Eliezer S. Yudkowsky wrote:
> > I would imagine from reading Hofstadter that he is fully capable of
> > being just as arrogant as Marvin Minsky.
>
> I can't for the life of me remember anything in _GEB_, _Metamagical
> Themas_, or _Fluid Concepts_ that could be taken as arrogant. Even when
> he was ripping apart the "Structure Mapping Engine" in _Fluid_, he was a
> lot nicer about it than I would have been.
Hofstadter doesn't make such bold claims as others in the field, but he is still in no position to claim that his models have any relation to human intelligence than any other arbitrary model. It is impossible to rigorously test if a model thought-process is the same as a real one. AI researchers have ignored this problem by moving the goal posts; they're not simulating the brain, they're simulating 'intelligence'. The problem with this becomes apparent when you consider that they are both defining and simulating intelligence simulaneously. Thus the simulation creates the definition and the definition creates the simulation, eventually we arrive with both a defintion of intelligence and a simulation, neither of which mean anything to anyone. To complicate this cycle even further the MIT AI lab has even stated one of it's goals as understanding the brain to help us think better. So these researchers are basing their models on observations of their thought processes at the same time as trying to think in terms of those models! I personally find the entire premise of AI research to be based on ill logic. As I have said before, the very idea that you can understand the brain while paying little or no attention to its structure is ridiculous. Based on this I think it's possible to say all AI researchers (and self-proclaimed cogintive scientists), Hofstadter included, are somewhat arrogant. Obviously I find other causes for thinking Hofstadter in particular could be arrogant, but a large part of that is due to what I state above. I'm open to disagreement.
> > As for Lenat, he couldn't be insulted enough. Cyc indeed.
>
> Yeah, Cyc is silly, they're pouring water into the bucket but they
> forgot to build the bucket first. I was talking about EURISKO. Cyc is
> just another failure.
I consider EURISKO to be one of AIs brighter moments but it was missing a foundamental element in that it could not function without constant 'tweaking' by its creator. And that which it lacked was probably worth more than its constituent parts. After starting the Cyc project it became obvious that Lenat's own thinking is obviously misguided. The flaws in the Cyc project are so obvious that either Lenat is a very stupid man or he had an insight with his previous projects that is beyond my own understanding. I find the latter to be unlikely (although if the ability to learn from scratch is intractable then a common sense database would seem to be the only option - perhaps Cyc is intended to be analogous to our own genetic inheritance).
BM