The Good Backgammon Bots
Sunday, June 28th, 2009With the battle against the misuse of backgammon bots, we almost forgot that these programs were originally purposed to help us become better players. I thought about it after reading an interview with Martin Smith, GameAccount bots buster and the creator of ProBot, not the heavy metal side project of Nirvana drummer and Foo Fighters lead-singer Dave Grohl, but the toughest challenger in the family of the games network’s robotic opponents which includes AmateurBot, RookieBot and Beginnerbot, on FT.com.
Backgammon Bots History Revisited
First, an old obligation. At the end of the first episode of the long and winding history of backgammon bots, after looking at the first experiments in backgammon computing, the winning BKG 9.8. and the commercially successful Expert Backgammon, I promised to leap forward to 1990 to review TD-Gammon, still the most revolutionary backgammon software in the history of the game and artificial intelligence.
TD-Gammon Backgammon Software
TD-Gammon was developed by Gerald Tesauro of IBM’s Watson Research Center in New York, as part of a reinforcement learning study. The TD stands for Temporal Difference learning, a prediction method mainly used for reinforcement learning, or in less scientific jargon, the computer taught itself to play backgammon and to play better and better. The eventual result was the first neural net to really take up the best human players. Moreover, TD-Gammon could have also improved the way its human opponents played backgammon.
Back to ProBot
Mr. Smith, whose doctoral thesis is based on Teasauro machine, had similar methodical aims when building his family of bots (which, by the way, can not only play backgammon but also blackjack, poker and other skill-luck combined games). That is why ProBot is not undefeatable; as in human vs. human backgammon games, a beginner can beat a pro with the right rolls.
What ProBot does is "it wins for a couple of weeks, then the humans figure out how to beat it and they win for a couple of weeks while Smith goes away and works on his software… Then he comes back with a new version of the program that wins for a couple of weeks, while the humans go away and think about it." While the bot’s younger, less skilled brothers are programmed to make stupid mistakes, just like humans do.
According to Smith, who holds a PhD in with Artificial Intelligence and a past in chess and poker hustling, these nonhuman players are the best opponents a human player can ask for; bots don’t bitch and whine as you win nor do they blow their horn when you lose, "they won’t be saying: ‘I’m just going to make a cup of tea,’ and not come back. There’s none of those annoying human frailties."

