X-Phi on Free Will Needs a Makeover

 The approach that experimental philosophers take to the free will problem relies on two claims, both true:

1. Compatibilist and incompatibilist philosophers defend their theories by appealing to the intuitions of their audience.
2. So, evidence about the nature and prevalence of these intuitions can have significant substantive implications about the plausibility of these theories.  

In order to draw out these implications, however, the studies have to probe for the correct intuitions—namely, the ones that philosophers appeal to in their theories.   Unfortunately, most studies on free will are not designed to investigate those intuitions.  And this raises serious doubts about the philosophical conclusions that they wish to draw from the results. 

So what are the intuitions that X-phil studies are designed to investigate?  Well, with a few exceptions, they probe for intuitions about the compatibility question itself, in other words, on the conclusions of compatibilist and incompatibilist arguments.    The first Nahmias et al studies offered various ways of describing determinism and then asked respondents if people could be free/MR in such a world.  Nichols and Knobe did the same: they describe two universes, deterministic and indeterministic, then asked abstract, concrete, and emotionally charged questions about whether agents (or a particular agent) could be MR in the deterministic universe.   Virtually all subsequent studies have followed suit—in large part perhaps because they wish to develop or respond to earlier work in the literature.  When subjects give, say, compatibilist answers, the philosophers often conclude that incompatibilism is not so intuitive after all (at least under certain conditions).   But in order to draw that conclusion, the intuitions that incompatibilists rely upon would have to be the ones investigated in these studies.  And they’re not.      

In one sense, this should be obvious.  Would this be a good argument for incompatibilism?

1. Intuitively, free will and MR is incompatible with determinism
2. Therefore, free will and MR is incompatible with determinism.

Of course not.  Incompatibilists do not simply assert that the truth of their conclusion.  They develop (often ingenious) ways of revealing the threat of determinism to free will and MR.  In doing so, they certainly appeal to intuitions, but to intuitions about cases and narrower principles (the principle of alternate possibilities or the ‘transfer of non-responsibility principle’ or Rule B etc.), never on the compatibility question itself.

 Interestingly, in their first 2006 article, Nahmias et al anticipate an objection along these lines.  (And to my knowledge, no one has addressed this issue since.)  But their response doesn’t work.  After conceding that their surveys do not probe for the intuitions that are supposed to underwrite key premises in van Inwagen’s consequence argument, they write:

However our results do offer some indirect evidence against the intuitive plausibility of the Consequence argument. Our scenarios present conditions in the past that, along with the laws of nature, are sufficient conditions for the agent’s action. So, the fact that most participants judged that the agent in the scenarios is free and responsible seems to suggest either (a) that they have the intuition that the Transfer principle does not apply to free choices, or (b) that—regardless of the soundness of the Consequence argument—the concept of choice the argument invokes to reach the conclusion that “we have no choice about any truth about the future,” does not accord with the concept ordinary people consider relevant to free will and moral responsibility

This reply appears to assume that philosophically unsophisticated subjects should have somehow internalized or implicitly grasped the consequence argument before ever hearing it!  Van Inwagen has received credit for reviving incompatibilism in the past century in large part because he found a non-obvious way of showing the threat of determinism to free will.  Again, the whole point of developing an incompatibilist argument (rather than just asserting the conclusion) is to draw out implications of determinism that might otherwise go unnoticed.   

None of this is to say that the work in the field so far has no value.  I think the results are fascinating.  Learning about the effects of factors like psychological distance, emotional salience, personal temperament on our free will judgments is hugely important.  But right now, a lot of the insight and ingenuity that experimental philosophers are justly famous for is focused on the wrong questions.  My biggest worry is that the current probes are becoming entrenched in the literature and will be hard to get away from just by virtue of how prevalent they are.  At a recent conference in New Orleans, I raised this kind of objection (inarticulately, it was New Orleans) to Adam Feltz, and he replied that his probes were “industry standard.”   And he’s right.  It’s time to change that standard, developing  new types of probes that target intuitions about the premises of philosophical arguments rather than the conclusions. 

Thoughts?

[Note: For a somewhat more detailed defense of the ideas in this post, check out my article in Philosophy Compass (http://www.experimentalphilosophy.org/papers/Sommers.pdf)  especially section 4. ]  

Comments

[Adam Lerner]

Hi Tamler,

I wish I had more to say than that I think you're exactly right. Well, maybe I can think of something…

Let’s say you run an experiment testing to see whether some version of the transfer of non-responsibility principle is intuitive. You decide to present participants with the following formulation of the principle, taken from your “More Work for Hard Incompatibilism”:

If an act is the result of processes that trace back to factors beyond the agent’s control, then the agent does not deserve blame or praise for that act.

Say you get the following result: pretty much everyone strongly agrees with it. Hurray! Folk incompatibilist victory, right? Maybe. What are compatibilists going to say? They might think participants are taking the principle to entail something it doesn’t—maybe something like epiphenomenalism or fatalism. How do you test to make sure they aren’t? Well, I imagine you would include some manipulation checks. So you run it again with the manipulation checks. Amazingly, everyone passes them and still agrees that the TNR Principle is super-intuitive. This, I think, would surely be an important finding.

But would it show that the folk are natural incompatibilists? Sure, as long as natural incompatibilism and natural compatibilism aren’t mutually exclusive. After all, there’s good evidence that people do tend to think determined agents can be morally responsible. If you wanted to test to see which of the intuitions the folk found *more* intuitive, you’d have to run another study in which you show people the transfer principle, show them an industry standard deterministic vignette, and then ask them whether they agree more with the transfer principle or with the statement that the determined agent is morally responsible. But this just amounts to your run-of-the-mill x-phi of free will study with an implicit incompatibilist argument thrown in at the beginning.

And would the responses to this study even count as pre-theoretic intuitions? Does folk incompatibilism mean, “would agree to incompatibilist conclusions after being presented with incompatibilist arguments”? If so, wouldn’t it be open to compatibilists to present participants with compatibilist arguments before asking them the compatibility question and then, if they got the desired result, claiming the folk are natural compatibilists? And couldn’t it just go back and forth like this, showing the participants more and more arguments from both sides and seeing what they think? But then it seems like the question about what the folk find intuitive just amounts to: “What would the folk believe if they themselves were philosophers who considered all of the relevant arguments and had to make a considered judgment?” This seems a long way off from trying to get at people’s pre-theoretic intuitions.

Perhaps my point is this: the test for any theory of moral responsibility is going to come down to directly asking the compatibility question. Incompatibilist arguments may direct our attention to features of determinism that we didn’t notice before, but we still always have to decide whether we want to apply the transfer principle in any particular concrete case of deterministic agency. I think, then, that the question of whether the folk are natural incompatibilists can be best answered by (1) making sure people completely understand what determinism is and is not, and then (2) asking them the compatibility question. Anything more than that (like presenting arguments rather than just clarifying what determinism is) and I’d have a hard time calling people’s responses “pre-theoretic intuitions”.

That all being said, I definitely do think studies should be run probing how intuitive people find the premises of incompatibilist arguments. I think these are interesting questions of moral psychology in their own right. I also think further studies exploring *why* people find these premises intuitive (if they do) could have a significant impact on the debate.

[Tamler Sommers]

Hi Adam,

Thanks for the long detailed reply. I agree that you can't ask people directly about whether they endorse the TNR principle. But there may be ways of ascertaining more indirectly whether people endorse that principle--through a series of cases (maybe modeled after the four case argument), including cases designed explicitly to show that the agent intentions and desires are not epiphemonena. If you get positive responses, then this will be some evidence that people endorse the TNR principle.

I take your second point to indicate that if you do what I suggest, you'll end up having to give the subjects a description of determinism and then asking for responsibility judgments anyway. But I don't think that's true. Think of the dialectic between Derk and Al about the four case argument. Derk presents a bunch of cases (before the deterministic case), and argues that the best explanation for our judgments is that we endorse the TNR principle. Al says that another equally plausible explanation is that we withhold responsibility because of a conscious manipulator. Derk comes back with his computer that's randomly assembled (something like that) and claims that we still don't attribute responsibility. Nowhere in that dialectic is determinism described and rightly so--because the goal of the four case argument is to establish incompatibilism and not merely assert it to be true. So I think there's a lot of room to test these kinds of premises without getting into the compatibility question itself.

One last quick note: one could test the theories that compatibilists offer too. Susan Wolf argues that the deep self view is insufficient and offers Jojo as a counterexample. Certainly, one could develop a JoJo like case to see if people really do absolve him of responsibility because of his inability to know right from wrong. (I think David Shoemaker has done this actually, not sure if it's published yet. And maybe Chandra Sripada too?) Anyhow, I think that's a really fruitful direction to go in.

[John Fischer]

Tamler,

Mark Ravizza and I have argued that TNR fails. Mark got the ball rolling in his Phil. Studies paper, and then we further developed the critique of TNR in our book, Responsibility and Control. Our argument against TNR makes use of Frankfurt-style cases.

So: Jones is not morally responsible for the fact that Black is standing by (with all the bells and whistles). And Jones is not responsible for the fact that, if Black is standing by (with all the bells and whistles), Jones will (say) vote for Obama (i.e., behave as he actually does). But nevertheless, Jones is morally responsible for voting for Obama (i.e., acting as he actually does). Hence, TNR is false.

Here's a little X-phi of my own: what do you think of this argument?

[Tamler Sommers]

Hi John,

Honestly, I'm not sure what to make of these kinds of overdetermination cases. I think people would certainly judge Jones to be responsible. But in some sense, these cases seem to sidestep the real force of the TNR principle since it could still be true that Jones is ultimately responsible for his decision in some libertarian sense. Nothing about the case suggests that Jones' actual decision to vote for Obama is caused by factors beyond his control, right? So that could account for people's judgments. Maybe what's required is to refine the TNR principle and restrict it to the actual causes of a person's behavior.

I have to go back and look--I'm at home-- but isn't the avalanche case different? In that case, Mary doesn't actually cause the avalanche, but we're supposed to judge her responsible nevertheless. Is that right? That might be an interesting case to test...

[John Fischer]

Thanks, Tamler. You do graciously note the existence of such purported counterexamples in your book ms--which I highly recommend to everyone, Relative Justice (Princeton U Press). Is it out yet?

Anyway, I was thinking that TNR would have to be used to get to the conclusion that when a decision is caused by factors outside the agent's control, he is thereby not morally resp. for it. So, if TNR is under assault, one is not entitled (yet--unless one has some other reason) to say that, because Jones's decision is cause by factors out of his control, he is not morally responsible.

Anyway, yes, the Avalanche case is about consequences, rather than actions, but I think the issues are similar.

My point: TNR is problematic, and (as the subsequent literature bears out), it is not straightforward to adjust it to get the incompatibilist result.

[Adam Lerner]

Tamler,

Thanks for your response. I agree that the best approach is probably to present people with something like the four-case argument, and I agree that it would be important to get people’s responses to each of these cases. I just don’t think simply asking people about cases like the first three (and getting the intuition that the agent isn't responsible) is going to be enough to show that they endorse a transfer principle wide enough in scope to include determinism, or that such a principle is the best explanation of their intuitions about the first three cases. I don’t even think asking people about the random machine case can do this. If participants say that the agent isn’t responsible, this doesn’t necessarily warrant the inference that they're implicitly committed to the transfer principle. After all, doesn’t Mele still think that the agent in the machine case (or any “blind forces” case) isn’t responsible for reasons that are independent of TNR? If there is a viable alternative explanation, then even non-responsibility judgments in the machine case can’t tell between the different explanations on the table.

I’m not saying getting the desired result in the first three cases and the machine case wouldn’t be suggestive. It definitely would be. I just think the inferences you can confidently make are limited unless you go on to ask people about the deterministic case.

As for testing compatibilist theories: definitely. In fact, I think it is probably easier to do this. I know Chandra has that paper where he uses structural equation modeling to argue that people’s intuitions about manipulation cases are tracing the perceived absence of some compatibilist conditions on moral responsibility. I think this is an incredibly valuable technique. I’m just not sure how one would use it to tell between TNR and alternative (compatibilist-friendly) explanations for non-responsibility judgments in cases where everyone understands that the manipulated agent satisfies all (or nearly all) compatibilist conditions on moral responsibility.

By the way, I believe David's paper is published and Chandra's is forthcoming.

[John Fischer]

Thanks for your thoughtful comment, Adam.

I myself think that TNR should be rejected. There are cases that appar to show that TNR is invalid. Further, I think it is an incredibly blunt instrument--way stronger than necessary in order to explain our intuitions (woops--I used "our") about manipulation cases. I think that in certain manipulation cases, the agent is not morally responsible, but I certainly would not invoke an instrument as blunt as TNR to explain what is going on in the manipulation cases.

A nice discussion is in Kristin Demitriou's AJP paper.

In any case, I resist the idea that the only, or best, way to avoid saying implausible things about manipulation cases is to invoke TNR, which is in any case already tarnished (at best).

(btw, when I use the term "our", I refer to Mark Ravizza, me, and my mother--maybe Mark's too).

[Tamler Sommers]

John, thanks for your kind words about the book! It'll be out in December.

You say the TNR principle is too blunt an object to explain our intuitions in manipulation cases. But what if instead of focusing on conditionals, TNR focused on actual causes. So (roughly speaking), it's the view that if we are not responsible for Y, and Y causes us to perform X--and we are not responsible for the fact that Y caused us to perform X-- then we are not responsible for X. The non-responsibility still transfers, but now it's not vulnerable (if I understand it properly) to your counterexample. Does that sharpen the instrumental a little? Or, to mix a metaphor, does it make it out of tune?

Adam, you write:

"I’m not saying getting the desired result in the first three cases and the machine case wouldn’t be suggestive. It definitely would be. I just think the inferences you can confidently make are limited unless you go on to ask people about the deterministic case."

I think agree with you--but how will asking people about the deterministic case make you more confident about your conclusion?

[John Fischer]

Tamler,

Good, thanks, but now I'm not sure that one could use the newly revised TNR to get to incompatibilism about causal determinism and moral responsibility. Say Y is some event in the remote past. Now in a deterministic world, I'm not morally responsible for Y. But insofar as I freely (in a compatibilist sense) perform X, that is, I (say) exhibit guidance control in performing X, then why is it obvious that I am not morally responsible for the fact that Y caused me to perform X?

Of course, it will be controversial that simply in virtue of meeting compatibilist friendly conditions I will be acting freely in performing X. But my point is that it seems to be essentially contentious in this context whether the second premise (I am not responsible for the fact that Y causes me to perform X) is true. Thus, one couldn't use (revised) TNR to get to incompatibilism.

[John Fischer]

Tamler,

Also, when TNR is interpreted as pertaining to conseuqences, there will presumably be simultaneous overdetermination counterexamples to TNR. So Sam and Linda independently and "intuitively freely" shoot the mayor. We can tell the story such that Sam in not morally responsible for the fact that Linda pulls her trigger, and also Sam is not responsible for the fact that Linda's pulling the trigger causes the mayor to be shot. And nevertheless Sam is morally responsible for the mayor's being shot.

Granted: the counterexamples, and the associated interpretations of TNR, are a tricky and delicate business. I suppose my main point is simply that it is not at all clear that TNR, in a form that clearly leads to incompatibilism, is acceptable. It is, in my view, an essentially contested principle, and, as such, cannot help to make much progress in certain dialectical contexts.

[Eddy Nahmias]

Just to pick up on John's point, it's also unclear whether Y (some state in the remote past) *causes* X (my performing some act). If determinism is true, Y (plus the laws) *entails* X. But I'd like to hear an explanation of why we should interpret Y to cause (to be the cause of, the be a cause of) my performing the act. There's a reason PvI says he eschews talk of causation in his consequence argument.

If we don't interpret Y as causing X, then the new TNR principle won't work (and it's hard to see how we can ignore the counterexamples). If we keep the causal reading of TNR, then we can't use Y (unless we have a good account of how Y causes X). So, instead we'd have to use some state of affairs R much closer to X. Once we do that, you won't get clear examples of R (recent states) that an adult agent will be entirely non-responsible for. So, you'd need to go back to infancy. And then, I think, our intuitions get murky (how shall we test them??). Can an infant have just a tiny bit of responsibility for an action even though she did not have any responsibility for the states that caused the action? This suggests a point we can probably agree on--that compatibilist accounts will need to allow (proto-)agents to "grow" freedom and responsibility by degrees from non-agency (whereas libertarians presumably want a spark of free will to emerge full-blown at some point?).

I'll see if I can muster some responses to the original post, but I'm happy to agree that we should try to move beyond (though not ignore!) the early attempts to test intuitions about free will by presenting descriptions of determinism (or "determinism-like). I've been trying some manipulation cases, and as you (Tamler) know, we've been trying to come up with ways to test people's intuitions about TNR or related principles.

[Justin Capes]

I'd like to see what John and others think about the following, which is in line with Tamler's earlier suggestion:

Consider this TNR-principle:

TNR-co: If S is not even partly to blame for a causal sequence C, S is not even partly to blame for the fact that C terminates in E, and there is no independent causal sequence that also terminates in E, S is not even partly to blame for E.

This is a version of a principle Michael McKenna suggested in his 2001 APQ piece. Now, John (in My Way) objects that the principle can't be used as part of a general argument for incompatibilism, since it won't apply to worlds in which all the morally relevant outcomes are overdetermined. But even if this is so, couldn't it be used to defend the premise in manipulation arguments according to which agents in suitably constructed manipulation stories (e.g., Mele's story about Diana and Ernie) are not to blame for the actions they perform?

John seems to think not, for in cases of unimpaired agency, why couldn't the agent be to blame for the fact C terminates in E? This is certainly a question the proponent of this style of argument would need to address, but one can see how the answer might go (the agent doesn't cause C to terminate in E, couldn't have prevented C from terminating in E, and so on).

Here's another TNR-principle:

TNR-c*: If S is not even partly to blame for any of the events or states of affairs that are among the causes of E, and if S is not even partly to blame for the fact that these events and states of affairs cause E, then S is not even partly to blame for E.

This principle too, it seems, could be used either in a direct argument or to augment manipulation arguments in a way similar to that adumbrated above.

Notice that neither TNR-co or TNR-c* is threatened by Fischer/Ravizza style overdetermination cases.

[Tamler Sommers]

John, right, the point of the modified TNR principle is to make it invulnerable to the Frankfurt style counterexamples you mentioned which I believe dodge the true force of TNR by targeting the conditionals. One would still have independently to defend the plausibility of TNR.

As for the new counterexample: here's where my intuitions may differ from yours, and so maybe this is ripe for experimental fodder. If the point is (like the avalanche case) that Sam's bullet doesn't actually hit the mayor but would have if Linda's hadn't gotten there first, then I don't hold Sam morally responsible for the mayor being shot (although he may be MR for trying to shoot the Mayor). If the bullets get there at the same time, then the proper way to frame it seems to be this: Sam is not MR for the mayor being shot by Linda, but he may be MR for the Mayor being shot by him. Insisting that we only consider the case of the mayor being shot simpliciter seems to get the counterexample on the cheap. As you say, interpreting these cases properly is a delicate business.

Eddy,

Right, the causal version makes it far less easy to establish that the agent is not responsible for what causes the action. But setting aside how useful the causal version is for establishing incompatibilism, do you think it's true?

On your second point, you're right that the key issue will be how it's possible to perform our first responsible action, which will take us back to the responsibility of very small children. So maybe that's the thing worth testing?

On the third point, yes, I do know that which has made me all-to-familiar with the difficulty of testing narrower principles. And of course I agree that no one should ignore the early studies. I think we learn a huge amount from them about the conflicting intuitions people have about free will and MR, and the various factors that sway judgments one way or another. What I would like to know is whether these factors influence our intuitions about the narrower principles concerning the conditions of free and morally responsible behavior. (And my guess is that they do!)

[Adam Lerner]

Tamler,

If you gave people the deterministic case and they judged the agent non-responsible, this would seem to me like pretty good evidence that they endorse something like TNR. At least, we can be sure there’s no compatibilist explanation for that judgment! (assuming they understand the case properly)

To be clear, I'm operating on the assumption that this would be a within-subjects experiment in which each person got all four cases in the typical order. If you were operating on the assumption that each person would get only one of the four cases, though, I still think getting the non-responsibility judgment would indicate they endorse something like TNR--I just don't think you'd be very likely to get it.

John, thanks for the Demitriou reference. I'm looking forward to reading it.

[Adam Feltz]

For what it's worth, Tamler convinced me in New Orleans to move beyond the "industry standard." So, our lab has just finished up a first round of studies based on Pereboom's Four Cases (thanks to Tamler for help with these). We have the data and are just trying to find time to analyze them.