From: Larry Kaufman WIZARD NET> Date: 13 may 1997 Subject: Re: Real chess computing At 04:40 PM 5/12/97 +0200, you wrote: > >>You are just being silly. >> >>As the programmers frequently pointed out, this victory had nothing to do >>with artificial intelligence. The computer does not think. It just runs fast. > >I don't think such a challenge would be silly. Programs in >Japan are starting to get better then 1dan, on a PC. So why >would it be silly to try it on a machine that is a few >thousand times faster ? Sorry to disagree, but I think it would be extremely silly. The top pc chess programs play at (weak) Grandmaster level, and Deep Blue plays about one class better, at the level of Candidates for the World Championship. One class in chess is comparable to one or two ranks in shogi (opinions differ on this); if I allow two ranks, that implies that if pc programs play shogi at amateur 1 dan level, if they were to run on a specially built shogi machine comparable to Deep Blue they would play at Amateur 3 Dan level. This in turn means that with years of effort and millions of dollars expenditure, IBM (or the Japanese equivalent) could construct a machine that could give Habu a good fight -- at Two Piece Handicap! Of course software may be improved in the future, conceptual breakthrus may occur, etc., but until a pc program can play at the Amateur 6 dan level (which I would expect to take about twenty years, and I am on record as forecasting back around 1985 that a computer would beat the chess champ in a match in 1995, only two years off), no super computer using similar algorithms could hope to defeat the Pro Meijin. > >>A shogi playing computer program would have to function much differently >>from a chess program. An exhaustive search of the type applied to chess >>simply would not work in shogi. >It's working allright. It's just a lot more difficult to >program. I'm thinking especially of endgame speed were shogi- >programs can get really good. I think the search would need to be somewhat selective to excel in shogi, but could still be similar to some of the selective search chess programs. >A chess-computer might win or lose the next match, but this test was >certainly a historic event. Most people tought it would last >at least another 10 years before the world champion would lose. Not me. >Matt Larry Kaufman, 301-309-0904, 9213 Wooden Bridge Road, Potomac, Md. 20854