The First Man-Machine
Poker computer program will try to outplay humans
Updated Mon. Jun. 11 2007 8:20 AM ET
John Cotter, Canadian Press
EDMONTON -- University of Alberta researchers are betting their poker-playing computer program will beat two of the sharpest human professional players in the world, and they're not bluffing - yet.
The developers of the Polaris program have challenged Phil (The Unabomber) Laak and Ali Eslami to 2,000 hands of Texas hold'em. The $50,000 man-versus-machine poker match will not only be fun - it will help test advances in artificial intelligence, said Jonathan Schaeffer, leader of the computer science team that created Polaris.
"We have developed a format that has helped us factor out luck and make it into a scientific experiment to determine how good humans are relative to the best program in the world," Schaeffer said.
"The goal is to eventually produce a poker program that is stronger than all human players."
Texas hold 'em is considered to be the most popular variant of poker played today, thanks in part to plenty of exposure on television and the Internet.
Players are dealt two cards face down and then share a series of five community cards to form the best hand. Some games feature unlimited betting.
The match will be played on July 23 and 24 in conjunction with the Association for the Advancement of Artificial Intelligence Conference in Vancouver.
Polaris is actually a number of different computer programs with different characteristics.
One is very aggressive, but doesn't take into account the playing style of opponents.
Another program actually learns from the strengths and weaknesses of other players and will adjust its style accordingly.
All of the programs are masterful at one of the most interesting aspects of playing poker - bluffing.
"There is a mathematically optimal rate at which you should bluff. Computers can calculate that. Humans don't understand the mathematics of poker. If they bluff too much, you can exploit them and win money," Schaeffer said.
But the mechanical card sharks do have a weak point.
Programs that don't learn can be exploited if human opponents are skilful and savvy enough to pick up on the computer's playing style.
But Laak and Eslami won't know which program they will be playing against this summer, and the researchers will be able to shift to a different program if the computer's game starts to fold.
"Humans are tricky and interesting opponents," Schaeffer said.
Laak, who wears his trademark hoody and sunglasses whenever he sits at the card table, is a veteran of the poker circuit and sometimes works as a television commentator.
With a degree in mechanical engineering, Laak has tangled with computer poker programs before.
Facing a check-raise from a computer during a match in 2005 he once exclaimed, "If that is a bluff, it's over for humanity," and promptly folded. It was a bluff.
Laak selected Eslami, a professional poker player who is also a computer consultant, to be his teammate in the July match.
Together they will play against the computer. Each match will consist of 500 hands played in four sessions over two days. The experiment is designed to squeeze as much luck out of the game as possible.
At the end of each session the combined bankroll of the humans will be compared with that of the computer to determine the winner.
"This will be a stronger and cooler version of the program," said Laak, who said he loves computers and solving puzzles almost as much as poker.
"Poker's indeterminate nature makes it a real challenge for programmers. It's a bunch of computers solving a puzzle. What is cooler than that?"
Eslami said playing the computer program will be a challenge, but he's confident of victory.
Top human players are better at quickly discerning patterns of play, he said.
But when you're playing a machine, important aspects of the game, such as reading body language, are not a factor.
"It is not so much a psychological game as it is with human opponents," Eslami said in an interview from Las Vegas, where he is competing in the World Series of Poker tournament.
"But in the long run, computers could become a good player for sure and maybe in the long, long run could be one of the best players, just like in chess."
Schaeffer is also mindful of the chess analogy, noting the 1997 match between IBM's Deep Blue' program and Gary Kasparov, the world chess champion at the time.
The computer program won that multi-game match.
Duplicating such a victory will be more challenging using the game of poker, where Lady Luck is always a factor.
"The difference is that chess is a game of perfect knowledge, meaning there is nothing hidden from the players," Schaeffer said.
"In poker you can't see your opponent's hand and you don't know what cards will be dealt.
"This makes poker a much harder challenge for a computer scientist from an artificial intelligence perspective."