The gauntlet’s been thrown. And let the chips fall where they may.
In the same man-versus-machine vein as the IBM computer Watson’s turn on the TV game show Jeopardy!, a new card-shark computer program developed at the University of Alberta is being challenged by the head of this country’s premier poker league.
“You can out-think an algorithm, I think you can outplay an algorithm,” says a defiant Kelly Kellner, president and CEO of the Canadian Poker Tour.
“I think we can pop a few circuits on that damn thing.”
That “damn thing” is a program known as Cepheus, which was created by researchers at the Edmonton institution, who say it has essentially solved a version of the popular card game known as Texas hold’em.
The program — whose creators say it has run more poker games than have been played in all of human history during its development — was described this week in the prestigious journal Science.
And its head creator has an answer for Kellner’s challenge: bring it on.
“We’d love to even publicize (a match) and have some spectators come along and watch,” says computer scientist Michael Bowling, leader of the U of A’s Computer Poker Research Group and lead author of the Science paper, released on Jan. 7.
“I have no doubt that our program is far superior to the best players.”
The program — which will beonline this week for anyone to play — was developed using 4,000 powerful computer-processing units (CPUs), each capable of playing six billion poker hands a second.
Employing some “cool mathematical tricks,” the Cepheus program honed its skills over two months by playing countless trillions of games against itself.
“So maybe it’s not surprising that our program can reach superhuman levels and actually be nearly perfect,” says Bowling.
But Kellner, who is also publisher of the Canadian Poker Player Magazine, says many of his top guns are willing and eager to give Cepheus a go.
And given the massive audiences Watson drew when playing Jeopardy! champions Ken Jennings and Brad Rutter four years ago — and the global interest in chess champion Gary Kasparov’s two tilts against IBM’s Deep Blue in 1996 and ’97 — Kellner says a poker challenge would be an easy TV or Internet sell.
The particular game Cepheus has essentially mastered is known as heads-up limit hold’em poker.
This simplest version of the Texas-style game pits two players against each other. Like all such hold’em games, the players are both dealt two down cards in the first round, three shared cards with faces showing in the second and one more shared up card in the third and the fourth rounds.
After each dealing round the players can check, bet or fold.
Like all hold’em games, this version rewards the best five-card poker hand that can be cobbled from the down and shown cards. Unlike most, however, it fixes the amount of each individual bet and limits the number of times that amount can be raised to four after each dealing round.
Computers have already mastered several simple board games, including checkers, which was solved by a computer program also developed at the University of Alberta by a separate team in 2007.
But unlike such parlour games, in which all moves can be observed and calculated, poker holds a large element of pure chance, Bowling says.
And no matter how adept a player the Cepheus program is, it can still fall victim to a hand of crappy cards.
“If we’re dealt really terrible cards in every hand, then we’re not going to be able to win those hands,” Bowling says.
“And so there is some probability that even if we were to play a long match that we would just continually get dealt really terrible cards.”
Just like a string of coin flips, however, dealt luck will even out for both sides in a poker game, Bowling says, because the probability of unequal bad luck gets smaller and smaller with every dealt hand.
That leaves his program’s vast expertise — it’s seen every possible combination of plays and bluffs — with an unbeatable advantage in the long run.
“What we’re actually saying is that if were to play a long enough sequence of matches so that the luck evens out, that we’re guaranteed that we can’t be losing,” he says.
“No matter what you do, no matter how strong a player you are, even if you look at our strategy in every detail . . . there is no way you are going to be able of have any realistic edge on us.
“On average we are playing perfectly. And that’s kind of the average that really matters.”
A professional player going up against a rank amateur would need to run through 30,000 hands before he or she could statistically prove they were the better player, Bowling says.
Going against professional players, Cepheus would be ahead 95 per cent of the time or more after 30,000 to 40,000 hands.
The value of the program, however, lies not in such theoretical marathon matches but in the artificial intelligence applications it may bring to real-world situations in which there are unknown factors.
Protecting vulnerable targets from terrorist attacks — where such things as methods and timings are hidden — is one potential application, Bowling says.
“The adversary could investigate our strategy and exactly how we’re running our patrols, (but) there’s still no hole that they can exploit,” he says.
“And that looks pretty close to the poker problem that we’ve solved, that we want to build a strategy that no matter how the other person plays, we’re unbeatable.”
Source:: Metro News
