Poker-Playing Computer Knows When You're Bluffing
- NHNE News List
Current Members: 656
Subscribe/unsubscribe/archive info at the bottom of this message.
ALBERTA RESEARCHERS DEVELOP A COMPUTER PROGRAM
THAT KNOWS WHEN YOU'RE BLUFFING
By Charlie Gillis
Saturday, July 27, 2002
EDMONTON - Ten years ago, Darse Billings was a small-time card shark who
paid his way through university with poker earnings, honing his skills as a
gambler while pursuing a computer science degree.
Little did he know those smoke-filled nights held the key to his academic
This week, Mr. Billings and a team of University of Alberta researchers are
publicizing a poker-playing computer program that does what many a putative
gambler cannot -- it successfully processes the mercurial and misleading
information it receives in the heat of a game.
The system represents a significant stride in artificial intelligence
because it effectively guesses whether an opponent is bluffing, wavering or
playing his hands arrow-straight.
By doing so, it goes beyond programs developed for games such as chess and
backgammon, which sift through finite sets of moves before choosing a
"I had no idea that the game would play such an important role in my life,"
says Mr. Billings, a PhD candidate who leads five other researchers on the
project at the university's computer science department. "I spent a lot of
time completing my master's [degree] because I was doing a lot of what I
called 'field research.' "
He chuckles at the memory. "It was a fun time."
Now 40, Mr. Billings started the way many poker players do, joining
nickel-ante games with friends when he was an undergraduate in Calgary.
Only after he embarked on graduate studies at U of A did he identify poker
as an ideal vehicle for artificial-intelligence research, as it incorporates
both mathematical probability and what he calls the black art of guesswork.
For his master's thesis, he developed program components capable of
evaluating poker hands, and in the process accumulated vast knowledge about
the game. He soon found he could outsmart most players and began joining
local money matches to pick up cash on the side.
"That's how I sort of fell into being a professional player," he says. "If
you're really good, then you can make a living at it.... I played here in
Edmonton at casinos and legal card rooms, maybe about 20 hours a week."
Mr. Billings is coy about his winnings during that period, describing them
only as "enough to pay the bills."
In any case, his expertise proved doubly valuable as his former thesis
advisor, Jonathan Schaeffer, invited him to continue his research in the
Mr. Billings began consulting with -- and later returned to lead -- a U of A
team that would parlay his hand-reading components into a fully operational
Dubbed POKI, that program now defeats 90% of opponents in games of Texas
Hold 'Em on the Internet. (The site can be found at
www.cs.ualberta.ca/~games/ poker/). It recently defeated Mr. Billings in a
POKI's success depends heavily on its "opponent modeller," a kind of
sophisticated profiling component that catalogues a player's habits or
biases as the game progresses. The program then uses algorithmic formulas to
mix that information with baseline probabilities, creating the effect of
both reason and intuition.
To Prof. Schaeffer, it is much more than a diversion. Such technology, he
says, might be applied to countless fields of human endeavour, from military
strategy to electoral politics.
"Poker just offers an environment that allows us to experiment and quantify
and determine whether we're actually making progress."
Mr. Billings, for one, is delighted with the team's progress. The group is
currently completing a more advanced version of the program it hopes will
compete with some of the world's greatest poker players, he says.
Someday -- possibly within the next few months -- the team hopes to line up
a trial against a prominent player, such as those who compete annually in
the World Series of Poker.
"That's probably how it will prove itself," he says. "If our techniques for
playing the game are sound and we can find a strong player ... then one of
the ways to demonstrate it would be to play against a human, one-on-one, and
show that it can hold its own against anyone."
NHNE News List:
To subscribe, send a message to:
To unsubscribe, send a message to:
To review current posts:
Published by NewHeavenNewEarth (NHNE)
NHNE Website: http://www.nhne.com/
Phone: (928) 282-6120
Fax: (815) 346-1492
Appreciate what we are doing?
You can say so with a tax-deductible donation:
P.O. Box 2242
Sedona, AZ 86339