Unbeatable AI program for poker

BANGALORE: A team of University of Alberta computer scientists is betting that it's only a matter of time before the world's best poker players are folding their cards in the face of a digital opponent.

The artificial intelligence (AI) researchers have moved one step closer to creating an unbeatable computer poker program. An account of their most recent program, called PsOpti — for pseudo-optimal poker program — will receive the top paper award at the world's premier AI meeting in August.

"This past January we arranged for one of the best players in the world to play PsOpti. After 7,000 hands it was clear that the human was the better player, but it took him a long time to figure out how to win against the program," says Dr. Jonathan Schaeffer, who heads up the University of Alberta's Games Research Group.

PsOpti is the latest version of the Group's decade-long NSERC-sponsored quest to build a world-champion silicon-based poker pro. Dr. Schaeffer already has one world man-machine games title under his belt — his Chinook program jumped to checkers world champion in 1994.

"Poker is much more computationally interesting and difficult than checkers or chess," says Dr. Schaeffer. "Checkers is a perfect information game. When you look at the board you know where all the pieces are. Poker is different. It's an imperfect information game. You don't know your opponent's cards. And that means you have to make inferences about what they're doing. That implies that it is in the opponent's best interest to deceive you. Bluffing is a critical part of the game."

Compared with Poki, its predecessor, PsOpti has a much better poker face.

"PsOpti is a tough opponent, and very difficult to learn against. Its play is quite inhuman, and that's a good thing," says Darse Billings, a former professional poker player, and now University of Alberta Ph.D. student who is the lead designer in honing PsOpti's skills in its poker game of choice, Texas Hold'em.

In the world of poker, PsOpti is designed as a defensive player. The program is based on a mathematical formula developed by game theory founder John Nash (the subject of the Academy Award-winning film A Beautiful Mind). This mathematical approach emphasizes the optimal strategy to ensure that you don't lose.

The seven-person University of Alberta team's technical description of PsOpti beat out more than 1,200 competitors to win the distinguished paper award at the International Joint Conference on Artificial Intelligence to be held in Acapulco, Mexico, August 9-15.

However, Schaeffer and Billings think they can do even better than PsOpti. They're now at work on a "maximal strategy"-based program, one that will watch for and then exploit its human opponent's weaknesses.

The researchers stress that they're not in this for some high-stakes cash-in and early retirement. "This is a research project," says Dr. Schaeffer. "We're exploring new ways for enabling computers to handle uncertainty, and poker is an ideal domain for scientifically exploring this issue."

The team's computer poker programs have never played for real money. But when their next generation program is ready for the cards to be shuffled, there could be real chips on the table.

"We've had phenomenal results against strong players," says Dr. Schaeffer. "But it's not clear that this extrapolates to a real game against a world-class person who has money on the line. Because then they will make different decisions than they would if they were competing for play money."