RPJesus

Yes, I partialy agree with you ,I have been a little bit too much semplicistic.
When testing a position the computer must assign a fitness score.
Also the programmer must know the basic of tree decision alghoritms to be enable to cut off the wrong branches etc.
The point is that the AI alghoritms are defintly not the main issues in this kind of application.
When designing the last generation super computer " Blue ...something " to play vs the chess world champion the designers focused on hardware , definitly not on software.
Not only but they even got rid of some previously developed sophisticated alghoritms , being of no use.
If you own AI : a modern approch, by Russel and Norwig it is clearly mentioned
For example a super computer miserably loses against a low level "GO" player , because the AI is not fitted to evaluate the position of the " the thousands white and black small stones layout " of the chinese game .
Same in a war game one of the most difficult task is to evaluate the position of the the opposing armies for the same reason.
cheers

AlbertoT,

The computer you are referring to is Deep Blue. And, in fact, they concentrated on software much more than hardware. The 2nd gen. deep blue was only 2x as fast as the 1st gen. model, and it could kick its ass: "Deep Blue will be significantly stronger too, searching twice as many positions per second and searching them with enhanced chess knowledge." They hired a few grandmasters to help with the evaluation function. As far as I know, they did concentrate on superior hardware for the 1st gen. model, but obviously realized that raw speed was not good enough.

Also, in regards to the algorithm, it is not simply a matter of searching all moves. For deep blue to search 14 ply in middle game where, on average, positions have 35 possible moves, it would take 35^14 = 4,139,545,122,369,380,000,000 positions to search. Even at 200,000,000 positions / second, it would take over 650,000 years to search to ply 14. Obviously, some AI comes into play here. What makes a good grandmaster is to know which positions to forget about, and the same works with computers. Many positions can be eliminated with superior searching algorithms and tree pruning.

Jason Doucette
www.jasondoucette.com

[edited by - Jason Doucette on February 21, 2003 3:15:21 PM]