Computer fails to beat humans in Carnegie Mellon poker faceoff
Carnegie Mellon鈥檚 poker-playing AI fell short of victory in an 80,000-hand faceoff with human pros.
A poker dealer is shown during a game at a Miami casino in 2011.
AP Photo/Wilfredo Lee
Claudico has the best poker face in the game. That鈥檚 because it doesn鈥檛 have a face at all.
Developed by聽a team of Carnegie Mellon computer scientists led by Prof.聽Tuomas Sandholm,聽Claudico is a poker-playing artificial intelligence that specializes in no-limit Texas Hold鈥檈m, a two-player variant with unlimited betting. And while the program is far from unbeatable, it is a worthy competitor.
At a grueling two-week competition at Rivers Casino in Pittsburgh, Claudico proved its mettle against four top poker players. Human professionals Bjorn Li, Doug Polk, Dong Kim, and Jason Les each faced off with the program, playing 80,000 hands between them. No real money was wagered, although a $100,000 donation from Rivers Casino and Microsoft was split between the players based on their performance.
At the end of the competition, Claudico came up short by a combined margin of $732,713. But since $170 million was "wagered" overall, the loss was considered statistically insignificant. So as far as researchers are concerned, the bout ended in an encouraging tie.
"We knew Claudico was the strongest computer poker program in the world, but we had no idea before this competition how it would fare against four Top 10 poker players," Dr. Sandholm said in a . "It would have been no shame for Claudico to lose to a set of such talented pros, so even pulling off a statistical tie with them is a tremendous achievement."
As an incomplete information game, poker is the perfect testbed for artificial intelligence. By developing programs capable of decision-making with limited information, researchers can then apply them to broader human problems. At the University of Alberta, computer scientists have already 鈥渟olved鈥� a simpler variation of poker with a program called Cepheus.
"The real world is a whole lot like a poker game," said Michael Bowling, lead developer of Cepheus, in a previous Monitor story. "One of the things we have to cope with in making any real world decision is uncertainty. Humans can deal with uncertainty. We鈥檙e still able to make reasonable decisions. Poker embodies that uncertainty. If we鈥檙e going to build artificial intelligence systems that can answer complex real-world problems, they need to deal with uncertainty. So looking at this through the lens of gaming is actually easier."
Cepheus plays Limit Hold'em, a game in which carefully-structured betting results in a finite number of possible in-game choices. But in No-limit Texas Hold鈥檈m, Claudico鈥檚 game of choice, there is no betting cap. Players can bluff with huge bets, thus skewing perceived odds. The result is an incredibly unpredictable game that will prove much more difficult to "solve."
To that end, Sandholm and colleagues are already improving Claudico鈥檚 algorithms. With 80,000 hands of data to work with, they hope their program will soon surpass human players. But that will only be another step in the quest for true artificial intelligence.
"Beating humans isn't really our goal; it's just a milestone along the way," Sandholm said. "What we want to do is create an artificial intelligence that can help humans negotiate or make decisions in situations where they can't know all of the facts."