debray@sbcs.UUCP (Saumya Debray) (08/09/85)
> Samuels (Samuelson?) had a checker program in the mid-fifties which ran > on an IBM 704, I think. It played a quite good game, they said. It had > an adaptive evaluation function so that it was self improving. Check > the literature. The program you refer to dates through the sixties. It had, as far as I recall, a learning component, but used the same evaluation function for its estimate of its opponent's estimate of a position as it did for its own estimate of a position, and hence was not really "adaptive". (A truly adaptive system would maintain two evaluation functions -- one reflecting its own strategy, the other its estimate of its opponent's strategy. It would use predictive-corrective methods to continuosly modify its estimate of its opponent's strategy, and then adapt its own strategy to this.) Talking of game-playing programs, Hans Berliner at CMU had a backgammon program in the late '70s - early '80s that beat the then world champion pretty convincingly. But I guess that doesn't really belong here ... -- Saumya Debray SUNY at Stony Brook uucp: {allegra, hocsd, philabs, ogcvax} !sbcs!debray arpa: debray%suny-sb.csnet@csnet-relay.arpa CSNet: debray@sbcs.csnet
stevev@tekchips.UUCP (Steve Vegdahl) (08/14/85)
> Talking of game-playing programs, Hans Berliner at CMU had a backgammon > program in the late '70s - early '80s that beat the then world champion > pretty convincingly. But I guess that doesn't really belong here ... I was in the CS department at CMU when that match occurred. The victory in the deciding game was anything but convincing. The (human) world champion had a very strong position, but Hans (rolling for the computer) rolled something like a double-5 and two double-6's right at the end in order to eek out the victory. The score might have appeared convincing due to doubling that may have occured during that game. I don't remember the details. I do remember Hans relating the story to some of us in the department; he gave the strong impression that his program was very lucky to have won the match. Steve Vegdahl Computer Research Lab. Tektronix, Inc. Beaverton, Oregon
debray@sbcs.UUCP (Saumya Debray) (08/19/85)
Steve Vegdahl: >> Talking of game-playing programs, Hans Berliner at CMU had a backgammon >> program in the late '70s - early '80s that beat the then world champion >> pretty convincingly. But I guess that doesn't really belong here ... > > I was in the CS department at CMU when that match occurred. The victory > in the deciding game was anything but convincing. The score might have > appeared convincing due to doubling ... I do remember Hans relating > the story to some of us in the department; he gave the strong impression > that his program was very lucky to have won the match. Berliner had a paper on this program in "Artificial Intelligence" in '80 or '81. I seem to remember reading that afterwards, they reversed positions to test the program: the "opponent's moves" fed to the program were in fact its own moves when it had played the world champion, and in almost all cases, it made exactly the move the world champ himself had made in that position. I was tremendously impressed by this. Again, I'm not sure how much a discussion on machine checkers belongs in net.chess -- I've changed the followup line to cross-post this article to net.ai: please send followups to this article to net.ai ONLY. -- Saumya Debray SUNY at Stony Brook uucp: {allegra, hocsd, philabs, ogcvax} !sbcs!debray arpa: debray%suny-sb.csnet@csnet-relay.arpa CSNet: debray@sbcs.csnet