Man Versus Machine at the Poker Table

by Lou on July 29, 2007

John Henry said to his captain,
“A man ain’t nothing but a man,
But before I let that steam drill beat me down,
I’ll die with the hammer in my hand.”

—American Folk Song

The man-versus-machine allegory is as old as time, or at least as old as the folk song, John Henry. But in a poker game between man and machine in late July, a software program running on an ordinary laptop computer fought a close match before losing to two professional poker players.
Billed as the First Man-Machine Poker Championship, the match pitted Phil Laak and Ali Eslami against Polaris, a program written by a team of artificial-intelligence researchers from the University of Alberta.
University of Alberta researchers created the program that won the world checkers championship in 1994. This year they trotted out a checkers program that could not lose, and at best could be tied. In short, they completely solved the game of checkers.
While chess has yet to be solved, in 1997 Deep Blue, developed by IBM researchers, beat then world champion Garry Kasparov. Given the complexity of chess and Kasparov’s copious skills, who at the time of the match was the highest rated chess player in history, it was a significant accomplishment. Ten years later, it still is.
But poker is not checkers or chess and is much more difficult to teach to a machine. Chess or checkers lend themselves to number crunching, which is why poker–a game of “incomplete information”–is hard to reduce to mathematics. While a computer can calculate the odds perfectly, it doesn’t know when a player is bluffing, or just on tilt.
Although computers can approximate these judgments by searching for tendencies after learning how an opponent plays certain kinds of hands in certain situations, it still hasn’t learned to handle incomplete information as well as it handles checkers and chess problems. Much of the artificial intelligence work currently being done with poker goes toward constructing algorithms that track opponents’ betting habits.
The Man versus Machine Poker Challenge eliminated as much of the luck as possible by playing duplicate poker, in which one teammate plays the same hand as his teammate’s opponent, and vice versa. By eliminating luck, duplicate poker reduces the game to its strategic fundamentals.
Phil “the Unabomber” Laak and Ali “the Prince” Eslami, edged out the University of Alberta’s computer poker program Polaris by about $400 over the course of the two day match.

Because in there isn’t always one best move, the game tree approach is tough to use in poker. Expert poker players will also adjust their play to exploit other players’ style, and Eslami and Laak were able to adjust and adapt to Polaris faster and more accurately than the bot was able to handle the pros’ play.

More work remains to be done before Polaris or one of its next generation descendants is a sure fire favorite to beat a skilled human player. And it’s a longer way still, before a series of algorithms can solve the game of poker as adroitly as the artificial-intelligence researchers from the University of Alberta solved the game of checkers.
While today’s drilling machinery can hammer their way inside a mountain a whole lot faster, harder, and longer than John Henry, their poker playing cohorts still have a ways to go.

Comments on this entry are closed.

Previous post:

Next post: