A team of researchers from the University of Alberta has won two tournaments for poker-playing programs held at the annual conference of the American Association of Artificial Intelligence.
"Poker is a game that involves skill, chance, and many forms of uncertainty", said professor Jonathan Schaeffer of the Alberta team. "It is a great problem for Artificial Intelligence, and we stand to learn a lot from competitions like this".
The victories were convincing, as the poker 'bot' from Alberta won every match it played, and amassed by far the most virtual money of any competitor.
"We've been writing good poker programs for many years", said Darse Billings, the lead designer for the Alberta team, "but we weren't overly confident, because there is still a lot of room for improvement".
"We used some of our older technology", added Neil Burch, the lead programmer for the Alberta team. "We didn't have much time to prepare for the competition, so we used programs that have been thoroughly tested, rather than our strongest bots".
A version of the winning program is included in the commercial software 'Poker Academy', under the name 'Sparbot'. Owners of the software can connect their own bots to Poker Academy for testing and evaluation.
"We thought that would be a huge disadvantage, because the other teams could practice against our programs 24 hours a day, every day", said professor Michael Bowling, who heads up the Alberta research group.
The game of one-on-one Limit Texas Hold'em was played in both computer tournaments. The first event used a normal pace of about one second per decision, as seen in games played by humans. The second event allowed a much slower pace, in the hope that it might produce a higher level of play. Somewhat surprisingly, the winning program from Alberta made every betting decision instantaneously, even in the slower event.
The AAAI event is the most intensive competition there is for poker programs. In order to reduce the effects of luck, more than a quarter million games of poker were played in the two tournaments. To make the results even more reliable, the matches featured an interesting twist: every series of deals was played twice, with both competitors getting a chance to play each side of the cards. This is possible because programs can be restarted with no memory of past events. The "duplicate matches" ensure that both programs have nearly equal opportunities, despite the lucky outcomes that can occur in each game.
The second place finisher in both tournaments was 'Bluffbot', written by Teppo Salonen, a hobbyist programmer based in Irvine, California. Originally from Finland, Salonen became interested in developing poker bots for the Poker Academy platform. Other competitors were from Monash University in Australia, and an independent entry from Denmark.
A well-publicized entry from Carnegie Mellon University in Pittsburgh failed to live up to expectations, losing all of its matches except for one win against the last place finisher. The CMU program only played in the slow event, because it required about a minute to compute its strategies.
The top programs were all based on mathematical game theory. The Bluffbot and CMU entries were developed over the past year, expanding on the ideas published by the Alberta team in 2002.
Poker is particularly interesting to computer scientists, because it has many properties not found in other games. "Poker is a nice well-defined problem for studying some truly fundamental issues, like how to handle deliberate misinformation, and how to make intelligent guesses based on partial knowledge", explained Billings. "Good solutions in this domain could have an impact in many other computer applications."
"This was the first AAAI poker championship, and next year is going to be a whole different ballgame", added Bowling. "We expect to see a lot tougher competition, with programs that do a much better job of learning and adapting to each opponent".
Normal pace event: UofA Blfbot Monash Teddy Average U Alberta (Canada) X +0.05 +0.72 +0.41 +0.39 Bluffbot (Finland) -0.05 X +0.53 -0.19 +0.10 Monash (Australia) -0.72 -0.53 X +1.17 -0.03 Teddy (Denmark) -0.41 +0.19 -1.17 X -0.46 Slow pace event: UofA Blfbot CMU Monash Average U Alberta (Canada) X +0.11 +0.18 +0.73 +0.34 Bluffbot (Finland) -0.11 X +0.12 +0.52 +0.18 CMU (U.S.A.) -0.18 -0.12 X +0.65 +0.12 Monash (Australia) -0.73 -0.52 -0.65 X -0.64The numbers in the crosstables reflect the average number of bets won per game played. For example, in the game of $10-20 Limit Texas Hold'em, the UofA defeated CMU with an average win rate of $1.80 per hand played (however, no real money was at stake).