Friday, March 9, 2012

Poker and Artificial Intelligence

Today, I was able to go to a talk by Michael Bowling from the University of Alberta.  He was visiting the University of Denver, and he spoke to a crowd of CS and other science types in Olin Hall. 

He did a good job of explaining how the game of poker can be used in the course of programming artificial intelligence into computing systems.  This paper (Poker as a Testbed for Machine Intelligence Research PDF) from his lab in 1998 covers some of the basics.

He showed a good clip from the movie Rounders, and then he explained some of the basic rules of the game and some of the strategy involved.  (I still can't believe that Matt Damon didn't consider that KGB might have had pocket aces during their initial game in the movie....) Poker is more like real life for a number of reasons.  You have to deal with incomplete knowledge of the game (you can't see the cards dealt to your opponent), and the opponent can try to bluff his or her way out of a jam.  This is not the situation for chess and other types of board games.

Anyway, I learned that it now possible for computers to beat world class professional poker players on a head to head basis.  They programmed the Polaris system to have different personalities, to be agressive or passive or bluff sometimes or whatnot.  The Alberta folk will continue to work on poker games that involve three or more players.  The programming gets a bit more complicated with more players.  Polaris was programmed to work with Texas limit hold'em so that the raises were set at particular levels.  There are just too many raise variables to consider in no limit hold'em, but they could work on this for future projects. 

Other links to check out.

No comments: