Humans best computer in poker match
Phil Laak and Ali Esmali, two professional poker players, face off against a computer program called Polaris and win.
In a poker contest of man vs. machine, the results are in. Humans-2. Polaris-1. (And one draw.)
On Tuesday night, Phil Laak and Ali Esmali, two professional poker players, faced off against a computer program called Polaris, which was developed over 26 years at the University of Alberta. The contest, held at the American Association for the Advancement of Artificial Intelligence (AAAI) conference in Vancouver, Canada, was one of the first scientific poker contests involving real players.
In the end, Polaris beat Laak and Esmali in the first match of 500 hands; tied the second match; then lost the last two rounds. Laak and Esmali walked away with almost $20,000 each, including a $5,000 honorarium for playing in the scientific game.
"It was very challenging and a far more sophisticated bot than what they put out before," Esmali, 30, said in an interview. "Previous versions were much more exploitable and you could adjust your strategy and beat it. (Polaris) used a much more varied strategy."
Esmali, a resident of Los Angeles, said that the play wasn't about the money for him or Laak, but rather about helping advance the field of AI. "It's the implication in society that this technology can be applied to like collaborative problems. That's what's so exciting is being at the beginning of this adventure."
For Jonathan Schaeffer, professor in the department of computing science at the University of Alberta and who's been working on Polaris since 1991, the game was also a win for his team.
"This is the first time any computer program has won a match against a strong human player, and from our point of view, that's a huge milestone for us and AI," said Schaeffer. "But we have a long way to go to beat the best in the world."
The consolation prize for Schaeffer's 12-person team was that Polaris won the Second World Computer Poker Championship, in which as many as 18 computer programs played each other over weeks. The results were announced Wednesday at the AAAI conference.
The University of Alberta team chose to work on an AI program for poker, rather than chess, because it's a game of imperfect information, just like life, Schaeffer said. That means that a player can't know what cards his or her opponents are holding, making the game one of luck and skill.
The man vs. machine poker contest was designed to eliminate the luck factor by dealing the same cards in each hand, but in reverse. So in one hand, Laak might get lucky with two aces, but in the other room Esmali would be unlucky as the computer was dealt two aces. To finalize a winner, officials added the two human scores and two computer scores separately, and the highest number won.
Even though the University of Alberta team lost, they can take heart that it was no easy task for Laak and Esmali.
Esmali said that the game wasn't like professional poker online, when he can typically watch TV or eat some food while playing. "This required really intense focus to play at the level to beat this bot."