Playing Your Cards Right: Poker Comes out of the Back Room and into the Computer Science Lab

Article excerpt

Poker comes out of the back room and into the computer science lab

Darse Billings plays poker for a living. Just a few years ago, he was a computer science student at the University of Alberta in Edmonton. Interested in games, he had chosen the development of a poker-playing computer program as his thesis topic.

"I discovered that very little work had been done on computer poker despite the many potential benefits of that research," Billings says. Although he did not create a working program, he learned enough about the game along the way to begin playing professionally--which he has done since completing his thesis in 1995.

"Understanding the theory and mathematics of poker gives you a solid foundation," he contends. "This alone can put you ahead of the vast majority of players. Beyond that, you learn methods of analysis. It teaches you how to think about a given poker situation and that enables you to make more effective decisions in the heat of battle."

Billings now also serves as a consultant to a team of researchers at Alberta intent on developing a program that can play poker at the level of the best human players. "It's wonderful working with him because he understands computers and he understands poker," says Jonathan Schaeffer, who heads the group.

In an earlier effort, Schaeffer and his coworkers had created a checker, playing computer program named Chinook, which could beat the world's top players (SN: 7/20/91, p. 40).

The current version of the Alberta poker program--named Loki for the Norse god of mischief and chaos--already plays a strong game, Schaeffer says. "But it isn't ready yet to win a world championship."

The first public demonstration of Loki is scheduled to take place later this month at the Fifteenth National Conference on Artificial Intelligence in Madison, Wisc.

Games have long been an important focus of research in computer science and artificial intelligence. With clearly defined rules and specific goals, "they are wonderful domains for testing ideas," says Matthew L. Ginsberg of the University of Oregon in Eugene.

Researchers have programmed computers to play chess, backgammon, bridge, Go, Othello, Scrabble, and numerous other games--in several instances eventually achieving a world championship level of play (SN: 8/2/97, p. 76; 8/16/97, p. 100).

Poker is an example of a game of incomplete information in which chance plays a role. Whereas a chess player sees the disposition of all the pieces all the time, a poker player sees only some of the cards--drawn or dealt from a shuffled deck--that are in play.

"You don't know what cards your opponent has," Schaeffer says. "All you can do is make educated guesses."

"One of the fundamental problems in computer science is how to deal with information that may be erroneous, unknown, or incomplete," Billings adds. Poker provides an excellent domain for investigating problems of decision making under uncertain conditions. Researchers can study such issues as risk assessment and management (betting strategy), opponent modeling (exploiting weaknesses in an opponent's play), and deception (bluffing).

Early efforts to model poker, some dating back to the 1950s, were generally unrealistic and rather limited, Billings says. Moreover, although there are many commercial poker-playing computer programs now on the market, they range widely in quality and lack the strategic flexibility and learning capability that a world-class player must have.

"Loki is probably better than commercial programs at evaluating the strength and potential of its cards," Billings notes. "Loki is also better at observing each opponent's actions over time and then adjusting its play accordingly."

Hundreds of books have been written about how to play poker. Billings contends that the vast majority give flawed advice on strategy. The typical level of human play is so low, however, that a contender can be highly successful despite serious misconceptions, he argues. …