Sign In

Communications of the ACM

ACM TechNews

How to Build an ­nbeatable Poker-Playing Robot

View as: Print Mobile App Share:
A poker game.

Researchers at the University of Alberta developed a poker-playing program that learned the game by practicing at a rate of about 2,000 games per second.

Credit: The Atlantic

The allure of building superior poker-playing computer programs is the chance to tackle the challenge of dealing with missing information.

A good game-playing algorithm backed by heavy computing muscle can bypass the problem of a lack of available data, and this was the strategy followed by University of Alberta researchers in their development of their Cepheus program.

Cepheus mastered poker by practicing the game over and over at a rate of about 2,000 games a second, and employed a "regret minimization" algorithm, which reflects on past games and learns from errors. The bot demonstrates it is possible to work out an optimal strategy in complex situations, and the range of scenarios to which such algorithms could potentially apply is vast, according to the researchers.

Still, the University of Alberta's Jonathan Schaeffer believes humans will continue to outclass game-playing algorithms by their ability to quickly make assumptions about opponents with little available data.

Fellow University of Alberta researcher Michael Johanson notes human players' practice of aggression to beat opponents is another skill computer programs are attempting to mimic. "An aggressive strategy that puts a lot of pressure on opponents, making them make tough decisions, tends to be very effective," he says.

From The Atlantic
View Full Article


Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account