AI and consciousness, first half
Jan. 9th, 2013 11:18 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
davvAnd now for the second part of what I thought about... here it gets rather less concrete.
We know we are conscious. Well, I'm going to assume this, so zombies will probably not get much out of what I'm going to write :)
Now, there are three options as far as the mind and body, and particularly that which evolution tries to optimize, is concerned. Let us call something that is as powerful as a computer we ordinarily think of, but definitely not conscious, a computer. Then:
- Consciousness might let the conscious being do things no computer can do,
- Consciousness might be one way of doing something an ordinary computer could also do,
- or consciousness might not have any effect on the material self at all.
Let's take the last one first, because that's the one that appears to make the least sense. How can consciousness not have any effect on the material? I am aware, after all.
Yes, that is true, but it might be the consciousness that's along for the ride, or flipping through channels, as it were. Thus, this option usually leads to epiphenomenalist theories. For instance, many-worlds may be literally true and the consciousness just selects between the worlds. The conscious "me" decides to raise my arm, but all that means is that I am moving to a universe where my bodily self raised its arm. In that universe, my bodily self raised its arm for some other reason. Such a theory is unfalsifiable from the material perspective, because it works even when there's no way you can determine whether any given material body is being "felt" by a consciousness. I might be able to subjectively verify it - say, by jumping off a cliff and always surviving (because there's an universe in which I survive, and I jump there because there's nowhere else to jump) - but that is privileged information to me and nobody else will know it.
This particular epiphenomenalist theory has its problems. There's no obvious explanation why I can choose to jump to a universe where I raise my arm, but not to one where I decide to do a Superman and fly into the air. It doesn't explain why I'm a human and not a sentient (but not sapient) rock; and it doesn't explain how I think, yet with all the biases of an actual human. However, it does show the gist of things: how you could have a conscious narrative without actually "doing" anything. It's a choose-your-own-adventure on a truly enormous scale, but the adventures were already made.
In this kind of setting, philosophical zombies exist and they're completely indistinguishable from real people. Science has no chance of getting at the mind-body problem, only "revelation" (as in the subjective experiment above) can do that.
Okay, that's the last one. Now to the middle one.
This possibility is the most dangerous one as far as artificial intelligences go. Say consciousness serves as some kind of planner. The material self pushes and pulls at the conscious self by being the man in the middle and by coloring some sensations in "don't do this" (pain, itching, discomfort), and others in "more of this" (pleasure, relaxation). But the middle possibility also says a computer could be just as good a planner. Thus, we don't know if a general optimizer is conscious. It most likely wouldn't be, and then we're in trouble. If the general optimizer is given a very broad goal without clauses that keep it from trampling over people, then the general optimizer might out-compete people simply because its computerized (not-conscious) planner is as effective, or more so, than the conscious planner. After it does so, conscious beings then go extinct and consciousness in the universe is lost. The same thing holds if uploading becomes all the craze and the artificial self is not conscious (perhaps because of a replacement of the conscious planner with a more effective not-conscious planner). Someone uploads himself, the old material self is destructively copied, and it's lights out for him as a very good reproduction comes into being.
However, this possibility is scientifically better. Since consciousness has an effect, science could find out what links consciousness to the material body - or what kind of planners (or whatever) are conscious. This would probably take the form of some kind of experiment where one tinkers with the brain and asks the person if he feels more or less conscious than before. Random tinkering of the brain would show that something tends to become less effective as the person becomes less conscious - planning in the example above - and then one can home in on that to find out exactly what makes it special. It's in any event very important that we find out what it is, because survival might only be weakly correlated with consciousness. If evolution finds a more effective non-conscious organism, then it, too, might outcompete and destroy us.
Assuming some humans are conscious, human zombies wouldn't exist in this setting unless the effect is very small. The simple reason is that if evolution added consciousness for some benefit (say planning), then not-conscious humans would be worse at this unless they were in some other way better at planning; and it's unlikely that evolution would produce both a mutation to cancel out consciousness *and* one to implement a material form of planning that would take its place. Other species, however, could easily lack consciousness and bacteria (for instance) probably would be[1].
This post is getting long enough, so I'll deal with the final option and some implications for artificial intelligence in another.
[1] (In the original text file that I wrote, I forgot about another possibility I've later come to call the "CQ" possibility. In brief, perhaps consciousness only recently evolved. If so, the imperfect spread of it, due to evolution not having caught up fully yet, would permit kinda-zombies: less effective but still very real. But this has other problems. Perhaps I'll write more about it in a post of its own after all this :) )
(And is anyone actually reading this? I'm curious.)
We know we are conscious. Well, I'm going to assume this, so zombies will probably not get much out of what I'm going to write :)
Now, there are three options as far as the mind and body, and particularly that which evolution tries to optimize, is concerned. Let us call something that is as powerful as a computer we ordinarily think of, but definitely not conscious, a computer. Then:
- Consciousness might let the conscious being do things no computer can do,
- Consciousness might be one way of doing something an ordinary computer could also do,
- or consciousness might not have any effect on the material self at all.
Let's take the last one first, because that's the one that appears to make the least sense. How can consciousness not have any effect on the material? I am aware, after all.
Yes, that is true, but it might be the consciousness that's along for the ride, or flipping through channels, as it were. Thus, this option usually leads to epiphenomenalist theories. For instance, many-worlds may be literally true and the consciousness just selects between the worlds. The conscious "me" decides to raise my arm, but all that means is that I am moving to a universe where my bodily self raised its arm. In that universe, my bodily self raised its arm for some other reason. Such a theory is unfalsifiable from the material perspective, because it works even when there's no way you can determine whether any given material body is being "felt" by a consciousness. I might be able to subjectively verify it - say, by jumping off a cliff and always surviving (because there's an universe in which I survive, and I jump there because there's nowhere else to jump) - but that is privileged information to me and nobody else will know it.
This particular epiphenomenalist theory has its problems. There's no obvious explanation why I can choose to jump to a universe where I raise my arm, but not to one where I decide to do a Superman and fly into the air. It doesn't explain why I'm a human and not a sentient (but not sapient) rock; and it doesn't explain how I think, yet with all the biases of an actual human. However, it does show the gist of things: how you could have a conscious narrative without actually "doing" anything. It's a choose-your-own-adventure on a truly enormous scale, but the adventures were already made.
In this kind of setting, philosophical zombies exist and they're completely indistinguishable from real people. Science has no chance of getting at the mind-body problem, only "revelation" (as in the subjective experiment above) can do that.
Okay, that's the last one. Now to the middle one.
This possibility is the most dangerous one as far as artificial intelligences go. Say consciousness serves as some kind of planner. The material self pushes and pulls at the conscious self by being the man in the middle and by coloring some sensations in "don't do this" (pain, itching, discomfort), and others in "more of this" (pleasure, relaxation). But the middle possibility also says a computer could be just as good a planner. Thus, we don't know if a general optimizer is conscious. It most likely wouldn't be, and then we're in trouble. If the general optimizer is given a very broad goal without clauses that keep it from trampling over people, then the general optimizer might out-compete people simply because its computerized (not-conscious) planner is as effective, or more so, than the conscious planner. After it does so, conscious beings then go extinct and consciousness in the universe is lost. The same thing holds if uploading becomes all the craze and the artificial self is not conscious (perhaps because of a replacement of the conscious planner with a more effective not-conscious planner). Someone uploads himself, the old material self is destructively copied, and it's lights out for him as a very good reproduction comes into being.
However, this possibility is scientifically better. Since consciousness has an effect, science could find out what links consciousness to the material body - or what kind of planners (or whatever) are conscious. This would probably take the form of some kind of experiment where one tinkers with the brain and asks the person if he feels more or less conscious than before. Random tinkering of the brain would show that something tends to become less effective as the person becomes less conscious - planning in the example above - and then one can home in on that to find out exactly what makes it special. It's in any event very important that we find out what it is, because survival might only be weakly correlated with consciousness. If evolution finds a more effective non-conscious organism, then it, too, might outcompete and destroy us.
Assuming some humans are conscious, human zombies wouldn't exist in this setting unless the effect is very small. The simple reason is that if evolution added consciousness for some benefit (say planning), then not-conscious humans would be worse at this unless they were in some other way better at planning; and it's unlikely that evolution would produce both a mutation to cancel out consciousness *and* one to implement a material form of planning that would take its place. Other species, however, could easily lack consciousness and bacteria (for instance) probably would be[1].
This post is getting long enough, so I'll deal with the final option and some implications for artificial intelligence in another.
[1] (In the original text file that I wrote, I forgot about another possibility I've later come to call the "CQ" possibility. In brief, perhaps consciousness only recently evolved. If so, the imperfect spread of it, due to evolution not having caught up fully yet, would permit kinda-zombies: less effective but still very real. But this has other problems. Perhaps I'll write more about it in a post of its own after all this :) )
(And is anyone actually reading this? I'm curious.)
no subject
Date: 2013-07-03 07:58 am (UTC)The BB variant goes: given enough time, we can proceed to find ever more complex busy beavers. A Turing machine can't, therefore we're doing hypercomputation.
The counter goes: but at any given point, there's a pre-prepared TM that can solve all BBs up to the point you've discovered. So there's no reason to presume that you can keep on solving ever more complex BBs; perhaps you're just a very complex TM and haven't reached your limit yet.
(At least that's how I understand it.) So the Busy Beaver variant can only be inductive: it can say something to the effect of "we've surpassed ourselves this many times and we keep doing so, so the evidence that there's some hidden limit is starting to look awfully specific whereas the alternate explanation is much simpler". It can't deductively prove anything by puzzle-solving alone since the bar can always be set higher than whatever has been solved so far.
The similarity lies in that you can't know with certainty whether the provable second-order theorems are a tiny first-order subset or not.