Crime, Neuroscience, and Free Will

Over on the new Law and Neuroscience Blog, Thomas Nadelhoffer has a post that might be of some interest to Gardeners.  He concludes with a series of questions that will probably need to be answered in the years to come.

Thomas wonders,

As we become increasingly better at predicting violent recidivism, drug relapse, and the like, should this affect our basic notions of agency, free will, and responsibility? Assume, for instance, that for a certain sub-class of offenders-e.g., psychopaths with substance abuse problems-we are able to predict that 90% will violently reoffend within two years of being released. Should this change our views concerning their free will and autonomy? If not, why not? If so, why? If we keep inching our way towards 100%, is there a point at which we should conclude these individuals are no longer free?

I'm inclined to think that our ability to predict such behavior shouldn't wholly undermine responsibility.  After all, compatibilists are committed to the claim that 100% predictability doesn't undermine responsibility--and not just 100% predictability of some action type (e.g. violence against others), but 100% predictability of some action token (e.g. violence against that person at that time).  Moreover, libertarians like Kane need not change their views.  If the reason we can predict recidivism so well is because the agents in question have  formed themselves into violent individuals, then mere prediction won't undermine their responsibility.  Of course, we might worry if (1) we could predict action tokens with perfect accuracy, and (2) the reason we could do so was because the relevant mechanisms producing behavior bypassed our powers of reflective self control.  But I seriously doubt neuroscience can satisfy (1) or (2).

In any case, you should head over to the Law and Neuroscience blog with your thoughts.

Comments

[Fabio Fang]

Hi.
You said: "After all, compatibilists are committed to the claim that 100% predictability doesn't undermine responsibility". Maybe some compatibilists are commiteted to that claim, but others aren't, or at least don't have to be: after all, it is arguable that determinism doesn't entail predictability. Right?
Of course, determinism implies predictability *in principle*, but I take it that you meant predictability *de facto* and not in principle.

[Eddy Nahmias]

Fabio is right that determinism certainly does not entail predictability in practice, so the "dream" of 100% predictability in the real world is impossible. The amount of information required to make such predictions grows astronomically larger the farther in the future you go (as wide as the light cone extends!). There's no way to predict with certainty that the future "criminal" won't get hit by a bus before s/he commits the crime (unless you control the situation).

But it's not even clear that determinism entails predictability *in principle* (unless you just define it in that way). There is a deterministic interpretation of quantum theory (Bohmian) that rules out predictability in principle, and there are some interpretations of (deterministic) chaos theory (non-linear dynamics) that rule out predictability in principle. I think both cases apply only to predictive agents within the predicted system, rather than an external God or something. And there have also been philosophical arguments against such prediction by an agent within the system, right?

In any case, I see little reason to think predictability (to the extent it is possible) should re-shape our theories of free will and responsibility. Given the impossibility of predicting specific crimes ahead of time, I think the relevance of these fascinating new actuarial methods is (a) to offer to the potential criminal ways to avoid doing harm (e.g., voluntary treatments) and (b) to raise the question of whether "risky" people should be watched more closely (without violating their Constitutional rights).

[Andrew Jaeger]

Also, (regarding (1)) isn't it the case that the predictability of action tokens made rather difficult given the influence of external stimuli that do not enter into the probabilistic outcome of an action token...
It seems at best that what could be predicted is a tendency (or inclination) to act in such-and-such a way...
For example, it seems that I could be predicted (with 100% certainty) to relapse on drug abuse.
But this seems impractical given the many external causes that do not enter into the calculation regarding whether I will perform action a [e.g., use drug x at time t].
Perhaps my mom grounds me from leaving the house and several years go by and I loose my addiction to drug x...my mom's grounding me did not, nor could(?) enter into the probability calculation of whether I would reuse drug x].

[Justin Coates]

Fabio,

I meant predictability in principle, but I see now that I wasn't clear about that. That's why I'm skeptical that neuroscience could make predictions of the sort that Thomas is wondering about (e.g., it is 100% likely that S will X).

Eddy,

I'm happy to concede the point that it is possible for a deterministic system to be in principle unpredictable. Of course, I suspect that when most of us became compatibilists, we had a deterministic system more like Leplace's in mind. And unless there is a compatibilist who claims that FW/MR are compatible with determinism only if the determinism in question precludes in principle predictability, it seems like she is would still be committed to the claim that Leplacean determinism is compatible with FW/MR (even though all human actions are predictable in principle on this model).

Your other points are quite good, and as far as I can tell, friendly to my skepticism about neuroscience's ability to make such predictions.

Andrew,

As far as I'm understanding you, I think that's right. And it's a further reason for thinking predictability of action types shouldn't undermine our responsibility for action tokens. I was trying to point out that even predictability of action tokens wouldn't necessarily undermine our responsibility, but for reasons you and Eddy have suggested, we might not even need to worry about this sort of predictability (at least from neuroscience) at all.

[Kip]

Eddy,

I have some comments about predictability, in response to your thoughts.

1. Your comment seems to generally say "well, it seems that there is a pretty good chance that were not predictable after all."

What if we are? In other words, your comment seems to speculatively hope that we're not predictable. Well, what if that hope gets dashed? Would that affect you views on free will?

Suppose, for example, that in the future, I can be implanted with an extra chip in my brain that makes me very smart - 1000 times smarter than those who don't have the chip. Suppose that you don't have the chip. Suppose that I'm now smart enough to predict your actions with 99% accuracy, even though you satisfy the Compatibilist Conditions of your Choice (CCC).

In other words, hoping that we're not predictable avoids the real question: is there a conflict between free will and predictability. I'd be curious to know if you think there is.

Personally, I think free will originated, and is mostly understood today, as the power that lets us be unpredictable, that lets us be the ones who decide our own futures, and that that power is exclusive to us - nobody else decides for us. In other words, when people threatened that "well, God just set up the whole universe in a deterministic way so that your whole life unfolds exactly as he planned" the retort was supposed to be "no way, because I have free will."

Of course, if I had read Dennett before thinking about free will, maybe I would have understood it to mean what Dennett means: the rational capacities for deliberation and moral choice that nobody denies human beings possess. But I doubt it.

So I think free will is intimately and fundamentally related to unpredictability. I think that is part of the huge appeal of libertarianism, and a major deficiency of compatibilism. Strangely, I rarely hear compatibilists ever talk about unpredictability - it's an aspect of the concept that (it seems to be) gets swept under the rug.

2. This leads to my second point: your comment emphasizes that human beings might be largely unpredictable in practice. In other words, it might be very, very difficult to predict human behavior, so let's not worry about it.

I think this misses the real, underlying concern: we don't want to be predictable *in principle*. The fact that it might be very difficult to predict our behavior in practice is, therefore, largely irrelevant. What is disturbing about predictability is not the thought that some other human might be exploiting my predictability to steal my credit cards. At least it is not just that. What is disturbing about predictability is the thought that we are, at bottom, the kind of creature that is predictable. We don't want to think that somebody could know at T1, what I would choose at T2, because we want to believe that the choice originated with us. Back at T1, there simply was no fact of the matter as to how I would choose, because I hadn't decided yet. It was up to me, and then I decided. If we're predictable, then that conception of ourselves as exclusively originating our choices, decisions and life stories is false. That's part of the worry about predictability, no matter how impractical the predictability is, and that's part of the tremendous appeal of "free will."

[Eddy Nahmias]

Kip, above I was not trying to say anything one way or another about the conditional, "If we were predictable in principle, that would mean we lack free will" (though I could see how you might draw such inferences). Now I will. I think the answer is that the conditional is false. I also think it's entirely unclear that most ordinary people think the conditional is true--after all, many people seem comfortable with the idea that God can know everything we will do (and most people in my surveys answer that free will and MR are possible in a universe with a perfect predictor).

Having said that, I also agree that unpredictability is very attractive for lots of reasons: e.g., it prevents certain types of control by other agents, it makes life interesting, it makes us able, in some cases, to know ourselves (and what we will do) better than anyone else does. However, here I'm inclined to think unpredictability in practice is sufficient to get us what we want, though we still tend to get disturbed if we learn we are more predictable regarding *types* of actions we will perform than we'd like to be/thought we were.

Finally, if, as I suggested above, determinism does *not* entail predictability in principle (much less, in practice), then I think we should not move from the conditional, "If we were predictable in principle, that would mean we lack free will" (even supposing it is true), to the incompatibilist conditional, "If determinism were true, that would mean we lack free will."

[tnadelhoffer]

Eddy,

If (a) determinism were true, (b) you had complete knowledge of the laws of nature, and (c) you had complete knowledge of the physical state of the universe, then (d) everything would be predictable, in principle, no? I think what you are talking about is limited to predictability in practice. I, for one think the truth of determinism--used here in the standard philosophical way--*does* entail 100% predictability in principle. If you disagree, I would like to hear an argument that concedes the truth of (a), (b), and (c) but which consistently denies (d).

As for the kinds of predictions made possible by state-of-the-art actuarial assessment tools such as Quinsey's VRAG or Monahan's ICT, they are not made at the individual/token level. They are made at the group/type level. For instance, if you are (a) a psychopath with a Hare PCL-R Score of 33+, (b) a drug addict or alcoholic, and (c)guilty of a previous violent offense, you are a member of a group such that 85% of the people in this group will commit a violent act in the future (the % is meant to be merely illustrative not an accurate statement concerning the actual predictive power of these tools). I take it this is why folks prefer clinical risk assessment to actuarial assessment--namely, that clinical predictions of violence are made at the individual level whereas actuarial predictions are made at the group level.

p.s. Needless to say I wish the Gardeners had decided to have this discussion over at the other blog rather than here!

[Mark Smeltzer]

Tom,

I laid out an argument that accepts (A), (B) and (C), yet denies (D) in my comments on this post. The argument only applies to worlds that are chaotically deterministic.

Moreover, the conclusion of the argument does embody a bit of semantic word play: the argument is that it would be possible, for argument's sake, to produce an exactly similar replica of a chaotically deterministic world in a massive supercomputer and run it forward from a known state at a faster than reality time rate. This would, in principle, allow you to see facts about the "future" in the represented world.

However, I am resistant to calling this a prediction. I would rather say that we have two worlds which happen to be exactly similar, and one of those worlds is running ahead of the other. I do not think it is clear that this is what we mean by "prediction".

Secondly, I am skeptical that such an experiment is logically possible. Where would this device be constructed? If the second world is constructed within the first world, then the second world necessarily will not be exactly similar to the first world -- due to a viciously infinite regress (the second world would at some point have to also contain a replica within itself, and this third world would follow suit, etc., etc.).

Moreover, if the future state of the replica world were observed within the host world, it would necessarily alter the state of the host world in ways that the replica did not suggest.

So, I believe that if the replica world were created outside of the target world -- completely outside the target world's space-time continuum, then the replica world reliably would be exactly similar to the target world. However, it is essential that no facts ever be communicated between either world, which means that they must be completely causally isolated from each other.

If the replica world is completely isolated from the target world, how would this allow for in principle prediction about the target? Only a casually isolated observer could use the replica to learn facts about the future states of the target.

So, what this worry boils down to is this:

• I am suspicious about calling facts about the present state of an exactly similar, but otherwise sped up, duplicate of the target world "predictions" about the target world's future.
• Even if we accept these facts as predictions, the only situation they possibly apply to is a causally isolated observer.
• I think we have something else in mind when we invoke worries about predictability.

To put it another way, if the question were, "Could a God-like observer know facts about the future state of a chaotically deterministic world?" The answer has to be yes. Is this what we mean by predictability?

My feeling is that when we invoke worries about predictability, they are rooted in natural resistance to the idea of logical determinism or fatalism. Since a chaotically deterministic world is not logically deterministic, that may be sufficient reason to dismiss such worries regarding such worlds.

[Justin Coates]

p.s. Needless to say I wish the Gardeners had decided to have this discussion over at the other blog rather than here!

This is no doubt my fault. I apologize.

[tnadelhoffer]

Justin,

Of course it's not your fault. It's no one's fault. But then again, I am a free will/MR skeptic, so I am never looking to spread blame around :) I was merely pointing out that you initiated an interesting comment thread and it would have been nice if it occurred at the other blog instead.

[Eddy Nahmias]

Thomas, you ask: "If (a) determinism were true, (b) you had complete knowledge of the laws of nature, and (c) you had complete knowledge of the physical state of the universe, then (d) everything would be predictable, in principle, no?"

No. It depends on who "you" is. If "you" are part of the universe itself, then I think there are cases where a-c do not entail d. For instance, the Bohmian interpretations of quantum mechanics is deterministic but allows (just as the Copenhagen interpretation) only probabilistic predictions (and I emphasize here again that there is every reason to think these probabilities of quantum mechanics, whether metaphysical or epistemic, have real effects at the macro level). But I think there are other interesting cases where predictability in principle might fail, including when the predicting agent is predicting its own future states or where the information in c approaches infinity.

So, maybe we should say that determinism means that a-c entail d if the "you" is outside the universe (or the system being predicted). Well, there are some problems about how this "you" would get the information while counting as distinct from the system (analogous with the problem of interactionism). The "you" could always be God--God can fill in whatever blanks we need, right? But if you posit an omniscient God to do the work, it is unclear why everything is not predictable in principle even if determinism is false. Surely, we can imagine God predicting the future without having to read it off from the present and the laws.

Again, all of this suggests that *if* determinism is supposed to be a threat to free will (or MR), we should not think so because of worries about predictability.

(Of course, all of this metaphysical jibber-jabber has little to do with the real-world issue you raise about the sort of predictability new methods might allow regarding the likelihood that people with certain *types* of backgrounds or brains will perform certain *types* of actions. Right?)

[Kip]

Eddy,

The predictor doesn't need to be divine, and the prediction doesn't have to be magical.

For example:

Suppose the entire known universe is running on an alien's very fast laptop. That is, the known universe is running inside of a very fast computer, the same way that a game of Quake or Halo runs inside of a computer.

The person running the laptop doesn't need to be divine. The laptop doesn't have to be infinitely fast. And there are no problems with interactionism, any more than there are problems with an Xbox player figuring out what state the game is in.

To predict what would happen, you could:

1. run the game through once, and then know how it will play out any future time;
2. use a faster computer to run the same determinism algorithm; or
3. use a faster algorithm.

All of these seem possible, at least in principle. It just seems to me that you're too quick to dismiss worries about predictability.

And I'll repeat my general points from above:

1. even if predictability in principle is unlikely, we should still address that question, because it is interesting in its own right (hoping that some interpretation of quantum mechanics makes us unpredictable seems to duck the real philosophical issue here); and
2. the worry is not about predictability in practice, the worry is about predictability in principle. Even if the predictability is very difficult, we do not want to believe that we are the kind of creatures whose life stories can be known before we make any life choices.

[Mark Smeltzer]

Kip,

Just so you know,

• Quake (and all other video games) cheat massively. I should know since I've work on the physics in some of them, and continually review the source code for many more. For instance, with Quake 2 I helped introduce a patch to the game's physics engine that allowed items to move more realistically: the game world moves in 0.1 second "pulses" and during those pulses, the game was only making one adjustment to each particle per pulse, and if a particle were to collide with another particle, it would normally do so before it had a chance to move the entire distance it would have moved if there had been no collision. However, the game did not account for this "lost" time. So, in an effort to make the game more realistic (we weren't sure up front if there would be any tangible benefit), I helped revised the system to allow a particle to moving to its full potential by performing a loop until it had used up its displacement potential for each timeslice. The net effect was that things bounced MUCH more realistically and the game took on an incredibly better feel (most modern games now use a similar technique).
• There is no such thing in existence that can be labeled as a physics emulator (only simulators so far, and no reason in principle to think that will ever change).
• Interactionism is relevant if the simulator is supposed to interact with the world it is simulating -- there will necessarily be information that the simulator does not have.
• In chaotic systems, predictions will vary wildly with even the slightest miscalculation at the 10th decimal place (which should allow you to infer that 99.999999% of the climate models distasterites feed us are based on pure fancy).

If you posit that the alien and his laptop are outside the space-time of the target universe, and the alien is able to "access" the state of the target universe introducing causal effects into that universe (Heisenberg's uncertainty principle is about prediction in *principle*), then the rest of what you say follows. But then, that's the same thing that Eddy and I said ;)

[Kip]

Mark,

I'll respond to your points in turn:

1. It's not clear to me that the game system "cheating" prevents it providing the analogy I suggest. Further, it's not clear that the universe doesn't also "cheat" in a similar way. Lastly, do you want to hinge your defeat of my argument on your current estimation that the universe doesn't work that way? What if, it turns out, it does?

2. No comment.

3. A person interacting with a computer running Quake or Halo can access every single bit of information about that system. It is all in the hard drive. There is no interactionism problem whatsoever. And it seems that our universe could work in a similar way (again, do you want to hinge your position on your current guess that the world doesn't work this way? What if it does?).

4. Who suggested that anything was a chaotic system? Are you suggesting the universe is? How would you prove that? Are you willing to hinge your position on that?

[Mark Smeltzer]

Kip,

1. My point about the fact that Quake cheats to simulate physics is important only insofar as you want to setup an analogy about it and the real world -- because Quake relies upon real world physics in order for it to exist: to move the electrons around on the hard disk, for example.

3. Future facts in game of Quake may seem predictable, but it depends on what you want to predict whether it is. If you are talking about the future state of the game world in which the player is playing, then it is de facto unpredictable because the game world does not contain all of the information. The interaction between the player and the game world creates new information, which changes the future state of the game world.

4. I would be happy to assert that the actual universe is indutibitably a chaotic system. This should be obvious since chaos was discovered why observing complex systems that actually exist in nature, such as fractals. Of course there are possible logically worlds that are deterministic and non-chaotic, but that doesn't mean the two types can be treated equally.