The First Man-Machine
Poker Championship

Poker players dealt date with Polaris the computer

Two humans face best machine in four-match competition

Tuesday, July, 24, 2007

Laura Payton, The Province

VANCOUVER - Two poker players are trying to prove they know when to hold and when to fold against the reigning world champion computer poker program.

Phil Laak, a professional poker player, and Ali Eslami, a player experienced in high-stakes games, are taking on the University of Alberta's latest generation of artificial intelligence poker-playing technology.

The program, known as Polaris, defeated all other programs last year at the Association for the Advancement of Artificial Intelligence annual conference. This year, Polaris' creators want to see how it handles human challengers. Laak has a degree in mechanical engineering and Eslami was a computer consultant before becoming a poker professional.

The two are simultaneously taking on the program in different rooms for a four-match competition yesterday and today in Vancouver. To eliminate the element of luck, one player in each room will get the same hands -- Polaris in one room and the human in the other room.

Polaris is programmed with multiple styles and will be more or less aggressive depending on how the programmers want it to play. In the first game, Polaris focused on the rules of poker and narrowly beat the humans by a combined total of $70. From there, Polaris will get more aggressive.

"In some sense, we [were] aiming for a draw on Match 1," said Michael Bowling, leader of the team that created Polaris. The program needs to get used to the humans' style of play and learn what mistakes they tend to make.

In 35 hands, Polaris can "build an opinion of somebody that's usually very accurate as to the kind of poker player [it's] playing against."

For the academics behind Polaris, the game is about more than just man versus machine.

"Academics would call [poker] a game of imperfect information," said Bowling. "It's a game where the player who has to make a decision at a point in time doesn't know the key piece of information they'd really want to know to be able to choose their action -- which is the cards the other player has."

Bowling and his team used a type of applied math called game theory to mathematically describe poker and allow Polaris to do things a computer program wouldn't normally do, like bluff. "Our goal is to study imperfect information games -- games in the sense of a mathematical model for interaction between intelligent entities," said Bowling. Real-world applications include helping businesses make crucial decisions when they're missing key information, such as what their competition is doing.

lapayton@png.canwest.com