The First Man-Machine
Poker pros out of luck in battle with 'bot
U of A team pits man against machine in experiment that negates effect of good fortune
David Staples, The Edmonton Journal
Published: Monday, June 11, 2007 [page A2, "Top Copy"]
EDMONTON - Poker is a game of luck and skill, but to find out who really is better at the game, man or machine, a group of University of Alberta computer scientists has come up with an experiment that negates luck as a factor.
In the experiment, the U of A team will pit its state-of-the-art poker-playing laptop program, Polaris, against two top-flight professional poker players, Phil (The Unabomber) Laak and Ali Eslami.
They will play four sessions of Texas Hold 'Em at an artificial intelligence conference July 23 and 24 in Vancouver.
Above, the U of A's Jonathan Schaeffer heads a team that will pit its poker-playing supercomputer, Polaris, against two poker pros, Phil (The Unabomber) Laak and Ali Eslami.
Above, Laak is shown playing the U of A's Vexbot in November 2005 as his girlfriend, poker player Jennifer (The Unabombshell) Tilly, watches.
"The question is: 'How good are humans in poker in relation to computers?' and we don't have an answer for that because the previous results have been clouded by this luck versus skill issue," says Jonathan Schaeffer, chairman of the U of A's computer science department. He designed Chinook, the checkers program that beat world checkers champion Marion Tinsley in 1994.
In one room at Vancouver's Hyatt Recency Hotel, Polaris will play Laak.
In another room, Polaris will play Eslami. Polaris will get the same cards that Laak is dealt in the other room.
"If Phil (Laak) plays our poker program and he gets incredibly lucky and the cards are just all going his way, in the other room the same cards are happening but are just reversed, which means Ali is going to get incredibly unlucky and all the cards aren't going to go his way," Schaeffer says.
In the end, the chips for the two humans will be counted up against the chips for Polaris, the reigning computer program poker champ, to discover who has won.
"The luck factor is almost completely gone. It's all skill now," Schaeffer says.
$5,000 and pride at stake
If the two pros win a session, they will split $5,000.
Laak says he hopes to break even against Polaris, which he refers to as a "bot." He plans to train relentlessly against other bots in order to prepare.
"There's no doubt a lot of pride is at stake," he says, adding he respects the program's capability. "For me to play that thing in the long run is no bargain. I'd probably do better not playing it if I was trying to beat it for money."
Schaeffer doesn't know what to expect. "It could be a great day or it could be an embarrassment," he said.
Like it or not, the bots are on the rise, Laak says.
"We're already at the point where artificial intelligence crushes players that are unsophisticated, beats handily intermediate players, and loses small or wins small against savvy opponents... For Round 1, I'd say the bots have it."
This is the second time Laak has faced a U of A poker program. In a 2005 match in Las Vegas, Laak easily beat Vexbot, a predecessor of Polaris, partly because he played better, but also because he had far more luck that day -- as Laak himself readily admits.
Laak will be a tough opponent because he trains against commercial versions of the U of A program, which have some of the same tendencies as Polaris, Schaeffer says. Eslami is a higher-rated and more consistent player than Laak at this particular poker game, which has a limit on betting.
Laak brings one more quality to the experiment -- star power. He's a flamboyant character. He dates actress and poker player Jennifer Tilly, who is known as the Unabombshell.
"Phil is a showman," Schaeffer says. "He loves an audience, he's got great humour, he likes to joke, he's very emotional, and since computers can't hear anything, he's agreed he's going to talk out loud as he plays ... .
"If he thinks the program is bluffing, he'll say so."
Program adjusts to human play
The U of A team, led by Prof. Michael Bowling, has 15 members and includes former poker pro Darse Billings.
The scientists are now fine-tuning Polaris. The program is likely to combine two elements of past U of A machines. First, it will have stored in its memory banks the optimal strategy for most poker situations. Second, its program will study the way its human opponents are playing and attempt to learn and adjust to take advantage of the style of play.
In Las Vegas in 2005, Vexbot adjusted its style in an attempt to exploit Laak's tendencies, but it didn't always make the best choices, Schaeffer says.
"It was aggressive, it had a learning component, but it had an unstable learning component. This new program is more stable."
Against humans, Laak says he has often won simply because his opponents make mistakes, but the computer is able to almost always make the optimal move, so he won't have that advantage.
"I have a solid understanding of poker and math, but their bot has seen so many more hands than I'll ever probably see in my lifetime.
"It's hard to beat game theory."
The poker challenge isn't as significant in the public imagination as IBM Deep Blue's victory over world chess champion Garry Kasparov in 1997, but it's more significant in terms of artificial intelligence science, Schaeffer says.
Chess has proven to be a much easier game for computers to play than poker because all the chess pieces are on the board and there are no unknowns. In poker, you don't know your opponents' cards, so you must make your decision based on imperfect knowledge.
"Poker is a much better representative of real-world situations with imperfect information, negotiation, bluffing and misrepresentation," Schaeffer says. "This makes it much more interesting. From a scientific point of view, it's a harder problem because of that."
The Edmonton Journal 2007