Are there any arguments against strong AI in principle, given physicalism?

Are there any arguments against strong AI in practice, given physicalism?

Basically, I want to be aware of any potential objections to an assertion that if physicalism is true, then strong AI is possible. Personally, I don't see how this assertion is false, but philosophers are creative, so there may be someone who tried to reconcile the idea of physicalism with a ban on strong AI.

  • Can you explore the qualification "given materialism" a little bit further? A bit more about the context and motivations of the question could help too -- maybe you could tell us something about what you've been reading or studying that's made this an interesting or important question to you? – Joseph Weissman Jun 12 '14 at 16:25
  • 1
    Just to be clear, by 'materialism' you mean 'physicalism'? – stoicfury Jun 12 '14 at 18:12
  • I am not sure what the current test of "strong AI" is, but there are many potential barriers recognized by "physics," if that's what you mean. Entropy, speed of light, and--strangely enough, though in my view important--gendered differentiation. If by "strong AI" we mean passing a Turing test administered with total transparency by an AI scientist, then it may be necessary that AI pass though a biological and even "gendered" production process. – Nelson Alexander Jan 29 '16 at 1:51
  • @NelsonAlexander given physicalism, we know at least one type of physical system (brains) is involved in intelligence despite the physics based constraints. Your comment has clarified something in my mind: any such argument would be a "uniqueness" one, that natural intelligence is a uniquely natural phenomenon, which then has a Wittgensteinian "family resemblence" type of problem. – Dave Jan 29 '16 at 18:10
  • True enough. I find AI and machine learning fascinating. But I like to annoy people by pointing out that proper AI shouldn't just play chess and dating games, it should be able to solve the big problems of NI...like reproducing itself. Preferably through a"gendered"bifurcation that can eliminate bugs, viruses, and programmers. Anyway, the whole scheme seems rife with paradoxes of self-reference. – Nelson Alexander Jan 29 '16 at 18:48

In principle, it could be the case that although materialism is true, there is some class of physical systems that can't be simulated on computers and this happens to include the human brain. In principle, it could be the case that the human brain could be simulated on a computer but is is impossible for anybody to understand the relevant algorithm and so impossible to set up the simulation.

In practise, the first possibility is ruled out by the best existing theory of computation, which states that any physical system can be simulated by a universal computer operating by finite means:


The second possibility seems implausible. It has to be the case that whatever algorithm our brain is running was created by rounds of variation and selection such that each stage it produced something that allowed genes to be copied. So there must be some sequence of problems we could solve that would lead to something that could be tested at each stage and would produce AI at the end. Why would we be unable to do this? It can't just be a matter of memory or speed or reliability of computations since computers can do that for us. So I don't see what can stop us from creating AI.

| improve this answer | |
  • With regards to the second possibility: Evolution does not solve problems. Evolution also does not have a goal nor must the things that evolve be useful. The only thing they must do is spread quicker, spread wider than others. – pandita Jun 13 '14 at 13:26
  • You are right. A gene need not have a useful phenotypic effect as long as it spreads. I have modified the argument to account for that. – alanf Jun 13 '14 at 15:21

Alanf is right to point to the fact that the Theory of Computation says that any physical system can be simulated by a universal computer. Those theorems don't address the cost associated with those simulations. So, it could be that in practice, the amount of time to completely emulate a human brain, or some other element of the cost of the simulation, makes doing so prohibitive for the foreseeable future.

| improve this answer | |
  • 1
    "the amount of time to completely emulate a human brain, or some other element of the cost of the simulation, makes doing so prohibitive" — but for what kinds of systems? Because it obviously isn't intractable for our brains to do themselves. – Niel de Beaudrap Jun 13 '14 at 10:25

Your assumption means that strong AI is possible; since it determines that we are some form of organic computer; we would simply need to reverse-engineer the 'bio-technology'; whether this is possible or not, now or in the future is besides the point; your assumption means it is possible in principle.

Were we to say, rather than physicalism, naturalism (recall the early modern scientists were known as natural philosophers - philosophers of nature essentially), this would leave open the possibility that this may not be possible, even in principle.

| improve this answer | |

In general I would say that neither naturally follows the other:

There might be a system, existing within another system, that cannot fully compute the same functions that the parent/sister system does.

I.e. assuming you can compute anything within a physical system that can be calculated just because you are within the physical system is simply not warranted.

Hence even if physicalism is true, you can't deduct that Strong AI is possible. That is unless you know physicalism to be true in a very specific sense.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.