james Rogers wrote: > I would, at the very least, agree that most published research on the > supporting mathematics is inadequate at best. This has been a > (the?) major > stumbling block since the beginning of computer science with > respect to AI.
Well, I think that the mathematical questions involved here are very hard. They have to do with self-organization, emergence, fuzziness and probability all mixed up -- the sort of thing existing math doesn't handle well at all. I often wonder if there's a catch-22 here: maybe we need intelligent computers to do this kind of math [as with some kinds of physics calculations, because the theorems just won't come out neatly and compactly enough for humans to formulate in realistic timeframes] ... but we need the math to create the intelligent computers... But I now believe that we can create generation 1 of the AGI programs without having the math, and then they can help us create (or themselves create) the math needed to make generation 2 AGI .... > I just think that it is a bit of "the cart before the horse" that so few > people doing AI research are actually tackling the fundamental theoretical > questions rather than trying to devise solutions. To a certain extent I > understand it -- the glory is in the solution -- but it seems that people > haven't been as lucky as they hoped. The vast majority of folks doing AI research are not focusing on AGI at all. There are plenty of AI theorists out there, as can be seen from the academic journal literature. But it's narrow-AI theory -- neural net stuff or formal logic stuff, convergence of evolutionary programs, etc. whatever -- not AGI theory. I think the problem is an imbalanced focus toward narrow-AI rather than AGI, not an imbalanced focus toward solution-devisement rather than theory... > > To put it another way: Even if we built a perfect model of the human brain > on silicon by perfectly copying the functionality of biological > neurons and > their interconnections and so forth, what have we really learned? In my > estimation, precious little. I agree, but I'm afraid this is a bit of a "straw man" argument, as nearly everyone involved with bio-inspired AI is actually also involved with trying to understand how the pertinent brain systems work. > I > guess in some > ways I'm less interested in building any old viable AI as I am in figuring > out the fundamental science of intelligent systems. Hmmm... I'm not fussy -- I think that either one would do nicely, for a start ;-) But I tend to think the two will grow up together -- one's understanding will increase as one moves along the path to AGI-construction (which is how it's been for me so far), and then one's understanding will increase MUCH FASTER once one gets to the phase of systematic AGI-testing... > One thing we've had to deal with is that in practice my best theoretical > models map closer to biological systems than I ever thought possible, > particularly since they were not derived from biology. I don't know the details of your work, but this kind of outcome is often found in complex systems research. "Dissimilar underlying systems, similar emergent structures/dynamics" -- that's what complexity science, such as it is, is all about, right? > And at > this point we > are still doing more with less (in a design complexity sense) > than any other > project that I'm aware of. Doing more with less is great ... but what really matters in the end is just doing the most possible with what's available, eh? ;) I don't know the details of your design but I'm guessing it's much simpler & more mathematically elegant than Novamente, which is large & diverse. I have a philosophical belief that a simple elegant design is not going to work for human-level AGI given the magnitude of computational resources at our current disposal. This belief is supported by plenty of experience on the part of myself and others. But of course, there is nothing like solid, absolute knowledge involved here -- we're all a lot more ignorant than we are knowledgeable - and I look forward to hearing the details of your design or seeing the behavior of your system one day -- ben ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/
