http://www.mcclatchydc.com/100/story/43658.html
Posted on Thursday, July 10, 2008
Battle of chips: Computer beats human experts at poker
WA****NGTON .. Human pride took a hit 11 years ago when IBM's Big Blue
computer beat world chess champion Gary Kasparov. Now it's poker
players' turn to be humiliated by a machine.
A computer system called Polaris outperformed some of the world's top
players last weekend at a human-vs.-machine competition in Las Vegas.
The score was computer 3, humans 2, with one draw.
If you think it should be easier for a computer to win at poker than
at the highly intellectual game of chess, think again. The human
element makes poker a much more complex challenge.
"Poker is a completely different game," said computer scientist
Michael Bowling, the leader of a Computer Poker Research Group at the
University of Alberta, Canada.
"In chess or checkers, you have perfect information. There are no
secrets on the board," Bowling said. "But in poker you don't know the
other person's cards. The basic computer techniques used in chess
can't help you in poker."
The poker computer project may have practical applications beyond the
card room. For example, Bowling said poker-like skills might be useful
in bidding auctions where multiple companies are competing for
government contracts or buyers are hunting deals on eBay.
"There is a lot of uncertainty there," he said. "Should you wait or
bid? The same things apply in poker."
Bowling's team launched Polaris five years ago as a project in
artificial intelligence. At first it did well against amateur players
but couldn't beat professionals. Last year, it narrowly lost a match
against two poker pros in Vancouver, British Columbia.
This year, a stronger version of Polaris — one that learns how to
adapt to an opponent's strategy in midgame — triumphed over seven
top-ranked humans drawn from the online poker-training site
www.stoxpoker.com.
So far, the system plays a relatively simple game of two-person Texas
Hold 'em. The next goal is to take on games of three or more players.
"That's very challenging," Bowling said. "There is no perfect strategy
to play against multiple players."
Unlike Big Blue's IBM supercomputer, the Canadian team used a cluster
of five small, off-the-shelf computers linked in a network to prepare
its strategy before the game.
The system repeatedly played 8 billion games against itself to devise
multiple strategies, each slightly different. Some strategies were
more aggressive, others more passive.
When it came time for the match, a laptop was sufficient to manage the
system. The laptop, of course, showed the perfect poker face.
During the game, Polaris analyzed its human opponent's style of play
and adjusted its strategy to meet it. For example, the system plays
more aggressively in order to get the human to give up and fold his
cards.
"The computer pushes humans to make more decisions. More decisions
mean more mistakes. Aggression raises errors," Bowling said.
Bowling conceded that it will take a few more competitions for human
poker players to accept that a computer can outdo them.
"Now that we've lost, I'm itching for a rematch," said Jay Palansky,
one of Polaris' opponents.
|
