Posted 18 May 2000 - 06:39 AM
I''m still not exactly sure how that discussion got so far without ANY interjections on behalf of ye olde gods of the Mesaje Boarde...oh wait there was one, but it seems like nobody noticed...Anyway there are a number of absolutely critical points to be made. Before I (try to) make them, I have to say that I do find it amazing that people on this board have stumbled on so many important concepts in genetic programming all over again. Congratulations (and that isn''t even meant to be sarcastic).
Part I: the original intent
The fact that much of the original idea for the thread constitutes the heart and soul of genetic programming as a whole field of study has been mentioned, although most people seem to have skipped over it. It is an important note, and the simplest way to decide if you feel that your ideas are significantly different is probably just to look at Steven Woodcock''s Game AI page (www.gameai.com) and check the links from there - ESPECIALLY Tierra, and probably Sugarscape as well. If after searching through that material and whatever else you run across you still are not satisfied, then I''ll say this: what you want is an interpretive environment, much like the Java VM. Except this machine has a lot of bytecode-like physical reactions, and has some kind of generalized struct for creatures and objects - like, for example, a list of components. Each component affects the others automatically via the physical environment of the interpreter. This in turn gives rise to all the stuff that has been conjectured - sex, evolution, eating, etc. If you start with nothing, for example, then one might evolve a creature with no physical needs. In a roleplaying game called HOL, these creatures are small white and jellylike. And tasty. But I digress...I believe that the described environment has been referred to in the thread with the term ''testbed'', which is also what most of the literature that I''ve come across calls it. Basic point, look around, and if you''re not satisfied, it''s a neat idea for a project if peeps are interested ( which is another forum entirely). You''d probably be able to come up with an A-life software toy, at least.
Part II: the rest
Since the topic has been covered exhaustively, I''ll keep this short. Minister: the point that sensing is more than receiving and processing an image is well-made. However, there''s absolutely nothing saying that this is the only thing that a computer can do with a visual signal. A properly constructed visual system would link into a larger array of senses that linked into a system which actually did the grunt work of being "intelligent". Emergence is the technical term: by linking into such a system, our eyes allow us to do what you refer to as seeing, but in actual fact there is as yet no evidence to indicate computers *can''t* do this.
Someone else suggested that dogs and flies are on the same level w/r/t self-awareness. WTF? really? Then the whole energy/cost tradeoff that makes the dogs brain much larger than it has to be for minimal control of its systems is doing...what exactly?
My point overall is this: although we can''t simulate the universe, we can simulate a universe (somebody already pointed this out, of course), albeit a very small one, and in that universe there''s no reason that things can''t develop with a kind of sentience. Nor is there any reason that a simple evolutionary physical system couldn''t eventually give rise to that sentience, with the caveat that resources would likely have to expand tremendously to get a properly sentient being inside the computer.
P(ost) S(cript): To the masters student guy and his respondant: It might be useful to consider the situation in terms of the effect of major catastrophe on the landscape of the environment (the fitness function map) as well as the actual organisms'' position therein, since a massive catastrophe would ideally change the shape of the fitness function ( that is, at least, the way it should probably work in world-sims. OTOH, for a simple static-landscape search things are prolly not as peachy - in a static landscape mutation rates would be more important.) Maybe the best course is to examine how a variety of cataclysmic/world-shaking events do shake the system up. And how adequacy fits in, since there aren''t many "optimal" solutions in nature as-is.
Don''t mean to harangue,
just adding my 3.5 cents in.