The First Man-Machine
Poker: Man still owns machine -- for now
Web Posted: 08/01/2007 10:34 PM CDT
Express-News Staff Writer
In 1996, world champion chess player Garry Kasparov scored one for the humans when he defeated IBM's computer program Deep Blue in a series of matches.
After losing to the program in the opening match, Kasparov figured out how to beat it and ended up winning the best-of-7 series 4-2.
So what did IBM do? It beefed up Deep Blue and set the stage for a rematch the following year. Deep Blue defeated Kasparov in 1997 by a margin of 31/2-21/2 to mark the first time in the history of chess that a computer was able to pin a loss on a world champion.
Now the attack of the machines is focused on poker. The computing team at the University of Alberta in Canada created a poker-playing computer named Polaris that's designed to compete at the game's highest level in limit Hold 'Em.
To prove the mettle of its machine, the team pitted Polaris against professional poker players Phil Laak and Ali Eslami last week in a series of matches deftly labeled the Man-Machine Poker Championship to see how it would fare.
This wasn't the first time humans squared off against a computer on the poker table. In 1984, in front of a live TV audience, Bob Stupak squared off against Mike Caro's computer program ORAC (Caro spelled backward) in a $500,000 challenge match. Stupak won the match, but it was a heated contest.
Instead of providing a single, formulaic approach to the game that would've been easily exploitable by the human professionals, the Alberta team programmed multiple strategies into Polaris. The program was designed to then determine which strategy was most effective, and use that strategy to attack its opponents.
Since it's a computer, elements such as physical tells, language glitches and nervous twitches were obsolete. The only information available to both human player and computer were betting patterns and standard odds. The strength of Polaris is that it's impervious to emotion.
The stakes were simple: Laak and Eslami would each play four, 500-hand $10-$20 limit Hold 'Em sessions against Polaris at the same time, heads-up. Laak would play in one room and Eslami would play in the other.
They would be dealt identical cards vs. the computer and the computer would also hold the same hand. If Laak and Eslami were to combine for a win of 25 or more small bets, they would take the session. A 25-bet win by Polaris would produce a win for the machine, and everything in between was considered a draw.
There were financial incentives for Eslami and Laak — they stood to earn as much as a split of $50,000 with a sweep of Polaris and $2,500 for each session won. A Polaris victory would simply void the money, since computers don't need to be paid, just plugged in.
Match 1 ended in a draw after the human team produced a meager win but failed to achieve the needed 25-bet advantage. The second-round match swung heavily in Polaris' favor, where the program destroyed Laak and Eslami by 95.5 small bets.
"Polaris was beating me like a drum," Eslami told the New York Times after losing $2,515 simulated dollars in the round.
The loss pinned the humans down 0-1-1, forcing victory in the final two matches to achieve total victory. Remarkably, Laak and Eslami did just that, producing small wins that barely covered the needed 25-bet margin.
Laak was so impressed with the play of Polaris, he said after the match to Card Player magazine that he'd put it up against any human in the world.
Maybe so, but for now, humans can brag that they still own the realm of limit poker. However, just as Deep Blue programmers only needed a year to reconfigure their machine to defeat the world champion in chess, it would not be a surprise to see Polaris achieve victory in the near future.