[lbo-talk] AI

Curtiss Leung curtiss_leung at ibi.com
Wed Nov 19 15:27:02 PST 2003


Whoa, whoa, whoa--I don't think anybody mentioned strong AI as such and the definition you offer is at odd with what I understand strong AI to be.

First, Searle can be right while his argument is unsound, i.e., machines cannot think, but the reasons for this are not the ones he gives. For myself, I think Searle is hiding behind the magic word "meaning," and I'm agnostic about whether a piece of software or even some kind of specialized hardware could pass the Turing test, let alone have internal states comparable to human consciousness. I think what he *does not* say about state, either in the Chinese Room or in an actual piece of software or hardware is telling, however.

Second, strong AI as I understand it means this: all mental processes can be represented as algorithms/idealized Turing machines. That's not the same as "mimicry of certain aspects of human cognition"; infinitely many Turing machines do not halt, and there is _a priori_ no way to distinguish between those that do and those that don't. This also has zero, zip, nada to do with the actual performance of hardware and software--it may well be the case that the only thing that can *completely and efficiently* implement a brain is a brain--and actually I incline not only to that, but that brain states called "consciousness" only develop in response to social stimuli. But that doesn't matter; people don't implement actual Turing machines either, but they're indispensible as formalisms.

As for saying that machines don't think--well, what is it to think?

Curtiss


> Having worked with AI based systems and huge computing
> clusters for a few years now, I find this faith in
> strong AI bewildering and disappointing. It is
> perhaps, an indication of the deep complexity of the
> systems we depend upon that many people believe almost
> any innovation to be possible - as if via magic.
>
> But strong AI is a failure and will remain a failure
> because its underlying assumption - that minds can be
> built or evolved using canned mimicry of certain
> (poorly understood) aspects of human cognition - is
> deeply flawed.
>
> Well phrased and on-point criticisms of Searle's
> arguments are fine and necessary but do nothing to
> change the fundamentals: machines do not think.



More information about the lbo-talk mailing list