The following article appeared in the June 12, 2003 issue of the Edmonton Journal newspaper (front page).

The text of the story (and errata) appear below.




Poker playing computer will take on the best

Ryan Cormier
Journal Staff Writer

There's a new poker player that never sweats, never gets tired, never tips a hand and can still bluff with the best of them.

University of Alberta artificial intelligence researchers bet their new poker computer program will be the best player in the world, perhaps within a year.

"We've made some really fantastic progress over the last year and a half," says Jonathan Schaeffer, who heads up the university's Games Research Group.

PsOpti -- the pseudo-optimal poker program -- is the latest version in the team's decade-long attempt to create the ultimate poker player. The program has some crucial tools, including the ability to bluff.

"You have to bluff," says Schaeffer, who already has a world-champion checkers computer program under his belt.

"If you do not bluff, you're predictable. If you're predictable, you can be exploited."

The program is based on a mathematical formula developed by Nobel Prize winner John Nash, who was featured in the movie A Beautiful Mind. Game theory is a branch of mathematics that studies the interactions between people, companies, or countries who are in competition.

[See POKER / A14]

[picture: Darse Billings, a PhD student at the University of Alberta, has written a computer poker program that is garnering international attention. - John Lucas, The Journal]

Computer will bluff, but plays carefully

[POKER Continued from A1]

"The formula tries to find a fair outcome for the game," says Darse Billings, a former professional poker player and a PhD student working on the project.

"It's a defensive strategy. It doesn't guarantee that you will win, but it guarantees that you won't lose."

The technical description of the program beat out 1,200 competitors to win the distinguished paper award at the International Joint Conference on Artificial Intelligence to be held in August in Acapulco, Mexico.

Schaeffer says the program has more benefits than playing smooth poker.

"A lot of the original research in games involved games with perfect information. Like in chess, you always know where the pieces are, there's nothing hidden," Schaeffer says.

"Games with imperfect information, like poker, are actually much more important in the real world than games of imperfect information."

Figuring out how to reason with imperfect information has many benefits: in international negotiations, in poker, or in buying a car.

Billings says using a computer to deal with unknown information, such as the dealer's bottom-line price of a car or how many aces your opponent is holding, is no easy task.

"How do you get the computer to handle deliberately faulty information?"

The program plays a version of poker called Texas Hold'em, in a style Billings says humans will not be used to: "At times, it seems rather conservative, but it is also very tricky, by design."


Correction / Omissions: Darse Billings is the principle architect, but did not write the program.

The Poki software system, including the web Java client, was written by M.Sc. graduate Aaron Davidson.

Programmer / Analyst Neil Burch did most of the coding for PsOpti, including the linear programming systems.

M.Sc. student Terence Schauenberg has contributed to the recent work on adaptive programs.

Professors Jonathan Schaeffer, Duane Szafron, and Robert Holte have made valuable contributions to design, algorithms, and scientific methodology.