The First Man-Machine
Poker Championship

Humans out-bluff poker-playing computer

Two Professional poker players have triumphed in a two-day battle with a machine called Polaris, but not before it took a game off them

Jonathan Richards

Humans are still better at playing poker than computers, but only just.

Two professional poker players have narrowly beaten what scientists described as the 'most intelligent' poker-playing computer to date.

Phil Laak and Ali Eslami, who rank among the world's best players, lined up against 'Polaris', a machine designed by Canadian academics, at an exhibition match in Vancouver.

Over two days in an overheated conference room, they eventually defeated their nemesis -- which was following in the footsteps of Deep Blue, the chess-playing computer that defeated Garry Kasparov in 1997 -- but not before losing one game and drawing another. Out of the four games of Texas Hold'em, the humans won two, with Polaris bettering them just once.

"I'm surprised we won," said Mr Eslami, 30, after the match, adding that it had been the most exhausting of his career. "It's already so good it will be tough to beat in future."

A crowd of weary computer scientists who were attending the annual conference of the Association for the Advancement of Artificial Intelligence broke out in cheers after the final hand was played at 11pm on Tuesday night.

The two players sat in separate rooms with the human and computer hands reversed, so that a poor hand in one room would translate to a strong one in the 'mirror' game.

At the end of the tournament the total chips won by the pair were compared against the computer's takings.

Darse Billings, the lead architect of Polaris at the University of Alberta and himself a former professional poker player, said that, despite losing, his creation had played "brilliantly". "I wouldn't be surprised if we can beat them tomorrow," he said.

Poker poses unique challenges for scientists devising machines that can play the game because of the myriad uncertainties regarding what cards the opponent has and how he or she may play them.

Other games, such as chess and draughts, always start the same way, and there is a large but finite number of moves in which it may finish, with play governed by a consistent set of rules.

In poker, "you dont have perfect information about what state the game is in," Dana Nau, a professor of computing science at the University of Maryland, said. "That means that when your opponent does something, you can't be sure why."

Earlier poker-playing computers made decisions based simply on the strength of the hand, which meant that human opponents could intuit from the aggressiveness of play what cards the computer had, and adapt their strategy accordingly.

"It's mandatory for you to understand how the other guy approaches the game. This is critical information in poker, and it's not true of any of these other games that we've studied in academia," said Mr Billings, who has working on Polaris for 15 years.

Last week scientists said they had created the ultimate draughts-playing computer, called Chinook, which could be battled to a draw with a perfect strategy but could never be beaten.