Free will blogging
Kevin Drum wonders whether hyperintelligent machines might undermine people's belief in their own free will*:
With this in mind, here's another possibility for what happens after we create fantastically advanced computing capabilities that are thoroughly merged with human consciousness: we discover — in a way that's truly convincing — that free will doesn't exist. And so we give up. Within a few decades, the human race chooses to put itself out of existence because there's really no point to its continued survival and our biological urge toward self preservation, honed over millennia by evolution, no longer controls our merged biological/machine selves.
The thing is, we've already got hyperintelligent machines in our heads. They're called "brains." These machines have already formulated some pretty convincing arguments against free will. It's impossible to say whether any super-computing prosthetics could provide decisive proof that free will doesn't exist. That depends on what you imagine these machines will have that ordinary thinkers lack.
You also have to make a lot of philosophical assumptions about what the problem of free will is before you can start asking whether a hyperintelligent-machine-mind could solve it any better than we could.
Furthermore, in order for Kevin's scenario to play out, we humans would have to be in a position to recognize and accept a definitive proof that free will doesn't exist. Philosophers have been generating very strong arguments against free will for thousands of years. But whenever we're presented with one of these stumpers, we assume there must have been some mistake and redouble our efforts to refute it.
Suppose that hyper-intelligence creates minds so powerful that they become LaPlacean demons. A LaPlacean demon knows where every particle of matter is in the universe and all the physical laws that govern transitions from one state to another. Therefore, in a deterministic system, the demon can predict every physical phenomenon, including human behavior.
Would the existence of LaPlacean demons undermine people's faith in their own free will? Logically, the existence of these creatures wouldn't change the status of our belief in free will. If you doubt that free will is compatible with determinism, it doesn't make any difference whether there is anything powerful enough to predict your behavior, all that matters is that you behavior is determined by prior causes.
So, I don't see why hyperintelligence is more likely to discredit free will than ordinary intelligence.
NB: The Stanford Encylopedia of Philosophy has an excellent introduction to free will.
*Kevin isn't arguing that super-intelligence would inevitably undermine faith in free will. Kevin's main point in the original post is that Ray Kurzweil's optimistic projections about the effects of AI aren't foregone conclusions.
This was more a fun post than a real argument, but FWIW, my thesis depended on Kurzweil's belief that humans will merge with super intelligence machines sometime in the next few decades. So the machines don't have to convince us verbally. Once we're merged, we'll believe what our machine implants believe as strongly as if we'd thought it up ourselves.
As for whether free will really exists or not, that's a very good question, isn't it? If it does, then presumably our super smart computer descendents will figure that out too.
Posted by: Kevin Drum | October 04, 2005 at 12:32 PM
Well, in fairness, I don't think Kevin was seriously arguing that those events would necessarily happen; he was saying that even if you accept Kurzweil's premise -- that superfantabulous computing/thinking machines are years away -- his conclusions don't necessarily follow, and any number of things could potentially happen. Many of them aren't especially good. Why Kurzweil thinks he is particularly well suited to peering into a singularity and seeing out through the other side is something of a mystery.
Of course, Kurzweil's whole argument is badly done, as has been well documented elsewhere, but (as you've shown here) that doesn't mean that considering the ramifications of intelligent machines is fruitless.
Posted by: Jonathan Dursi | October 04, 2005 at 12:34 PM
It's an excellent question. I'm just not sure we're likely to get dramatically better answers than we already have.
If we did get better answers, I suspect our super-smart descendents would react more or less the same way we do to strong arguments against free will--shrugging them off, revising our initial assumptions to hold onto free will, redoubling our efforts to disprove any putatively definitive proof.
The proof would occur to some super-brained individual, but in order to have massive cultural effects, he or she would have to recognize and popularize that insight. So, the machines wouldn't have to convince the lone super-philosopher, but he or she would have to sell the idea to everyone else.
Posted by: Lindsay Beyerstein | October 04, 2005 at 12:44 PM
Lindsay wrote:
"If you doubt that free will is compatible with determinism, it doesn't make any difference whether there is anything powerful enough to predict your behavior, all that matters is that you behavior is determined by prior causes."
I think Kevin's point is that the supersmart/LaPlacean computer demons will switch people to this point of view, not how people who already have this point of view will react to the new evidence. As to convincing people, it shouldn't be hard, by proving to people that the computer knew what they would do before they "made" their decision.
I would guess there's still a big gap between even the AI that Kurzweil is talking about and a LaPlacean demon, however.
Posted by: Brian S. | October 04, 2005 at 01:05 PM
But do we not already have a 45 decade experiment in the popular rejection of free will? That humorless frog was exiled to Geneva in the first half of the 1500s, and convinced the majority of European Protestants to accept predestination and reject free will; to admit that life was pointless, humans were worthless, and most of them would live in a God-made concentration camp for all eternity. They lived, reproduced, and if you believe Max Weber, were responsible for the success of capitalism.
Dostoyevsky engaged in the same thought experiment a century ago, and concluded that, when a person's actions could be worked out by reference to a chart or calculation, he would do precisely the opposite, just to be contrary. No scientist but like the great Russian authors a keen observer of people, Dostoyevsky pointed out another hole in the logic: the impact of the measurement and knowledge on the actor.
Predictions that, given too much information about our destiny, we will lay down and die are premised on the unsupportable conclusion that humans are creatures of reason rather than creatures that reason.
Posted by: Thomas | October 04, 2005 at 01:59 PM
"As for whether free will really exists or not, that's a very good question, isn't it?"
I don't think so. I don't think it's any more sensible than the question of whether morality "really exists." I plainly does exist, just by the fact of people acting as though it does and talking about it as such. What more do you want? A description of it in terms of particle physics? What for?
Also, doesn't the absence of determinism in physics pretty much show that arguments and thought experiments about "free will vs. determinism" are badly posed? Nobody still thinks that the Unabridged History of the Universe is even in priniciple predictable from a complete description (in whatever terms) of subatomic start-conditions, do they? I guess what I'm trying to say is, the concept of a LaPlacean demon seems incoherent, (eg., does it predict it's own future states?), and based an outdated, mechanistic conception of the universe.
I guess I just can't take seriously the idea that there is a problem of free will, but if anyone (especially Lindsay) feels like describing why they feel it is a problem, I'd love to read it.
Posted by: Scott PM | October 04, 2005 at 02:20 PM
I don't think that the painless extinction of humanity would be such a bad thing (I don't really care one way or the other, though naturally I wouldn't like the painless extinction of everyone except me), but I think that what you mean by free will. If you mean some form of agent-causation, well, that was never really easy to understand (which isn't to say that it isn't true, though it isn't, but it's hard to understand regardless of its truth), so I doubt many people believe that (maybe I'm transposing my own prejudices on others here, and I'm uniquely bad at understanding this sort of free will and others grasp it more easily). If it means "the exact future isn't mapped out from the first moments of the universe," then I don't think it would be an abuse of quantum mechanics to say that uncertainty really does make that form of "free will" plausible, though having your behavior determined by a roll of the dice seems no more like free will as used in the common language than having it determined by a rigid set of non-random rules.
I think many people think that they're talking about agent-causation when they talk about free will, but when they get into the specifics of what could actually refute free will, they mix up agent causation and everything except rigid, non-probabiblistic determinism. (i.e., arguments for free will based on quantum mechanics)
Posted by: Julian Elson | October 04, 2005 at 02:23 PM
Thanks for posting about this, Lindsay. Very interesting stuff.
"Once we're merged, we'll believe what our machine implants believe as strongly as if we'd thought it up ourselves."
Sorry, but this wording really bugs me. "Our machine implants" don't "believe" anything, it's *us* that's doing the believing.
There's two things you might mean by this. One is that the machine implant is "something else" or "not us" and therefore "its" thoughts can be spoken of separately. But unless you want to make the case that incremental modification and enhancement of the human mind somewhere introduces a discontinuity of personal identity--and the question's open--then you can't really make statements like this honestly. To just assume it is wrong, I think.
Second, you might mean that one's non-biological mind could believe one thing and one's biological mind another, implying two different consciousnesses. I don't think that would really ever happen to any significant degree with machine enhacement of intelligence. On the contrary, I imagine our current problems with non-unitary intelligence (our perception of consciousness as unitary is largely an illusion, remember?) would be decreased with machine enhancment instead of exacerbated.
"If we did get better answers, I suspect our super-smart descendents would react more or less the same way we do to strong arguments against free will--shrugging them off, revising our initial assumptions to hold onto free will, redoubling our efforts to disprove any putatively definitive proof.
"The proof would occur to some super-brained individual, but in order to have massive cultural effects, he or she would have to recognize and popularize that insight."
When everyone has a trillion times more intelligence than they do now, arguments about free will and such will be very obvious, elementary, and non-controversial. The proof wouldn't "occur" to a super-brained individual, it would be as clear as day the moment they set their mind to thinking about it. There might be controversy and questions about the issue, but they would be so far above the level of what we're familiar with now that we couldn't even recognize the arguments. Any questions we're even capable of formulating now will be answered with little doubt.
Super-enhanced minds will be able to understand themselves in the same way that a programmer understands the computer program she writes. This level of understanding, (as far as I can see with my non-enhanced mind,) would seem to lead to a much different worldview. For one, questions of morality and purpose become either meaningless or tautological. I think Kevin is right to try to approach it from a systems perspective, though. Instead of viewing post-humans as human, view them as unpredictable, chaotic systems. Will this particular system end up inevitably destroying itself?
To really get a grip on this problem, I think, you have to step back a little and take a different perspective. I'm going to write this up in a new comment and continue.
Posted by: pdf23ds | October 04, 2005 at 02:25 PM
"Also, doesn't the absence of determinism in physics pretty much show that arguments and thought experiments about "free will vs. determinism" are badly posed?"
Not necessarily. Quantum effects in all liklihood have virtually no randomizing functional influence on the brain at all. One can probably understand a human mind as a completely deteministic system at some level of abstraction and lose almost no information at all. If this is true, arguments from physical determinism still have quite a bit of maneuvering room.
Posted by: pdf23ds | October 04, 2005 at 02:30 PM
"Quantum effects in all liklihood have virtually no randomizing functional influence on the brain at all."
I agree with this.
But even if there were randomizing quantum effects in the brain, it's hard to see why this would tell in favour of "free will" rather than "determinism" -- our supposedly free choices would simply be the result of chance instead of necessity.
My point is that looking for free will in the brain, or any physical system as such, and hence pairing it off against "determinism", is a sort of category mistake. I very much agree with your point above that *people* have beliefs and not their cybernetic implants. I would generalize it to the claim that our brains don't have beliefs (or free will) either. Reduction, in both cases, destroys the phenomena.
Posted by: Scott PM | October 04, 2005 at 03:40 PM
Re: Quantum events and consciousness - Given the complexity of the brain and the extremely nonlinear dynamics of neural phenomena it seems like a major assumption that quantum effects are ignorable. Not that the brain should be considered a fundamentally quantum mechanical system, but for at least some quantum effects there are reasonable mechanisms by which they could affect large scale neural activity.
All of this seems to me only marginally relevant, since I don't really understand the problem. Free will seems to me a meaningless concept based on a mystical approach as much as anything else. Consciousness is either deterministic or probabilistic, and neither option affects what I'm going to wear tomorrow or whether it is OK to eat babies.
Posted by: togolosh | October 04, 2005 at 03:57 PM
Hello,
Quick question regarding the comment "Quantumm effects in all liklihood have virtually no randomizing influence on the brain at all". I remember something regarding this when they used the computer to get the proof of the 4-colour map problem. Some mathematicians argued that there was a probability that a component could've failed resulting in a error in the proof. Flaky, I know but why couldn't you apply this with regards to QM and the brain? I.e. given the exact same conditions as before, can you guarantee the exact same outcome? Simple probability says no.
Regards,
Sanjay
Posted by: SBV | October 04, 2005 at 04:04 PM
I don't agree that determinism precludes individual freedom. Determinism requires that I am constrained by external causes; but it allows that I too am a cause. I believe this agrees with my experience. I am a free cause of my own actions; but my freedom is limited by external causes.
Posted by: Gary Sugar | October 04, 2005 at 04:30 PM
Sort of off topic, but to me the weirdest part of that post was the idea that the human race will "choose" (how?) to commit mass suicide after receiving this definitive answer to a question I'm pretty sure most people don't care about. As a chronically depressed person to whom suicide has always presented itself as an attracive life choice, I can say the possible lack of free will has NEVER been one of the myriad reasons that life is clearly not worth living. It's not like the car not starting.
So, is every one of the eight billion people on earth going to choose this individually of their own lack of free will, or is Kevin supposing that by then we'll have a one-world theocratic dystopia that will do it for us?
Posted by: Disgruntled Attendee | October 04, 2005 at 05:09 PM
My guess is that Kevin Drum was talking about the possibility that the machines themselves could be demonstrated to have no free will--say, because they are running on deterministic computer simulations, so it is possible to run a given mind multiple times from a given initial state and identical sensory inputs, and see that its behavior is identical on each run. If the intelligent machines otherwise behaved in ways just like humans--showing emotions, intelligence, creativity, spirituality, and the feeling of choosing their own actions--this would lend a lot of support to the idea that we don't have free will either. The argument would be even stronger if the intelligent machines were uploads, precise simulations of biological human brains that had been mapped out in great detail at the synaptic level, and we could see that the upload's personality and behaviour was completely continuous with the original human's. I think most reasonable people (read: everyone except religious fundamentalists and new age cultists) would take this as compelling evidence that humans have never had free will or a supernatural soul.
Posted by: Jesse M. | October 04, 2005 at 06:26 PM
"To really get a grip on this problem, I think, you have to step back a little and take a different perspective. I'm going to write this up in a new comment and continue."
Oops. I tried, and it turns out it would take 1000+ words without assuming a bunch of technical AI jargon. Maybe later.
I think Jesse M. is on to something important. The human soul and human free will seem to have a lot in common. Perhaps they're more or less the same thing, historically? I believe that the traditional (i.e. since Medieval Christian philosophy) justification for the idea that humans have free will is that they have a spirit or soul, unlike animals. Scientific naturalism makes the idea of "soul" pretty meaningless, unless you do some radical redefinition, and probably does the same to free will. So why is everyone talking about whether we have free will or not anymore if we're not talking about whether we have an immaterial spirit?
I think it's only because "free will" has a prima facie meaning that doesn't openly conflict with scientific naturalism, so people who hold that worldview don't automatically dismiss the idea as foreign and incoherent when they think about it, as they do with the soul concept. But upon closer inspection, the concept of "free will" turns about to be every bit as incoherent as that of "soul".
As a side note, Ray Kurzweil mentioned in the book his thoughts about spirituality. In his mind, the essence of humanity, and what will become post-humanity, is order and complexity. The more structured, ordered, and purposeful this or that bit of matter is, the more spiritual it is. It's an interesting concept.
Posted by: pdf23ds | October 04, 2005 at 07:14 PM
What? I have no problem with scientific naturalism; but I know for sure that I have an immaterial soul. You must mean something very different than what I mean.
Posted by: Gary Sugar | October 04, 2005 at 07:25 PM
You don't need to invoke QM to have nondeterministic behaviour of the human brain. That said, I suspect it does, at least to the degree that chemisty is fundamentally stochastic, with this randomness having origin in QM. But as far as non-QM nondeterministic systems, even the old chestnut of the gravitational three body problem doesn't have an explicit closed form solution. It can be solved in an iterative fashion using computers, but the solution to it ends up being a very chaotic system with strong dependence on extremely small changes in initial parameters. Likewise with the extremely simple system of a Lorenz attractor. The solution to these things can only be computed stepwise, and cannot be written in a final form that spells the exact state at all times, even though the initiator of this chaotic behaviour can be a very simple set of equations. So, the root equations act somewhat like Schroedinger's equation in QM in that in one sense it gives a complete description of the system, but it's unable to say what the absolute outcome is unless you actually go take the measurement or perform the iterative simulation, whichever is relevant. If you can produce that sort of chaotic system with just a few simple differential equations and three items, just imagine what you can do with 10^15 or so synapses, especially since these synapses are mediated by relase of very small amounts of chemicals, the gating of very small amounts of ions, and other things which operate directly on the scale of QM which is not at all deterministic. You can have a minute quantity of neurotransmitter dump into the synaptic cleft, diffuse across the cleft, hit some receptors, open some ion channels, and whatnot, and have a reasonable expectation of roughly how many ions will go through and what the signal strength is, but you cannot say exactly how much, and these very minor changes in initial conditions could have radically different effects if the system being played out has a degree of chaotic behaviour.
Posted by: Jason E. | October 04, 2005 at 07:27 PM
Here's something entertaining: a nice simulation of a gravitational menage a trois. As you can see, the slightest change in initial conditions gives comletely different results.
Posted by: Jason E. | October 04, 2005 at 07:33 PM
pdf23ds wrote:
What? I have no problem with scientific naturalism; but I know for sure that I have an immaterial soul. You must mean something very different than what I mean.
By "immaterial soul" do you just mean something like a pattern or algorithm that is instantiated by your physical brain, or do you mean that you disagree with the reductionist view that your behavior can in principle be explained completely in terms of physical interactions between the particles that make you up? Or none of the above?
By the way, I posted my previous comment before I noticed that the first comment was by Kevin Drum, so I guess I misinterpreted his meaning. I still think that the evidence of intelligent, humanlike minds running on deterministic computers would make a lot of converts to the "no free will" view though.
Posted by: Jesse M. | October 04, 2005 at 08:34 PM
I seem to remember some comment by Hume, that a person believes in free will because he can remember a decision he'd made, and can imagine having decided differently -- but another person who knows the first person well would have been able to predict that he'd have made the decision he did make.
As far as the feeling of unease about determinism goes, the conclusion I eventually came to was that my thinking and decision-making are real, not illusions -- that part of what determines the course of events in my life is my own process of making decisions.
The idea of free will is a bit scary to me, actually, because I imagine "free" would require that one's mind be utterly distinct from causal relationships. Setting aside the problem of how an immaterial substance could affect a material substance, this notion of "free will" strikes me as the ultimate alienation.
Posted by: FoolishOwl | October 04, 2005 at 09:14 PM
Jesse, by my soul, or mind, I mean my immediate experience, including, for example, my intellect, volition, and opinions, as well as my perception of my body, senses, and imagination.
My certainty of this experience, and my certainty that my experience is distinct from the body that I perceive is way prior to any kind of theory about it like the ones you mentioned. But to avoid any mystery, I agree mostly with guys like Parmenides, Descartes, and Spinoza.
Posted by: Gary Sugar | October 04, 2005 at 09:18 PM
Jason E. said:
"and these very minor changes in initial conditions could have radically different effects if the system being played out has a degree of chaotic behaviour."
Oh, I agree. Whether the brain has choatic features isn't an answered question, and QM interactions aren't necessary for it to be one. But QM doesn't prove that it is.
In fact, I'd go as far to say that we know for sure that a large number of our brain's circuits are definitely *not* chaotic--though whether they're stochastic to a significant degree is a harder question--because they produce results that are so consistent and predictable given the inputs. I don't see how one could possibly say that our visual or aural processing circuits are chaotic, for instance. On the other hand, a significant degree of chaos isn't nearly as hard to imagine existing in the symbolic-processing regions of the brain, so I don't think the existence of non-chaotic visual or aural circuits is evidence against chaos in other parts of the brain.
Jesse M. wrote:
I did? No, that was Gary.
Gary, first, by "naturalism" I mean the belief that no forces outside of the universe enter into its causal chain in any anthropocentric manner; roughly, that all events can be explained by reference to physical matter and energy alone. (Wow. Even "naturalism" is a kind of fuzzy term. What if other universes in the multiverse can sometimes affect this universe? If one believes that is one still a naturalist? I would think so.) Second,
Gary said:
"Jesse, by my soul, or mind, I mean my immediate experience, including, for example, my intellect, volition, and opinions, as well as my perception of my body, senses, and imagination."
So, basically, you're thinking that the existence of qualia is evidence for the existince of the existence of a immaterial soul?
I can conceive of a "soul" taking this role in two ways. At the very least, it would be some sort of passive receptacle or inhabitor of the human body that receives the experiences, not participating, except as an endpoint, in any causal chain of any physical matter in this universe, affected by the universe but not affecting it at all--a non-participating observer*. I think this is a tempting view, probably compatible with naturalism, but probably ultimately a rather meaningless kind of speculation.
* To see how this is possible, imagine that the universe is really a simulation running in a computer program. Since all particles are simulated, complete information about them at any point in time could be obtained without modifying that information.
My own view on qualia is unsettled, but I hardly think that the above amounts to what is normally called a "soul". I think most people believing in a soul would hold that it is capable of affecting matter in the universe. Otherwise, it's not really necessary for completely human behavior, and you get into the zombie problem, and things go downhill. So then, one has to believe in something that can affect things in the universe, i.e. the atoms in one's brain, and is affected by things in the universe, so that it can receive the sensory data of one's direct experience, but does not itself consist of matter or patterns of matter. I don't see why one wouldn't simply reduce this view and say the soul is simply the physical brain and the mind it instantiates, thus obviating the need for a belief in any immaterial phenomenon participating in this universe. And I think a naturalistic viewpoint would require this.
Posted by: pdf23ds | October 04, 2005 at 10:53 PM
Suppose that hyper-intelligence creates minds so powerful that they become LaPlacean demons. A LaPlacean demon knows where every particle of matter is in the universe and all the physical laws that govern transitions from one state to another.
And where, he asks caustically, would it store this information?
Posted by: Phoenician in a time of Romans | October 05, 2005 at 12:51 AM
pdf23ds, I think the non-participating observer model is close, but it misses the inverse view, that the appearance of bodies might equally be called a non-participating image.
Consider an equation in analytical geometry. It is an idea, namely, the abstract idea of an abstract shape or body. If the equation is modified, the corresponding shape is also modified in a corresponding way. And vice versa. Neither the equation nor the shape is in any way prior to or a cause of the other. Neither a modification of the equation nor a modification of the shape is in any way prior to or a cause of the other.
So that's my understanding of the relation between my mind and body. My mind is a non-participating observer of the life of my body. My body is a non-participating image of the life of my mind. My ideas interact with other ideas; my body interacts with other bodies; and for whatever reason that I won't try to demonstrate, these interactions correspond simultaneously.
Posted by: Gary Sugar | October 05, 2005 at 07:00 AM