Bill Robertie

TD-Gammon was an experimental neural network program developed at IBM’s research labs in the late 1980s. TD-Gammon taught itself to play starting only with a knowledge of the rules of the game. After playing thousands of games against itself, the program reached strong Open player level. Within months it approached what the developers thought was world-class play. To evaluate TD-Gammon, the developers invited Bill Robertie to travel to White Plains and play an exhibition match against the program.

Learning from the Machine is the story of that match, which lasted 31 games. The neural net technology developed in that project eventually led to the development of Jellyfish, Snowie, and Extreme Gammon, the commercial backgammon bots of today.

Level: Intermediate/Advanced
Info: 51 pages, softbound
Price: $20 + S/H