the poor box. He inserts into the agent all the relevant desires and beliefs that will motivate that choice. Now, why can't the demon make the agent as reasons-responsive and reactive as you like? He could set the mental machinery so that *were there to be* reasons to respond otherwise, the agent would recognize them, and respond accordingly. But because (by hypothesis) the demon knows what the state of affairs prevailing at the relevant time will be, he knows that there *won't* be such reasons. He has constructed an agent that meets Fischer's criteria - indeed, if you like, stronger criteria - for reasons-responsiveness, but knows just how the agent will act. There's no need for the demon to intervene once he sets things going.

[Tim O'Connor]

Neal,

You asked why the Humean can't justifiably have confidence now in what the laws are (roughly), absent direct confirmation that future events conform to the patterns our actual evidence indicates. I myself doubt that they can, consistent with their Humeanism, but I wasn't relying on that position in my comment, as my point was metaphysical, not epistemological. What I claimed to be deeply implausible is that I freely determine what I do over a time interval t1-t2 if the truth of the statement that I performed any action at all over time t1-t2 strictly depends on what occurs in the far distant future.

Think of it this way: Assume you're a Humean about laws and causation, along the lines of Lewis's picture. A sequence of events unfold in Sue, events that you find natural to describe as Sue's briefly considering whether to insult Harry, forming the intention at time t to do so, and acting on that intention. These events are not necessitated by anything, including the entire past, although they perfectly conform to non-statistical ('deterministic') regularities suggested by all our evidence to date. And nothing of what we observe gives a compatibilist any qualms about Sue's freedom.

Now you reflect on the implications of your Humeanism. You take yourself to have good reason to believe that future patterns will conform (and past patterns have conformed) to those directly observed, though such future events are in no way constrained or necessitated by those of the past. So you have good reason to believe that Sue has performed a deliberate action satisfying your favored compatibilist account of freedom. But then you consider that everything that has occurred to date leaves it entirely open, metaphysically, what the future events will be. In those possible worlds exactly like ours (in the relevant sense) up to the events shortly after time t, any number of futures unfold thereafter, such that in 'a great many' of them, it would be false to say that Sue performed any action at all around time t, or any other time. You then reflect on the fact that the truth of an action statement strictly requires the future's unfolding in a certain way -- a way that is totally unconstrained by what is past and that is clearly outside the control of the agent in question. Whether Sue, in particular, really did deliberate, decide, and act at t as you take her to have done does not supervene on the events occurring at the general time in question (only some of which you observed, but *that* is not what drives my point). It depends further on the nature of a huge number of future events that are 'loose and separate' from what has already occurred.

I suggest that, reflecting in this way, you should begin to have grave doubts about whether it is really true that Sue freely and responsibly made it the case, at the time in question, that she decided and acted as she did. Can you feel the grip of this doubt at all? If so, what can the Humean say to alleviate it?

[Al Mele]

Because Neil and Tim have already sorted some things out, there’s no need for me to comment on those things. Just a few remarks for now.

1. Manipulation

Tim writes: “Why not consider a case where it is causally open to the agent to undertake two or more different courses of action, with the manipulation consisting in the fact that the motivational package driving the consideration of these options constitutes a complete change-over from what was the case previously. In this scenario, we might be inclined to credit the agent with some (albeit highly diminished) degree of responsibility for the choice he makes.”

I suggest that readers consider a version of my Beth story on pp. 9-10. Imagine that the setting is indeterministic and that the manipulators can’t do enough to determine Beth’s choice of a weapon. It’s open whether she’ll select a knife or a gun (both of which are in George’s house); but it’s not open whether she’ll kill George. Beth selects a gun.

I wonder how many of you think that Beth has some moral responsibility for one or more of the following: killing George; killing George with a gun; selecting a gun as her murder weapon. Obviously, the view that a person has some moral responsibility for A-ing only if more than one course of action is genuinely open to her (i.e., consistent with past + laws) at the relevant time does not commit one to judging that Beth has any moral responsibility for any of these actions. I’m curious what theoretical work this openness might do for you beyond being a necessary condition for being morally responsible for an action.

Neil writes: “My experience is that people are more easily persuaded that the formerly bad agent who has a new motivational structure installed by brainwashing is not praiseworthy for their first actions than they are for the analogous cases of blameworthiness.”

This is interesting. I tell stories of both kinds in *Autonomous Agents* and in *Free Will and Luck*. My thought was that my good-to-bad stories would be more effective for some readers than my bad-to-good stories, that the reverse would be true of some other readers, and that yet others would find stories of both kinds persuasive. I’m curious what readers think about Neil’s claim. In the target article, there is a mediocre-to-good manipulation story (Paul’s story, pp. 6-7). Are there readers who find this story persuasive but not my good-to-bad Beth story? Are there readers who find the latter story persuasive but not the former?

2. Reflective agnostics & the zygote argument

Tim writes: “I agree that an antecedent judgment about the prospects for indeterministic free agency will skew the judgments of many reflective agnostics about Mele's premise 1. Here I'll simply note that I am not persuaded that it is reasonable to allow the former judgment to drive the latter one.” Neil echoes this: “I think Tim is right to worry that though Al might be right in thinking that people's antecedent judgments about whether moral responsibility exist will influence their judgments about whether responsibility is compatible with determinism, the influence may not be rational.”

I don’t say that this influence is rational or reasonable. My concern in this connection is descriptive rather than normative: it’s with what intuitions readers of different kinds are likely to have and with some likely intuition-influencers. For relevant discussion, see my *Free Will and Luck*, pp. 205-6.

Eddy wrote: “Since a manipulator has ends of his own he is seeking to ensure and is using the manipulated agent to bring about (and he is powerful enough to make her do it), then were circumstances different such that the agent would have good reasons not to act to secure those ends (and if she is reasons-responsive such that she might act on those reasons), the manipulator would force her not to act on those reasons (by changing her mental states or directly intervening).”

I was planning to reply to this, but Neil pre-empted me. Here’s another way to see the point. Just imagine that Diana entirely loses interest in Ernie after she creates him – or that she dies right after she creates him.

Kip writes: “I love Mele’s Zygote Argument, and I’m glad his sympathy for it makes him nervous about compatibilism. But the differences between it and Pereboom’s argument are not so great.”

As I explained in the paper, the zygote argument and Derk’s 4-case argument have very different logical forms. By my lights, that makes a great difference.

In another connections, my expression “original design argument” misled Kip, but it was my fault. My “original” was meant to modify “design,” not “argument.” I should have written “original-design argument.”

3. Degrees

Kip writes: “Anyone who thinks freedom and moral responsibility, in the philosophical sense traditionally at stake in this dispute, admits of degrees, needs to explain, I think, why these things were traditionally regarded as all or nothing.”

Is “moral responsibility . . . traditionally regarded as all or nothing”? In any case, for an argument that it shouldn’t be so regarded, see *Free Will and Luck*, ch. 5.

P.S. I see that Tim has just posted something. I'll read it now.

[Kip Werking]

Thanks for your response, Al.

1. Wouldn't the praise/blame side effect suggest that people would be more likely to still hold good-to-bad agents as blameworthy, but less likely to hold bad-to-good agents as praiseworthy? From J. Doris, et al.'s Strawson Variations:

"Most subjects in the harm condition said that the chairman deserved blame, but very few subjects in the help condition said that the chairman deserved praise. Subsequent research has found similar effects using other vignettes (Hauser et al. unpublished data; Knobe 2003a, forthcoming; Sverdlik forthcoming)."

2. You write that "As I explained in the paper, the zygote argument and Derk’s 4-case argument have very different logical forms." I've read your paper and I don't think you explained how their structures are so different (at least not in much detail). As I mentioned in my previous post, your analysis does not strike me as very charitable.

In particular, you seem to have zeroed in one aspect of Pereboom's argument: the best explanation premise. But it is not clear to me that this premise is crucial to his argument. You have not addressed what I think is an especially moving consideration here: that even if Pereboom's argument contains a best explanation premise, it *also* contains a no difference principle.

Pereboom's elaboration of this no difference principle is even clearer in LWFW:

"Given that we must deny moral responsibility to Plum in Case 3, what reason do we have for holding him morally responsible in this more ordinary case? There would appear to be *no differences* between Case 3 and Case 4 that could support the claim that Plum is not morally responsible in Case 3 but is responsible in Case 4." (p. 115).

This quote appears in the paragraph *immediately preceding* the paragraph which begins "The best explanation..." It strikes me as uncharitable to only focus upon the latter paragraph, at the expense of the former, and to characterize Pereboom's argument as so distinct from the Zygote Argument (with its no-difference principle), when Pereboom's argument plainly invokes an identical or near-identical no-difference principle.

3. I understand now what you mean by "original design argument" (original-design argument). That makes much more sense, and I'm not sure the fault was entirely (!) yours for my misunderstanding.

4. The original quote about degree involved "free will" and not "freedom" or "moral responsibility". Your response inspires me to ask: do you think that one would owe others an explanation for using a degrees-type concept of free will, but would not owe an explanation for using a degrees-type concept of moral responsibility? I think that distinction is on the right track...

[Paul Torek]

Some of the stories on manipulation or brainwashing raise interesting questions realted to personal identity. Unfortunately, those questions are answered only implicitly and, I would argue, wrongly.

These cases involve psychological transformations so drastic and sudden that they are nearly textbook cases of the sort designed to contrast 'psychological' vs 'physical' criteria of personal identity. But rather than taking sides directly in that battle, I propose that as Gardeners we should focus on some of the moral and metaphysical questions of responsibility and agency. In other words let us pick and choose among the concerns that drive debates on personal identity. We may not have to answer whether the person who emerges from Beth's brainwashing is Beth, full stop. (I certainly hope not, because I doubt it has one univocal answer.) We can instead decide that this person does not continue the thread of agency that began with the Beth we all knew and loved. Similarly, we can deny that this emerging person's misdeeds tarnish the character of the Beth we all knew and loved. That character is gone, though something may remain of her. We might still dearly hope, for example, out of our love for Beth, that Beth's cancer is cured soon, even though in many other important ways we are inclined to say that Beth is gone.

If you are willing to consider different personal-identity-related concerns separately, I think you will find that concerns about responsibility and agency favor the broadly 'psychological' side of the personal identity debate. At least, that is my finding. Therefore I must object to any question which presupposes that we can identify a single agent both before and after Beth's brainwashing -- about which single agent we then proceed to ask, "responsible?" Rather, we should consider the two agents, the one before and the one after. And in that case, it becomes pretty clear who is responsible for what.

[Al Mele]

I’m just back from a pleasant trip to the University of Maryland. It seems that it’s my turn again to say something.

1. Folk responses

Kip writes: “Wouldn't the praise/blame side effect suggest that people would be more likely to still hold good-to-bad agents as blameworthy, but less likely to hold bad-to-good agents as praiseworthy?”

Not really. One needs to be careful here. The asymmetry we find in folk reactions to *side-effect actions* is not matched by a similar asymmetry in responses to stories in which agents are *trying* to do something good or bad and luckily succeed in their attempts. An agent who luckily succeeds in doing something very bad is given a high blame rating for doing it, and an agent who just as luckily succeeds in doing something very good is given a high praise rating for doing it. The difference between side-effect actions and actions one is trying to perform makes a big difference here. The side-effect results leave it wide open whether Neil Levy’s idea “that people are more easily persuaded that the formerly bad agent who has a new motivational structure installed by brainwashing is not praiseworthy for their first actions than they are for the analogous cases of blameworthiness” is on target or off target. Bear in mind that the actions featured in these manipulation stories are actions that the agent is trying to perform and not side-effect actions.

2. 4-case argument again

Kip, regarding your remarks on the 4-case argument, see the last paragraph of section 2 of the target article. Also, notice the differences between straight manipulation arguments and the zygote argument.

3. Degrees

Consider the expressions “S was morally responsible for A-ing” and “S freely A-ed.” Philosophers do seem happier, on the whole, with degrees in the first connection than in the second. Working out an attractive view of degrees in the second connection would, I think, be an interesting project. (Of course, we do use the expression “degrees of freedom” in several connections. There are, for example, degrees of social and political freedom. But the present issue is about acting freely.)

4. Personal identity

Here’s a quotation from my *Autonomous Agents* (Oxford UP, 1995, p. 175):

The "good" [Chuck] case and others developed in this chapter might have prompted worries about personal identity. Is the transformed [Chuck] the same person as the pre-transformation [Chuck]? Is Beth, after becoming a [Chuck] "twin," the same person as the earlier Beth? This is not the place to advance a theory of personal identity. But, surely, the pre- and post-transformation agents have much in common? [Chuck] just before his transformation (t-[Chuck]) is much more similar, on the whole, to [Chuck] just after it (t*-[Chuck]) than he is to neonate [Chuck] or toddler [Chuck]. Still, t-[Chuck] is the same person as the neonate and toddler [Chucks], in a familiar "personal identity" sense of "same person." So what is to prevent him from being the same person, in the same sense, as t*-[Chuck]? It is worth noting, further, that t-[Chuck] and t*-[Chuck] may be strongly psychologically connected, in Parfit's sense (1984, p. 206). They may be such that the number of direct psychological connections between them "is at least half the number that hold, over every day, in the lives of nearly every actual person."

Concerning manipulation stories of the kind I spin in the target articles, Paul Torek writes: “These cases involve psychological transformations so drastic and sudden that they are nearly textbook cases of the sort designed to contrast 'psychological' vs 'physical' criteria of personal identity.”

I’m unsure why Paul says this. And notice that it’s strong *psychological* connections that I emphasize.

Paul also writes: “I must object to any question which presupposes that we can identify a single agent both before and after Beth's brainwashing -- about which single agent we then proceed to ask, ‘responsible?’ Rather, we should consider the two agents, the one before and the one after. And in that case, it becomes pretty clear who is responsible for what.”

Question: “Two agents” in what sense of “two agents”? Consider me now (Al-n) and me 24 hours ago (Al-y). (I add that nothing special happened to me in the last 24 hours.) Do Al-n and Al-y count as two agents in the sense at issue? And, if so, (1) are these two agents such that one of them can be responsible for things that the other one is not responsible for and (2) how long does each of these agents exist?

[Paul Torek]

It's not clear that the severe brainwashing cases meet the criterion of "at least half the [connections] that hold, over every day, in the lives of nearly every actual person." Suppose all the (quasi-)memories are intact. Still, there are at least two other categories of psychological connection: the relation between intention and act, and the continued holding of personality traits.

The personality change is quite drastic. From the point of view of either t-[Chuck] or t*-[Chuck], the importance of the changed personality traits probably significantly outweighs that of the continued ones.

And the change wreaks havoc on the relation between intention and action. Given his new motivations, Chuck will probably question every intention. Not just the planned kidnappings and torture, but even a planned trip to the grocery store: "forget the darn groceries, I need to see a psychologist to figure out what has been wrong with me."

Out of the three categories of connectedness, Chuck fares quite badly on two. But wait, there's ... less. I don't think we should treat all three categories as one lump sum score of connectedness. On this point, I feel that Parfit is not Parfitian enough.

Being a "reductionist" in Parfit's sense about personal identity ought to make one ask whether "what matters" is different for different concerns. For example, what matters in my concern to end a physical pain might be different from what matters in my concern to make a difference in the world. The latter concern involves agency in a way the former does not. Suppose for the sake of argument that there is a personal-identity-like relation that gives special reason for concern about future pains, and a personal-identity-like relation that gives special reason for concern about future agency. Parfit seems to assume that these must be the same relation. I suggest we leave open the possibility that they are not.

Of the three aforementioned kinds of connectedness, the relation between intention and act is very directly and obviously tied to agency. Personality traits and memory are less so, although obviously, for example, you have to remember an earlier intention if you are to act on it later. If we are concerned about agency, then, the intention/act connection weighs very heavily.

What kills off an agent qua agent? A psychological disruption that wrecks the foundations of the intention/act connection, such that few or none of the intentions can be taken seriously by the ensuing actor. That level of disruption clearly does happen to Chuck, and clearly does not happen to everyday Al.

And what kind of event makes it indeterminate whether an agent survives? Psychological disruptions which derail a very large number of intention/act and closely related connections - perhaps, following Parfit, more than half - but still leave a significant number more or less intact.

[Al Mele]

A careful attempt to resist my manipulation stories on grounds that feature considerations of personal identity might be very interesting. But one has to be careful not to ignore pertinent details of the stories. For example, my characters endorse their new values and are entirely satisfied with them. Consider, in this connection, the following by Paul Torek: “the change wreaks havoc on the relation between intention and action. Given his new motivations, Chuck will probably question every intention. Not just the planned kidnappings and torture, but even a planned trip to the grocery store: ‘forget the darn groceries, I need to see a psychologist to figure out what has been wrong with me.’” Not so. In Chuck’s opinion, nothing is wrong with him.

Notice also that there are variants of my stories in which the radical value-change is undone in, say, a half hour – just after the featured dirty (or good) deed is done. If we say, that the original agent (Beth 1, say) went out of existence owing to the manipulation and a new agent (Beth 2) popped into existence, should we say that Beth 1 popped back into existence after the change is undone, or should we say that we have yet another new agent (Beth 3) – one who is a heck of a lot like Beth 1?

[Paul Torek]

Al,

In Chuck's new opinion, nothing is now wrong with him, but something used to be. And that's all I have him saying. And in light of his recent unexpected transformation, isn't it reasonable for him to worry that a second, contrary transformation might happen at any time - and to see a psychologist in order to assess and to head off such a possibility?

The question about the double-transformation, restoring the qualities of the original personality, is very interesting! A lot depends on the causal relationships between the early and late segments of the person's (persons') life (lives). Specifically, when we focus on agency, whether a normal or at least reliable causal relation between earlier intention and later act could be traced through such a double-transformation. It's not obvious, but you might be able to tell the story in such a way that a reliable causal relation does exist.

Although Parfit has implied that such causal relationships don't matter, most of the commentators (and I add myself to this bunch) think they do. On this view, whether there is a surviving (temporarily knocked out of commission) agent depends on the details of storage and restoration of personality.

[Kip Werking]

Al,

Some points.

1. I see the difference between the side-effect and trying cases. But I am wondering what basis you have for saying:

"An agent who luckily succeeds in doing something very bad is given a high blame rating for doing it, and an agent who just as luckily succeeds in doing something very good is given a high praise rating for doing it."

Are people really just as quick to praise as to blame? That seems counter-intuitive to me. Here is one relevant excerpt from the paper I wrote last summer:

"Another cognitive bias which may aggravate the FAE is the asymmetrical attribution of blame. Research shows that observers of immoral actions are more likely to infer correspondence between the actions and the actor’s character (Reeder & Spores 1983). In contrast, observers of moral actions are more sensitive to the situational constraints which may have caused this behavior. These results are consistent with other research showing that people identify words implying social costs more readily than words implying poor skill or positive qualities (Ybarra et al. 2001). The result is that the FAE may be especially insidious in the context of observing immoral, rather than moral, behavior."

This was the research that inspired my earlier comment (I had mistaken the side-effect studies for being essentially similar). My paper goes on to summarize some research on demonization processes and the eagerness with which people blame others. Are there similar “angelization” processes? Are people ever eager to praise others (hero worship, perhaps?)? I am skeptical, but perhaps I am wrong.

2. My discussion of the four case argument will be limited by my not-having-read your 2005 and 2006 publications, where you discuss the three manipulation cases in depth. I will say this: there seems to be some ambiguity about the four case argument. Earlier, I had assumed that you were discussing Pereboom’s four case argument, as a whole. But looking over your paper, I see that you repeatedly refer to “Pereboom’s argument to the best explanation.”

Of course, one can divorce whether-the-argument-works from the best-explanation-for-why-it-works. Note that Pereboom’s goal in that chapter of LWFW is to undermine compatibilism. In that sense, he does not *need* the argument to the best explanation. The four case argument can work even if the best explanation offered is mistaken (or just poorly worded). As I described in my previous post, Pereboom’s four case argument uses a *no-difference* principle and, in a later paragraph, the argument to the best explanation uses a best-explanation principle.

So my question would be: do you acknowledge the no-difference argument in Pereboom’s work? Were you just intending to target the best-explanation argument, as separate from the best-explanation argument? Or have these two been, to some degree, conflated?

3. Finally, I would say: earlier I had tried (at length, and inelegantly) to argue that the Zygote Argument cannot be kept separate from the rest of one’s view on autonomy/freedom. I would like to repeat that appeal here. The Zygote Argument, as I understand it, is like the smoke that betrays the fire, or the canary in the coal mine. If the Zygote Argument works, and contradicts one’s compatibilist theory of autonomy/freedom, then one must have made a misstep somewhere along the way while building that theory.

Earlier I suggested where the misstep took place: your view allows for agents to autonomous possess sheddable (and unsheddable?) proattitudes that bypassed the agent’s not-yet-existent capacities for rational control just because the agent was built or constructed with those pro-attitudes either existing, or latent, within it. Although this may not be entirely wrong (wrong in every sense of freedom/autonomy), it does not strike me as entirely right. The Zygote Argument seems to support this suspicion. My question would be: if (as you sympathize) there is something right about the Zygote Argument, and if this shows that something must be wrong with your compatibilist view on autonomy, what would that misstep be?

[Derk Pereboom]

As Seth Shabo pointed out a few threads ago, the four-case argument against compatibilism has three stages. 1. The first two cases, in particular, are intended to pose a direct challenge to the sufficiency of the various compatibilist conditions. 2. I then argue that it will not be possible to draw a principled line anywhere among the four cases, largely because all of the prominent compatibilist conditions are satisfied in each – this is the “no difference” part of the argument. 3. Finally, I contend that the best explanation for why Plum isn't responsible in these four cases is that he is causally determined by factors beyond his control in each. The no-difference part of the argument and the best explanation part are linked in this way: it’s partly because there’s no difference in the cases that can explain why one would judge Plum non-responsible in Cases 1 and 2, but responsible in Case 4, that we are driven to the conclusion that Plum is not responsible in Cases 1 and 2 because he is causally determined by factors beyond his control. This verdict then generalizes to Case 4.

Al suggests that there is a competing and better explanation for Plum’s non-responsibility in the earlier cases, viz., that Plum is manipulated. This claim can be supported by the fact that our intuition in certain indeterministic manipulation cases is that Plum is not responsible, together with the fact that compatibilists have intuitions of responsibility in certain deterministic cases. What’s common between the deterministic and indeterministic cases is the manipulation, and the suggestion that causal determination is not the best explanation is supported by the compatibilist intuitions about responsibility. And so the better explanation for non-responsibility for these cases is not causal determination, but manipulation.

(1) I argue in the book that the best explanation in the four-case argument for non-responsibility in Cases 1 and 2 cannot be manipulation by other agents, since if in those cases the manipulators are replaced by machines that randomly form in space and that have the same effect on Plum as the manipulators do, our intuition that Plum is not morally responsible remains. It may be that compatibilists have intuitions of responsibility in some deterministic examples. That counts against my assessment, and it does suggest that we should look for an explanation for non-responsibility other than causal determination But I contend that manipulation by other agents can’t explain non-responsibility in Cases 1 and 2, and that we are forced to conclude that it’s causal determination after all. This is a key part of the argument, and to overturn it, a stronger case needs to be made that manipulation can be the best explanation for non-responsibility in these examples.

(2) The fact that we can substitute an indeterministic for a deterministic case and still get non-responsibility (I in effect endorse this in ch. 2) does not show that determinism isn’t the best explanation for non-responsibility in the deterministic cases. Compare: most of us think that the best explanation for the WTC collapse is that it was hit by a plane. Someone might say: tweak the case a little – you can get the same collapse if you substitute a rocket or an asteroid. So the best explanation for why the WTC collapsed isn’t that it was hit by a plane. This is clearly incorrect – these reflections shouldn’t prompt us to withdraw the claim that the plane explanation is best. There is a more general and perhaps deeper explanation that these tweakings point to – that the WTC collapsed because it was hit in a certain way by an object with such-and-such general characteristics. But this being a more general and deeper explanation doesn’t show that the plane explanation isn’t best – at least in the assumed original context of explanation. Note that the plane explanation doesn’t compete with the more general and deeper explanation – they are explanations at different levels.

By analogy, in Case 2 Plum’s decision’s being causally determined by factors beyond his control best explains why he is not morally responsible, and this is compatible with the existence of a more general and deeper explanation for his non-responsibility common to the deterministic and indeterministic manipulation cases – viz., that Plum is not the ultimate source of his conduct, or that he lacks the sort of control required for moral responsibility.

[Neil]

Derk writes

I contend that the best explanation for why Plum isn't responsible in these four cases is that he is causally determined by factors beyond his control in each

But aren't you - by virtue of your endorsement of the luck argument against event-causal libertarianism - committed to a different claim; viz. that the reason why Plum isn't responsible in these four cases is that he is caused to act as he does by factors beyond his control (whether these factors are deterministic or not)? After all, you don't think that if Plum had acted as he did as a consequence of some genuinely indeterministic event-causal process that he would act responsibly. In that case, there seems to be the basis for a principled response to Al: you can agree that adding indeterminism to the cases doesn't yield responsibility. But you can resist the claim that it is manipulation that is the explanation. And that can give you the sceptical thesis you want.

[Derk Pereboom]

Neil, you’re right to suggest that I’m also committed to a general sort of explanation of non-responsibility that covers both deterministic and indeterministic cases – I indicate this sort of commitment at the end of the previous post. But the truth of such a general explanation does not preclude that the best explanation for Plum’s non-responsibility in Case 1 and Case 2 is that his action is deterministically produces by causal factors beyond his control. The compatibility of a more specific best explanation with the truth of a more general explanation is illustrated by the WTC analogy. The four-case argument aims to undermine the notion that moral responsibility and causal determinism are compatible, and given that this is the goal, the thesis that causal determination best explains Plum’s non-responsibility in Cases 1 and 2 is especially germane. So I could (and do) endorse the type of answer to Al that you suggest, but I also endorse a reply that cites the more specific explanation, and this reply is especially pertinent in the dialectical context.

[Neil]

Obviously, this takes us back to Al's claim about how we determine what the best explanation is. I have nothing to add to that debate. But there seems to be a disanalogy between the WTC case and Plum. A plane *did* hit the WTC. Determinism itself didn't exert any causal force on Plum. Determinism is not the kind of thing that has causal force. Plum was caused to act as he did, and the fact of causation without control explains why (if he was not responsible) he wasn't responsible. Determinism is simply a way of characterizing the nature of the causal relation. After all, we might know that Plum was manipulated in a near-future actual world, and therefore feel disposed to excuse him of responsibilility, even though we do not know whether the actual world determinism is true or not.

[Derk Pereboom]

I’m not arguing that the WTC analogy helps us see what it is about causal determinism that precludes moral responsibility – so the point of the analogy is not to claim that just as the plane exerts causal force on the WTC, determinism exerts causal force Plum. (I do think that an action’s being causally determined by factors beyond the agent’s control precludes moral responsibility because this is one specific way in which an agent can lack the sort of control over action that’s required for moral responsibility, and that an action’s having a certain kind of indeterministic history is another way an agent can lack this sort of control.) Rather, the WTC analogy is intended to show that a more specific explanation ("the WTC collapsed because it was hit by a plane”) can be the best one in a certain context, while there is also a correct more general explanation (“the WTC collapsed because it was hit in a certain way by an object with such-and-such general characteristics”) that covers more kinds of cases.

[Neil]

Well, I may be doing nothing more here than expressing my own puzzlement over incompatibilism. I have never had an incompatibilist intuition, and I'm not sure I even understand them. But I'm not a compatibilist either. The WTC collapsed in virtue of the impact of a massive body; having a plane fly into it is one way in which such an impact can come about. Plum lacks responsibility because he fails to control his actions. It seems to me that causal *determinism* is not one way in his lack of control could come about; the determinism itself is a simply an abstract description of the nature of the causal relations which really brought about his lack of responsibility. It is the causation by factors he did not control that explains the lack of responsibility. I just don't see the work that determinism is doing. But then I *never* see the work that determinism is doing.

[Eddy Nahmias]

Neil, could you explain how you are not a compatibilist? If you don't think that determinism precludes free will, then I would think you are a compatibilist by definition of the terms. That is true even if you think we don't have free will--even though every compatibilist I know of seems to think we have free will.

Derk (on a related note), on your view, is *determinism* incompatible with free will? Assuming the answer is yes, is it only because determinism is one (of several) theses that would entail that an agent is not the source of her actions? Would this thesis also be incompatible with free will: the falsity of agent causation? That is, on your view, is there anything special about determinism or is it just one way to suggest the more general problem for free will?

[Derk Pereboom]

Eddy, I do think that determinism is incompatible with the sort of free will required for moral responsibility, and that determinism is one of several theses that entail that an agent is not the source of her actions in the way required for moral responsibility.

My view is that determinism is incompatible with this sort of free will, since if determinism is true, then all actions would be what I call alien deterministic events: events for which there are factors beyond the agent's control that inevitably produce their occurrence. If actions are events of this sort, then agents cannot be their source in the way required for moral responsibility.

On the kinds of indeterminism that preclude this sort of free will, actions could be truly random events: events that are not produced by anything at all. Then the agent wil not be the source of the action at all, let alone the source of the action in the way required for moral responsibility. Or actions could be partially random events: events for which factors beyond the agent's control contribute to their production but do not determine them, while there is nothing that supplements the contribution of these factors to produce the events. Then, for instance, factors beyond an agent’s control might result in its being open whether she will decide to stop and help or to refrain from doing so, but given the causal contribution of these factors beyond the agent’s control, there is nothing that settles whether her deciding to stop and help occurs or not. Suppose she does decide to stop and help. Then again she will not be the source of the action in the way required for moral responsibility.

So the way in which determinism precludes the sort of free will at issue is different from how indeterminism can preclude it. If determinism is true, then factors beyond my control produce my actions in the sense of making them inevitable. This isn’t so for truly or partially random events. Here the absence of the requisite sourcehood results from there ultimately being nothing that settles whether the action will occur.

Agent causation is the only hypothesis about agency that I’ve come across that (if it turns out to be coherent and possible) I think would allow the agent to be the source of her action in the way required for moral responsibility. Perhaps there are other hypotheses about agency that would allow this -- I don't think that we're in a position to rule this out. But I haven’t encountered them; maybe there are such ways but no one has conceived of any of them. So I’m agnostic on whether the falsity of agent causation is incompatible with the sort of free will required for moral responsibility.

[Neil]

Eddy,

I can see why you would want to call me a compatibilist. You carve up the territory in such a way that if you are not an incompatibilist - ie, you do not think that determinism precludes free will - you are a compatibilist. But I don't want to carve up the territory in that way. It seems to me that worries are determinism are really worries about (variously) mechanism, control and luck (or just confused). I think that these things do preclude free will, and that determinism has nothing to do with it. Moreover, 'compatibilism' to me is a success term.

[Al Mele]

Here are some comments on some of what came in after I last replied. I realize I’ve been slow in replying, but (1) I’ve been grading papers and (2) you seem to have been getting along fine without me.

1. Side effects and lucky success.

Kip writes: “I see the difference between the side-effect and trying cases. But I am wondering what basis you have for saying: ‘An agent who luckily succeeds in doing something very bad is given a high blame rating for doing it, and an agent who just as luckily succeeds in doing something very good is given a high praise rating for doing it.’”

See, e.g., Nadelhoffer, “On Praise, Side Effects, and Folk Ascriptions of Intentionality” (JTPP 2004).

2. The 4-case argument.

Kip writes: “do you acknowledge the no-difference argument in Pereboom’s work? Were you just intending to target the best-explanation argument, as separate from the best-explanation argument? Or have these two been, to some degree, conflated?”

Some people definitely do confuse the two kinds of argument with one another. As I explained in the paper, the logical forms of the two different kinds of argument are very different. And, as I say at the end of section 2, in a paragraph I mentioned last time, “A fallback position is available to a proponent of Pereboom’s manipulation argument to the best explanation who concedes the elementary points that I have made in this section about that argument. It is to rewrite the argument as a straight manipulation argument from a trio of cases. (In Mele 2005 and 2006, in addition to rebutting Pereboom’s manipulation argument to the best explanation, I examine the three cases he used to generate the Manny premise. I forego doing so again here.)” Recall that a Manny premise appears in arguments of both kinds.

Derk says in one of his posts that if, in cases 1 and 2, “the manipulators are replaced by machines that randomly form in space and that have the same effect on Plum as the manipulators do, our intuition that Plum is not morally responsible remains.” I agree that, as I put it in the target article, “the responsibility-undermining and freedom-blocking effects of intentional manipulators can be matched by blind forces” (on this, see, e.g., pp. 168-69 of my *Autonomous Agents*). Regarding cases of this kind too, we should ask whether “*The best explanation* for the intuition that Plum is not morally responsible in the[se] cases is that his action results from a deterministic causal process that traces back to factors beyond his control” (Pereboom 2001, p. 116). Given what I say in the paper, my way of testing the claim is predictable.

3. The zygote argument.

Kip writes: “If the Zygote Argument works, and contradicts one’s compatibilist theory of autonomy/freedom, then one must have made a misstep somewhere along the way while building that theory. . . . [W]hat would that misstep be?”

I probably misunderstand this question, because, as I read it, the answer is obvious. If the zygote argument succeeds, then the crucial problem with the compatibilist set of sufficient conditions I offer is that they are satisfiable in deterministic worlds. (Did you mean to be asking something else, Kip? The conclusion of the zygote argument is that “determinism precludes free action and moral responsibility.” I don’t have the impression that you were thinking that, in my view, determinism can preclude these things without precluding autonomy.)

4. Personal identity.

Paul’s latest thoughtful post on personal identity provides confirmation for my thought, expressed in an earlier post, that a careful attempt to resist my manipulation stories on grounds that feature considerations of personal identity might be very interesting. What would work best is an account of personal identity that we know to be true such that the combination of it with details of stories of the kind I spin entails that the pre- and post-manipulation agents are not the same person. Of course, one needn’t aim this high. Showing that a very plausible account of personal identity has this result would be a significant achievement.

Some traps should be avoided: for example, coming up with an account of agent-identity according to which the pre- and post-manipulation agents are not the same *agent* while leaving it open that they are the same person. At least, I regard this as a trap. (But I have no use for a notion of “agent” according to which it can happen that although x at t1 and x at t2 are the same person and that person acts at those times, x at t1 and x at t2 are not the same agent.)

[Kip Werking]

Al,

About the Zygote Argument: I'm not sure how to rephrase my question. Obviously, if the Zygote Argument (ZA) works, then it conflicts with your theory of autonomy. (You're right that I am not here splitting hairs between fw/mr/autonomy). You might better understand my question if you consider my suggested answer:

"your view allows for agents to autonomous possess sheddable (and unsheddable?) proattitudes that bypassed the agent’s not-yet-existent capacities for rational control just because the agent was built or constructed with those pro-attitudes either existing, or latent, within it."

In other words, if we reach the final problematic conclusion that some agents in determistics worlds are autonomous (according to your criteria), and yet the ZA works, *how* did we reach that conclusion? What went wrong? If the conclusion of our valid argument is unacceptable, which premise was false?

It might be that sufficient conditions for autonomy should include that all unsheddable pro-attitudes, and not just later ones, do not bypass the agent's faculties for rational control. For example, suppose that a heterosexual male has a pro-attitude of desiring women, and suppose that this is largely or entirely a result of his genetics (I realize this is controversial; let's put aside that worry for now). As I understand it, such a man can still autonomously possess that unsheddable proattitude, even though it is in a sense innate and bypassed his rational control faculties. My suggestion is that *this* might be the flase premise, the logical mistake, that led one to the mistaken theory of compatibilist autonomy (if it is mistaken, if the ZA is correct).

Suppose that it is. Then a true set of sufficient conditions for compatibilist autonomy would include that all unsheddable pro-attitudes, and not just later ones, do not bypass an agent's faculties for rational control. This would be a correct set of conditions; the only problem is that no deterministic agent could ever satisfy them (because rational control presupposes pro-attitudes; some of them *must* be given or else the agent is just a blank slate, and blank slates don't do anything).

But perhaps I am wrong, and this is *not* the misstep. Perhaps your condition about bypassing is perfectly fine, and it is some other condition or aspect of your view that is causing the trouble (if the ZA is true and causing trouble). I doubt this. But I wanted to ask you what you thought. Hopefully you can better understand my question now.

[Kip Werking]

I realize that my question, as articulated, could still be tighter. So here, in one last effort to lure Al into replying, is a final, and concise, statement:

If your sufficient conditions for autonomy have multiple aspects or parts, and if we suppose that the ZA is at partially correct, and so your sufficient conditions for autonomy must be at least partially incorrect, which aspect(s) or part(s) of your sufficient conditions do you think is too weak? Would you agree with my modest suggestion: it is the BYPASS condition, and not any of the others, that is too weak?

[Al Mele]

I've been meaning to reply. I don't know what the suggestion that the Zygote argument is "partially correct" is supposed to amount to. Suppose the Zygote argument is sound. Then compatibilism is false and any set of proposed sufficient conditions for free action and moral responsibility that is satisfiable in a deterministic world is false. I don't see that the Zygote argument poses any specific problem for anything that I say specifically about bypassing.

Kip mentions "the BYPASS condition." Might he have in mind what Cuypers calls "BYPASS"? On that, see pp. 17-19 of the target article. BYPASS certainly isn't a principle of mine.

[Kip Werking]

Al, thanks for the reply.

I think an analogy might help get my point across. Suppose that a typical table chair consists of at least four legs, a seat portion, and a back portion. Then partially correct sufficient conditions for a typical table chair might be:

1. a seat portion
2. a back portion
3. two legs

The set of conditions is not correct because it is not sufficient. But it is partially correct just in the sense that it seems closer to being sufficient than, for example, the following conditions:

1. has hair
2. is blue

These conditions have nothing to do with being a typical table chair. So the earlier conditions seem partially correct, and closer to being sufficient, than the later ones.

In this way, we can imagine degrees of correctness. The set of conditions above is partially correct, but is less partially correct than the following:

1. a seat portion
2. a back portion
3. three legs

Suppose that someone came along and said "this set of conditions is not correct." What would the problem be? The problem is not the first or second condition. Every typical table chair has these. The problem is with the third condition. It is clearly too weak. To be a typical table chair, the thing must have at least four legs. Three will not suffice.

[Note that by BYPASS, I do *not* have Cuypers' definition of mind. I was just referring to your historical condition for autonomy, and I had forgotten the name of it. According to this condition, the agent cannot have any unsheddable pro-attitude that bypassed the agent's faculties for rational control. Let's call this HISTORICAL.]

Now, if I remember correctly, your sufficient conditions for autonomy look something like this:

1. HISTORICAL
2. reliable deliberator
3. ideally self-controlled
4. beliefs conducive to informed deliberation
5. [other(s) I may be forgetting]

Now, my suggestion is that your sufficient conditions for autonomy are like the partially correct conditions for a typical table chair: they are not quite sufficient (given the success of the ZA), and it is only one condition that is really problematic---HISTORICAL. So HISTORICAL is like the condition "two legs" or "three legs": it is too weak. And your conditions about being a reliable deliberator, being ideally self controlled, and so forth are like "seat portion" and "back portion": they are just fine, and one would not need to touch them in order to revise your conditions and make them actually sufficient (given the ZA's success). Note that saying they are “just fine” does not imply that they are necessary, because they may indeed be *more than necessary* (e.g. perhaps the condition that the agent be ideally self-controlled).

Now, you note that "Then compatibilism is false and any set of proposed sufficient conditions for free action and moral responsibility that is satisfiable in a deterministic world is false." This raises the possibility: what if any sufficient conditions for autonomy cannot be satisfied?

Suppose, with me, that HISTORICAL is too weak. It allows the agent to autonomously possess too many pro-attitudes without having rationally chosen them. We would need to say that the agent has to chose more such pro-attitudes in order to be autonomous (let's put aside the question of how many such pro-attitudes must be chosen). Then this stronger condition would be HISTORICAL*.

Now suppose, with me, Nagel, and G. Strawson, that freedom/autonomy is logically impossible, and that the problem involves luck and self-creation and nothing especially about determinism (indeterminism would be just as bad). And finally suppose that HISTORICAL* is one aspect of this freedom and is the aspect that makes it impossible.

How might it be impossible? The key is to realize that to rationally choose pro-attitudes, one must already have some pro-attitudes. One cannot choose between, for example, love and hate, without already having some sense of whether love is better than hate. An agent with no pro-attitudes, a blank slate, could not choose anything at all.

In this case, as you say, "any set of proposed sufficient conditions for free action and moral responsibility that is satisfiable in a deterministic world is false." This is because (I suggest) HISTORICAL* is a necessary condition for freedom/autonomy, and so any proposed sufficient set of conditions must contain it, and HISTORICAL* cannot be satisfied in any world---deterministic or indeterministic.

But this is perfectly consistent with, and leaves us with the possibility that, (given the ZA's success), it is HISTORICAL alone, and not any other of the items in your set of sufficient conditions for autonomy, that is causing the problem. To return to the analogy, HISTORICAL is like the "two legs" or "three legs" condition for being a typical table chair; the "back portion" and "seat portion" conditions are just fine, just as the reliable deliberator and other non-historical conditions on autonomy may be (I suggest) just fine.

But maybe you disagree. Maybe you think HISTORICAL is just like the harmless "back portion" and "seat portion" conditions, and it is some other (non-historical) condition that is problematic. Perhaps it is the ideally controlled agent condition. I doubt that it is, but I wanted to ask for your opinion.

I hope you can better understand my question now.

[Mark Smeltzer]

Kip,

It seems rather lazy to simple assert that Mele's set of conditions is false or even only partly true. Typically, a counter-example is considered the academically fashionable way to trounce someone else's proposal.

Two different methods for counter-exampling come to mind: one is to present an example of something that the conditions exclude that should not be (e.g. a chair that the conditions fail to identify), and the other is to present an example of something that the conditions include that should not be (e.g. a non-chair that the conditions identify).

In both counter-exampling techniques, an actual *thing* is presented that invalidates the conditions. Since you have in mind an agent-power that is logically impossible by definition, the first technique is necessarily unavailable since it is impossible to present an agent with a logically impossible agent-power.

So that leaves only the second technique. You would have to give examples of agents that do not have the logically impossible power, meet all of Mele's other criteria, are still perceived as genuinely non-free and the only reason for being non-free is that they lack the logically impossible agent-power. Hum... I cannot even imagine what that kind of case would look like.

Perhaps this challenge is part of the reason that neither Nagel nor G. Strawson have had much luck marketing their ideas... but who knows, maybe you will succeed where they failed! On the other hand, maybe you will realize they were off their rockers :)

[Kip Werking]

Mark,

I do appreciate your fiesty spirit during our exchanges. :)

In response to your inquiry, I hope it is apparent from this thread that both Al and I are considering the agent from the Zygote argument as posing a threat, if it does, to the sufficiency of his criteria for autonomy. So this would be an example of the second method of counter-example you describe.

[Note that I only ask these questions in the context of the success of the Zygote Argument; Mele does not feel that it clearly succeeds although he remains sympathetic to it.]

When you ask that I also show that "the only reason for being non-free is that they lack the logically impossible agent-power", I do not presume to have a definitive proof of this (not yet, at least). Rather, this is my hypothesis!

That is why I wanted to ask Al what he thought. If Al is sympathetic to the ZA, then the lack of the impossible power might explain where his sufficient conditions went wrong. Alternatively, he might think that the impossible power is not necessary and, rather, it is some other aspect of his conditions-for-autonomy that needs revision. I am interested in learning where he feels his view went wrong (if it did).

Perhaps I could summarize: it's not enough to say "this doesn't work." The question, for me, is: why didn't it work? If it's dead, let's perform an autopsy.

[Mark Smeltzer]

Kip,

Let's grant for a moment that the ZA is a counter-example to Mele's list of conditions. What would that prove? It would simply show that Mele's list is wrong.

Would it then follow that the deficency in Mele's list is that his list lacks a condition that the agent must have a logically impossible agent-power? It seems you would have to provide an argument (not a case) to make that point, and I'm sure champions of freedom would find lots of points to disagree with you on in that argument. For instance, as you suggested, maybe Mele's list is simply missing some other less radical condition that can account for the ZA-case.

My point is that I simply cannot imagine a counter-example that would, without an accompanying argument, provide sufficient intuitive force to convince a champion of freedom that freedom is logically impossible. I may simply have a poor imagination.

To date all of the arguments I have seen from the skeptical crew constitute either refutations of specific views or a seemingly dogmatic commitment to some kind of assertion of logical impossibility. Were I on that side of the debate, I would certainly have greater ambitions...

[Kip Werking]

Mark,

I think we agree with each other. If I wanted to prove that Al's conditions are wrong just because they lack an impossible power condition, I would need an argument. But if I just want to ask Al whether he shares my suspicion that this is how his conditions went wrong (if they did), I don't need an argument.

We agree on this much, right?

[Mark Smeltzer]

Kip,

Of course, but in that case, there's not much at stake regardless of the response ;)