bwm@ccice2.UUCP (Bradford W. Miller) (10/09/84)
Why not 'just' build an expert system? Brad Miller -- ...[rochester, cbrma, rlgvax, ritcv]!ccice5!ccice2!bwm
robison@eosp1.UUCP (Tobias D. Robison) (10/09/84)
It's not reasonable to assume any linear increase in playing strength for ech doubling in machine speed. As the search goes through more and more plies, a doubling in speed has less and less effect, because the number of positions to search increases exponentially (roughly) with the number of plies. Above a rating of 2200, doublings in speed might be worth 50, 25, 15, 8, 6... respectively. Adding depth to the middlegame search may allow a computer to become world champion if computers simply start to outstrip the accuracy of any human searches. This doesn't seem likely to happen for a while. When it does begin to happen, humans will use the best computers as "trainers", and boost their own performance to match the computers for a while; humans have shown over and over an ability to boost human performance if there is any way to push them. As computers become able to press human beings in the middlegame, endgames will become more important. Does anyone want to comment on Karpov's judgment in arranging to win collect his 4th win by exchanging his bishop for knight, and then temporarily sac'ing a pawn, to wina knight-vs-bishop ending? Some computers may be able to play endings like that moderately well, but when will they be able to decide to steer into them? - Toby Robison (not Robinson!) allegra!eosp1!robison or: decvax!ittvax!eosp1!robison or (emergency): princeton!eosp1!robison
kinman@garfield.UUCP (Kinman Tam) (10/12/84)
There has been quite a bit of work done on endgames for computer chess
recently. Endgames allow more controlled experimentation than the
middlegame. More importantly, the endgame has traditionally been the part
of chess that computers are the weakest at. I was quite astounded when I
learned a few years ago about the difference in their caliber of play between
the middlegame and the endgame. Computers could outplay 90% of non-tournament
chess players at the middlegame, but this figure dropped drastically for the
endgame. I can't remember what the percentage for the endgame was, but it
was evident that there was definitely room for improvement. Does anyone
have some up-to-date figures on this?
--
Kinman Tam
{allegra, ihnp4, utcsrgv}!garfield!kinman