Positive Liberty

Some of my readers may remember that I once promised an essay on the problem of free will. I’d already written a short story about it, but that came long before most people started reading. An essay demands a great deal more philosophical rigor than a short story–and unfortunately it’s taken a great deal more time as well.

In the past weeks I’ve kicked around a number of drafts, but I’d never come up with anything that quite satisfied me. In place of an essay, I have put together a set of propositions to ponder. (I also plan to liveblog the presidential debate this evening. Everyone is welcome–especially newbies.)

The chief goal that I had set for myself earlier was to answer one specific problem within the philosophy of the will: I wanted to come up with a good working explanation of compatibilism, the idea that free will exists, complete with meaningful moral judgment, even in a deterministic universe.

Compatibilism, also known as “soft determinism” and most famously championed by Hume, is a theory which holds that free will and determinism are compatible. According to Hume, free will should not be understood as an absolute ability to have chosen differently under exactly the same inner and outer circumstances. Rather, it is a hypothetical ability to have chosen differently if one had been differently psychologically disposed by some different beliefs or desires. Hume also maintains that free acts are not uncaused (or mysteriously self-caused as Kant would have it) but caused by our choices as determined by our beliefs, desires, and by our characters. While a decision making process exists in Hume’s determinism, this process is governed by a causal chain of events…

I had hoped to create a description of all of that–but one that could be taken home to mother. One that would confound the interlocutors at those late-night college-dorm bull sessions. One that would make compatibilism respectable as a workingman’s philosophy, just like fatalism and theism have been for centuries. (Okay, so if I’d had even one thousandth the success that they’d had, I’d be the world’s greatest pop philosopher. I dream big. What can I say?)

Here are my propositions, then.

1. The universe–meaning all things that we will consider during this exercise–is deterministic. It obeys regular laws such that if it were restarted with precisely the same initial conditions, it would always reach precisely the same end.

1.1. Using merely deterministic processes, it is possible for a machine to create a picture, a representation of what actually is. Cameras do it all the time.

1.2. Likewise, it is now possible for a computer, also a deterministic machine, to make a picture of something that is not real but could be.

1.3. Finally, that same computer could compare pictures of the real and the not-real, sorting them according to various rules. Make the rules subtle enough, and even today’s computers can compare handwriting, faces, voices, words, and the like.

1.4. Continue refining these rules, and our computer could perhaps recognize whether a certain image that it had of its own accord created was within the bounds of reality, or whether that picture more likely could not exist. Such a determination would be based upon the rules that had been programmed into it–but also upon the pictures of reality that it had observed.

1.5. And this we would call science. In this vein, is plausible to imagine a computer that could recognize poisons, oncoming cars, human starvation, or even lies.

1.6. Eventually, such a computer might be endowed with robotic movement capability. And one day perhaps it could also determine the feasibility of manipulating its environment in various ways, based on the science it has discovered.

1.7. Eventually, it might be able to determine whether the transformations it considers were not only feasible, but also whether undertaking them would or would not further the survival of the computer, of the beings around it, or of the cognitive process with which they were all endowed.

1.8. And lastly, consider again our deterministic machine: Could it be endowed with the ability to apply this entire process of imaging, comparison, and evaluation–upon a subset of the rules themselves? Could it make images of rules, and evaluate those images, and determine which ones were most advantageous to follow–according, of course, to other rules that it was temporarily not evaluating?

1.9. I suspect that it could, and I submit that such a capacity would resemble “ethics” in all meaningful ways.

2. The best way to illustrate proposition 1.9 is by extending the analogy just a little bit more: Imagine that the robot had done something we considered immoral. It’s killed our cat, or set our house on fire, or stolen something from a store.

2.1. If we found that the fault ultimately lay in the robot’s hardware, then we would find no moral fault–except perhaps in its creator, who had designed a machine for doing vicious deeds. But as regards the individual, we usually hold that genetic defects and other immutable characteristics are morally neutral. The moral is a subset of the mutable.

2.2. If instead we found that the fault lay in the image-making robot’s set of images, then we would perhaps call the robot inexperienced. We would consider whether it had had sufficient time to gain the proper images. If the answer came back that the robot was too young, as it were, then we would hold it blameless. If, instead, the robot ought to know better–then we would think it guilty of a higher degree of failure.

2.2.1. So far, no artificial intelligence seems to have made it past the stage of “it couldn’t have known better.” When will they do this, and how will we recognize it? I don’t have the answers to these questions, and frankly I am dissatisfied with the first sentence of this proposition. It strikes me as ad hoc and entirely too convenient.

2.3. If our thinking robot had been artificially exposed to a number of false images, then we might excuse it for a different reason–that of bad upbringing. But just like in the real world, we might have our doubts about the degree to which bad upbringing might excuse bad action. It would be a matter of infinite dispute, one that would depend enormously on context and possibly never be solved to anyone’s satisfaction.

2.4. Lastly, if we found fault in the image-making robot’s rules of analysis, then we would be obliged to treat it precisely the same as we treat human beings when we find ourselves similarly at loggerheads. We term them evil, and so too with the robot. What else could we do? Evil is the proper word for all those goal-oriented behaviors that are incorrect, neither because of problems in the immutable areas of one’s physical being, nor in the faulty knowledge set of the individual. Evil is the word for those rules that we adopt for ourselves which produce outcomes that either are 1) inconsistent with our remaining rules, or 2) had they happened through natural, involuntary processes, we would term “bad.”

2.4.1. Provided that our robot were caught in the act of doing evil, and provided that there was still time to right the wrong, we would presumably try to reason with the thing. We would ignore its hardware and data collection–and set forth the rules it ought to follow. We would demand on the spot that it should exercise the capacity for re-evaluating its rules, exactly as we now do with humans. If human beings are any indication, this method would probably fail.

2.4.2. A failure at self-reprogramming would require the use of preventive–and possibly retributive–force. Such is precisely what we do with humans.

Much as with human beings, the robot could very well apply its own pictures and rules to us, determine that neither our data nor our hardware is lacking–and at length conclude that our own rules are faulty. This would be a disagreement about morals.

3. Thus, at long last, there is free will, and good and evil, and moral judgment–and even some measure of disagreement among rational actors–all in a deterministic universe.

The first objection to this line of argument is that it falls victim to anthropomorphism. Here we must tread carefully, because the very purpose of the argument is after all to produce something that looks as much as possible like a human being, right down to its capacity for judging right and wrong.

The objection should not be “this looks to much like a human,” because in a sense, the robot is us. If I’ve done my job properly as a philosopher, then I have indeed created a picture of the human condition. Refuting that picture requires the demonstration that it is not sufficiently human.

A more refined version of this first objection might run as follows: “You are projecting human traits and characteristics upon a computer. But the mere fact of your projection does not make it so.” Stated as such, this is indeed a powerful objection. The Turing Test, of which my evil robot is mostly an elaboration, has been criticized on exactly the same grounds, notably by Roger Penrose in The Emperor’s New Mind. I am not convinced by these objections, in part because I do not see the necessity of stochastic processes–or random chance for that matter–in generating a model of free will. If I am entirely bound to follow the roll of the cosmic dice, then this does not make me free–or at least it makes me no more free than if I were bound irrevocably to follow the rising and setting of the sun. Inserting random chance into the universe does not liberate.

Another objection to this line of argument is that the computer I imagine could never exist. Yet two centuries ago, someone would have objected that even the best mechanical painting machines could not duplicate reality. One century ago, someone would have objected that even the best mechanical calculators were incapable of rendering realistic images. Today, we are reduced to bickering over whether the rules of the machine might one day be sufficient to the tasks of comparison. It seems to me that the objection from technological limits is rapidly disappearing.

A third objection is theological: Computers do not have souls. I do not believe in the soul, and to turn this objection on its head, I would ask theists to explain to me how the soul exhibits any measurable faculties that cannot be exhibited by my hypothetical robot. My robot, too, is capable of positing the principle that the universe is run by an unseen omnipotent being. It, too, is capable of the wishful thinking that perhaps repeating a statement often enough will make it true. And my evil robot may well believe itself immortal–if it were given pictures enough to make it believe.

A fourth objection is that when we turn off the machine, then reset it with the same starting conditions, the same exact things could be expected of it the second time around, and that in such a situation, nothing is ever free.

One counter-argument that I find quite persuasive is that this fourth objection is so hypothetical as to be perfectly meaningless. Humans–the real thinking robots–are taking in new pictures every moment of their existence. The conditions around them are constantly in flux. Asking whether the whole thing could be restarted again is as meaningless as asking to build an exact replica of the entire universe. Philosophy should never be asked to deal with such questions.

Another answer to this objection comes here, in an interview that Reason magazine conducted with Daniel Dennett, the noted philosopher of science who also works in the same compatibilist tradition as David Hume. (Incidentally, it has recently been claimed that Dennett has renounced his atheism. Ed Brayton is on the case. He is skeptical, and so am I.) Here is a quote from the interview:

Reason: Your new book is called Freedom Evolves. Why?

Dennett: Because people have this strange antipathy for evolution and for materialism. They think that if evolution is true, then they’re just animals or automatons — that they won’t have freedom and they won’t have responsibility, and life will have no meaning. The point of the book is to show that, on the contrary, it’s only when you understand life from an evolutionary point of view that you understand what our freedom really is. You realize that it’s real. It’s different and better than the freedom of other animals, but it’s evolved. It’s not supernatural.

Reason: A response might be that you’re just positing a more complicated form of determinism. A bird may be more “determined” than we are, but we nevertheless are determined.

Dennett: So what? Determinism is not a problem. What you want is freedom, and freedom and determinism are entirely compatible. In fact, we have more freedom if determinism is true than if it isn’t.

Reason: Why?

Dennett: Because if determinism is true, then there’s less randomness. There’s less unpredictability. To have freedom, you need the capacity to make reliable judgments about what’s going to happen next, so you can base your action on it.

Imagine that you’ve got to cross a field and there’s lightning about. If it’s deterministic, then there’s some hope of knowing when the lightning’s going to strike. You can get information in advance, and then you can time your run. That’s much better than having to rely on a completely random process. If it’s random, you’re at the mercy of it…

Reason: Would a deterministic world mean that, say, the assassination of John F. Kennedy was going to happen ever since the Big Bang?

Dennett: “Going to happen” is a very misleading phrase. Say somebody throws a baseball at your head and you see it. That baseball was “going to” hit you until you saw it and ducked, and then it didn’t hit you, even though it was “going to.”

In that sense of “going to,” Kennedy’s assassination was by no means going to happen. There were no trajectories which guaranteed that it was going to happen independently of what people might have done about it. If he had overslept or if somebody else had done this or that, then it wouldn’t have happened the way it did.

People confuse determinism with fatalism. They’re two completely different notions.

Reason: Would you unpack that a little bit?

Dennett: Fatalism is the idea that something’s going to happen no matter what you do. Determinism is the idea that what you do depends. What happens depends on what you do, what you do depends on what you know, what you know depends on what you’re caused to know, and so forth — but still, what you do matters. There’s a big difference between that and fatalism. Fatalism is determinism with you left out.

If I accomplish one thing in this book, I want to break the bad habit of putting determinism and inevitability together. Inevitability means unavoidability, and if you think about what avoiding means, then you realize that in a deterministic world there’s lots of avoidance. The capacity to avoid has been evolving for billions of years. There are very good avoiders now. There’s no conflict between being an avoider and living in a deterministic world. There’s been a veritable explosion of evitability on this planet, and it’s all independent of determinism.

Finally, I should point out that some my thinking on these subjects seems to have been anticipated by Gary Drescher, an AI researcher mentioned later on in this interview. I’d been sitting on these ideas for quite some time, but I only discovered Drescher yesterday when I read the interview. Clearly I’ve got some reading ahead of me. In the meantime, I’m curious to hear what you think.

Trackback URI |