The first strong computer opponent was BKG 9.8. It was programmed by Hans Berliner in the late 1970s on a PDP-10 as an experiment to evaluate board positions. Early versions of BKG played poorly even against poor players, but Berliner noted that the critical mistakes the show made were always in phase shifts. He applied the basic principles of fuzzy logic to smooth the transition between phase changes, and in July 1979, BKG 9.8 was ready to play against then-current world champion Luigi Villa. He won the match, 7-1, becoming the first computer program to defeat a world champion in any game, although this was primarily a matter of luck, as the computer had better dice rolls than its opponent in that match.
Starting in the late 1980s, backgammon software makers began to have even more success with a neural network approach. TD-Gammon, developed by IBM’s Gerald Thesaurus, was the first of these software to play at or near the expert level. The neural network of this program was trained using Time Difference learning applied to the data generated from self-play.
This line of research has resulted in two modern commercial programs, Jellyfish and Snowie, the BGBlitz shareware (implemented in Java) and the GNU Backgammon free software, which play alongside the best human players in the world. It is worth noting that without their associated “weight” tables, which represent hours or even months of tedious neural network training, these programs do not perform any better than a human child.
It is interesting to contrast the development of backgammon software with that of chess software:
For backgammon, neural networks work better than any other method so far. For chess, brute force searching, with sophisticated pruning and other enhancements, works better than neural networks.
Every advance in the power of computer hardware has significantly improved the strength of chess programs. In contrast, the additional computing power appears to improve the robustness of the backgammon software only marginally.
For both backgammon and chess, it is currently unclear whether the best computer or the best human is the best overall. For most other games, one or the other is unequivocally stronger.