Card sharps clean up as computer counts its losses
26/07/2007
Dan Glaister
They
didn't blink. They didn't tell. Even after Polaris took the second round, the
Unabomber and his accomplice stuck to their game. Finally, deep into the night,
after four rounds of Texas Hold 'Em, Polaris folded once and for all. The game
was over. The humans had beaten the machine. Just.
Far away from the
kitsch glamour of Las Vegas, with not a showgirl or a hustler in sight, two
professional poker players from Los Angeles took on a computer programme in a
hotel in Vancouver on Monday and Tuesday.
Billed as the "First
Man-Machine Poker Championship", the event staged at the annual meeting of the
Association for the Advancement of Artificial Intelligence offered prize money
of $50,000 (£25,000) to the winner of four hands of poker.
For Ali Eslami and Phil "the Unabomber"
Laak - so named because he wears a hooded sweatshirt and sunglasses - the money
was small change, but the stakes were high.
"I literally felt the same
feeling that you would have if you beat 500 people in a tournament and won
$1m," Mr Laak said after the game, which ended to the sound of whoops and
cheers from the watching crowd of hundreds as the humans vanquished the
computer.
"We won, not by a significant amount, and the bots are
closing in."
Playing against the computer was more exhausting than any
other game he had played, Mr Eslami said. "I really am happy it's over. I'm
surprised we won ... it's already so good it will be tough to beat in future."
Polaris has been 16 years in development at the University of Alberta
in Canada. While computers have previously mastered humans at games such as
chess and draughts - chess world champion Gary Kasparov was no match for IBM's
Deep Blue a decade ago - researchers have been keen to develop poker software
in the belief that it might be more applicable in other fields.
"I
contend that poker is harder than chess for computers, and the research results
that come out of the work on poker will be much more generally applicable than
what came out of the chess research," Jonathan Schaeffer, the lead scientist
behind the programme, told the New York Times.
His team won the world
draughts championships in 1994 and said earlier this month that they had
developed a programme that could not lose at draughts.
But poker is
altogether more sophisticated. The computer, like its human counterparts, knows
how to bluff. And unlike its chess and draughts counterparts, which require an
enormous memory to consider every potential move, Polaris runs for weeks before
a game, creating 10 different "bots" which have their own playing style.
Organisers of the event attempted to remove the element of luck by
putting the human players in separate rooms. The computer played both humans
simultaneously, with one human and the computer dealt identical hands.
If Mr Eslami was dealt a royal flush in one room, Polaris would have a
royal flush in its game against Laak. The separation also eliminated one of the
most important aspects of poker, the "tell" or giveaway signal when a facial
tic or restlessness can reveal a player's strategy.
The first round -
each round consists of about 500 hands - ended in a draw, although the
computer's winnings were marginally higher than those of the humans. But
Polaris won the second round heavily, leading Mr Eslami to comment that
"Polaris was beating me like a drum".
But on the second day, things
changed, and a strategic gamble by the computer's human handlers may well have
cost Polaris its $50,000.
For the first two rounds, Dr Schaeffer's team
ran a single bot. But for the third round, they substituted a more
sophisticated programme that was supposed to add a level of "learning",
deploying different bots as necessary. The strategy backfired. The humans
easily won the third round, levelling the contest.
In the final round,
although Polaris won a sizeable pot of $240 with a royal flush beating Mr
Eslami's three-of-a-kind, the humans again prevailed, winning $570.
Darse Billings, a scientist on the Polaris team and a former
professional poker player, thought the humans had played "brilliantly". But he
had a prediction for the future: "I wouldn't be surprised if we can beat them
tomorrow," he said.