The hardware part would be interesting. I suppose you could do SMP or SMT and thread the program so that any AI processes would work on one of the other processors, but that''d take some extra work I imagine.

As for AI engines, has anyone checked out the OpenAI engine at Sourceforge? There''s also two commercial ones...one is by louderthanabomb, and I forget who makes the other one. www.gameai.com and aidepot.com are cool links.

I wish I understood more about AI concepts. I was browing amazon.com for good AI books...sheesh, and I thought computer books in general were expensive. They all seemed to average about 100$

you also don''t want to buy one of the technical maunals which are all about the theory of intelligent processes and so on... very annoying stuff, I wouldn''t touch it.

George D. Filiotis
Are you in support of the ban of Dihydrogen Monoxide? You should be!

I recently designed a hardware (and firmware) only AI system to control a robot. It was basically a rule based neural network, with an expert system actually handling the evolution of those rules, all stored in memory and sorted by processor directly, like an OS. But then I realised that the program is so simple (yes, really) that an entirely software based system should be just as fast as an equivalent dedicated AI computer, because the basic functions the machine does will be virtually identical. It's just number crunching, at least the way I plan to do it.

I could be wrong, however. I haven't actually finished any of the various prototypes yet, what with also making a game and doing exams and stuff.

With regard to the original question, any true AI system is simply a black box, an appropriate response generator, that will learn to adapt it's responses to input in order to maximise "success". Thus if you can translate between the I/O of the AI software and that of the game you want it to control, any "proper" AI should eventually learn how to play, but it could take quite a while.

If the so called AI fails to do this, then it quite simply isn't true AI.

If it's books you're after, the only good book I EVER found was one titled "machine learning", with specific regard to applications in expert systems and information retrieval. I forget the exact title, but it reads something like that.

If anyone's particularly interested, wait a week for me to get back to school and I'll find it again in the library, and get you the full title. I'd strongly recommend it for anyone interested in serious AI.

[edited by - Captain Insanity on April 2, 2002 12:56:24 PM]

I''m interested in the book

The reason the MIT boys went into the whole hardware side of things is that their system was reliant on a different processing paradigm. For that reason, I really think that systems like that, which are designed to relate through approximation, and make fuzzy calculations, may have to be made separate from software run on the CPU, because so far, current linear execution software is very bad at fuzzy logic.

Which is the major difference between what you''re trying to do (cpt. insanity) and what they were trying to do. Your hardware is making faster functional programming possible, but it isn''t doing anything radically different.

George D. Filiotis
Are you in support of the ban of Dihydrogen Monoxide? You should be!

Check out BioGraphic Technologies'' AI Agent plug-in...

quote: If the so called AI fails to do this, then it quite simply isn''t true AI.

Well, at the risk of bringing down the wrath of the insane... I think that the definition of AI is that it can only learn so much. It doesn''t have billions of cells that link and unlink from one moment to the next in a furious mix of hormones, chemicals and bio-electric excitement. The amount that one animal takes in and records, updates, changes and deletes in one second is usually more than any cluster of supercomputers can hope to even consider taking in during an hour.

A computer is still just ones and zeroes and no matter how hard it works, it can''t compete with the awesome computing power of an animal''s brain. We can make intuitive leaps that aren''t based upon any one or even 50 things that are related. Einstein may have figured out E=MC^2 not due to anything that was related to physics, but due to the fact that he had all that physics knowledge, when combined with the fact that his girlfriend at 15 years old told him that she liked Hanz more because he was blonde, it turns out to be E=MC^2. There''s no way to equate that seemingly unrelated leap to a computer program or even anything that isn''t built similarly. Sure, they can add 1 and 1 all day long to come up with 2, or even come up with a variant of mathematics that comes up with something that looks like real intelligence.

The brain is so completely beyond non-biochemisty that it''s not even funny. Not just the human brain, any brain. Every time you remember something, you''re not just pulling a record, you''re pulling that record and portions of everything that is related to it, then re-writing that record as you remembered it, complete with all that junk thrown in. The next time, different related and semi-related stuff will get tossed in and after you remember it about a few dozen times it will actually only generally resemble the event that actually occured.

Anyway, that wasn''t why I started this post... I started it because I think that AI is just that... artificial intelligence. Don''t expect too much of it. Don''t stop trying to simulate real intelligence, but don''t be disappointed when you fail, it''s kind of inevitable with the medium that you''re dealing with.

I found the book. It''s one of a large series on AI by Ellis Horwood.

It''s called "machine learning applications in expert systems and information retrieval", by Richard Forsyth and Roy Rada. It takes a strong "how to do it" approach. I strongly reccomend it. Another book in the series is entirely about game play theory, interestingly enough.

Did anybody notice my post?