[lbo-talk] Re: AI

Chuck Grimes cgrimes at rawbw.com
Fri Nov 21 09:22:18 PST 2003


Would that not in fact be the proper test to apply? Computers are thinking if and only if they are making mistakes and disagreeing with each other???? Carrol

-----------

This might be a nice thought, if computers didn't make mistakes now. Unfortunately, computers don't compute all that well and therefore they make mistakes, and therefore they disagree already.

There are serious computing problems running modelling and simulations, where the errors in computing make it questionable whether the validity of the simulation is good enough to draw valid conclusions.

I suspect Dwayne Monroe was on the verge of developing an argument like this before Justin cut him off with an undeserved fuck-off.

Then there is the problem that even if you grant that thinking can be modelled as a machine replicable routine (computable function), you can also show that such a routine can't do all mathematics if it only has these routines to work with.

This latter argument was developed in The Emperor's New Mind, Roger Penrose. Unfortunately, I hated the writing style and thinking method so much, I gave it away after I finished it. So now I can't re-produce the argument.

Basically it linked up Goedel, Turing, Church, Russell and some other stuff to show all mathematics was not subsummed in computable functions---anymore than you can write down an axiom set that is both complete and consistant. Penrose saves the AI day, by proposing quantum computers that will be able to `think' like humans. This involves building in error in the form of probability, phase spaces and uncertainity. (I could be horribly wrong here since I don't have the book to check. Maybe Ian Murray remembers this better than I do.)

So the point here is that we don't need to make an argument that involves a transcendent subject, because AI can not succeed even on its own territory, since computers don't work all that well to begin with.

And then there is the problem that a Turing machine isn't a machine. It is a logical routine that is presumed to be able to work indefinitely. I can't remember now, but I think there is still a problem in logic of whether or not a computer can in principle be made equivalent to a Turing machine.

Well that ought to be enough mis-remembered bullshit for one morning's post before work...

Chuck Grimes



More information about the lbo-talk mailing list