> From: Dan Minette <[EMAIL PROTECTED]>
> From: "Erik Reuter" <[EMAIL PROTECTED]> > > > On Sun, Mar 10, 2002 at 09:26:51PM -0600, Dan Minette wrote: > > > > > Before I answer this Eric, let me ask you a question that will help me > > > frame an answer. Do you think science is about the Truth or is it a > > > means by which we model, predict, and manipulate phenomenon. > > > > Sniff. Either this is a trap or you must not know my views as well as I > > would have guessed. > > No trap at all. I now know I need to argue from a practical point of view. > Let us consider the practicalities of AI. I remember when they were first > touted 20 years ago; when LISP was a hot language. Expert systems were come > close to replacing various human experts within 5 years. Anyone from a log > analyst to radiologist could be replaced by an expert system. > > Its now 20 years later, and the horizon for such uses of AI appears to be > further away than when they were first touted. Yes, there are uses for > expert systems, and for neural networks. But, they are far more limited > than they were expected to be when computer power was almost a million times > as expensive. > > Second, when computers are used, the algorithms that are used are carefully > written, debugged, tested, rewritten, redebugged, retested, etc. In every > step there is a designer/designers who is figuring out what went wrong and > what has to be fixed. Thus, algorithms seem to be an expression of how > carefully a designer thinks, not something that just happens. > > With real AI, programs would have to be self modifying. This has been > touted to be just around the corner, again, for a couple of decades. From > what I've heard, the success has been limited to very restricted toy models > (a technical term, not inherently derisive). In most cases, the program > soon blows up. > > Third, humans appear to be able to do things that have been proven to not be > capable of being reduced to algorithms. Handling self-referential > statements is one of these things. I realize that Dennett argues that > humans only appear to be able to do this, its just that they have algorithms > that search the possibilities until they stumble over them...like a chess > playing program. However, it is very curious that humans would have such an > algorithm in their heads without being able to access most of the results: > since such access would be evolutionarily favored. In other words, a human mind do Polynominal calculations in Non-Polynominal time, like a quantum computer. > Fourth, from the time of Bohr's early writings on the implications of QM on > biology (in the '30s IIRC), it has been thought that QM renders the > predictions of brain states impossible. Even though, as Zimmy pointed out, > the neuron is rather large on a quantum scale, it is also very complex (as > Zimmy also pointed out.) Thus, quantum effects can rise quite quickly. > Indeed, in a post a while back, I showed how on an idealized pool table, > quantum chaos takes effect within 1-2 seconds of the positions and momenta > of the balls being well established.. > > Once we saw how computers worked, it was reasonable to suppose that our > brains might also work algorithmically. However, the last 50 years of > experimentation have not yielded the results expected by the AI enthusiasts. > Instead, very serious arguments have been raised against that hypothesis. > Practically speaking, while the hypothesis has not been falsified, it > doesn't seem to be a leading candidate. > > > Anyway, I am a pragmatist, and an experimentalist. Truth be damned! :-) > > As far as I can see, the last 20 years of experimentation shows that the > claims for AI are vastly overstated: that if it will work, it will be in a > much longer time frame than touted. My practical experience is when someone > starts talking about AI solutions, one should count the silverware _very_ > carefully before they leave.
