The First Man-Machine
Poker Championship

Computer takes on poker aces to see who's the busted flush

Ian Sample, science correspondent
Monday July 23, 2007
The Guardian

A showdown pitting human brains against artificial intelligence goes ahead this evening when two professional poker players take on a computer in the world's first such man-machine challenge.

Phil Laak and Ali Eslami will play Polaris, the most sophisticated poker-playing program yet written, the product of years of research and refinement by a team of artificial intelligence experts at the University of Alberta in Canada.

The challenge will play out over two days and 500 hands of Texas hold 'em at the Hyatt Regency hotel in Vancouver, with the players gambling for a total prize pot of $50,000 (£23,000).

Jonathan Schaeffer, the lead scientist behind Polaris, said that, even though his program had the perfect poker face, it was not the favourite to win.

Nevertheless, he promised to make his opponents work for their prize money. "I'm not nervous," he said. "Everyone expects the humans to win."

Last week, Dr Schaeffer published details of an artificial intelligence draughts-playing game that cannot be beaten.

The poker challenge has been organised by the American Association for the Advancement of Artificial Intelligence as part of its annual meeting. The two poker players will play against Polaris simultaneously in adjoining rooms.

To avoid either side later blaming a loss on bad cards, the games are designed to eliminate the influence of luck.

Whatever cards are dealt to Mr Laak will automatically be dealt to the computer playing Mr Eslami and vice-versa.

Polaris has been written to learn its opponent's playing strategy and identify its weaknesses.

"The program knows it has to bluff. In poker, if you don't bluff you're playing a bad game, and if you bluff too much, you're playing a bad game," Dr Schaeffer said.

Unlike draughts and other games, developing computer programs to play poker is difficult because of the number of possible decisions at each stage, and the lack of information a player has on an opponent's hand.

The games will be watched by an audience with the players encouraged to talk aloud about their decisions and the computer's strategy. "I won't be able to read its face, but equally, the computer won't know if I'm having a manic moment or if I'm starting to rush," Mr Laak said. "I can say out loud: 'Computer, I'm going to bluff you now.' But it's a strong program. It's going to memorise my betting patterns right away and my game is not perfect," said Mr Laak.

"I think we'll be surprised, confused and saddened if it slaughters us, or we slaughter it."