Poker-Playing Robots Battle For $100,000 PotPoker-Playing Robots Battle For $100,000 Pot
It's not just fun and games and cash. The winner of the World Poker Robot Championship will help artificial intelligence meet a tough challenge: solving problems in the face of incomplete and inaccurate information.
July 12, 2005
When the first hand was dealt Tuesday at the inaugural World Poker Robot Championship 2005, the stakes were higher than merely the $100,000 pot that goes to the winning computer programmer. The tournament, which had originally been called the World Series of Poker Robots, could shed light on one of the most challenging aspects of artificial intelligence: getting computers to solve problems when they're given incomplete, even misleading, information.
A poker program, or "bot," can play the game perfectly, making its calls based upon the odds. "But playing the perfect mathematical poker game is only part of it," says Kurt Lange, president of Poker Academy, a maker of poker gaming software. The tournament--held at Binion's Horseshoe Hotel & Casino, the site of the final two rounds of the real World Series of Poker--is significant in that it draws out different ideas and techniques for creating artificial intelligence. IBM made a name for its Deep Blue supercomputer by occasionally beating then-world champion chess player Garry Kasparov. Unlike chess, which Lange calls, "the perfect information game," poker is a "misinformation" game. "It's a whole new challenge and way of thinking for the computer," he says. "You have to find your way through the woods using bad information." Poker also introduces luck as a key factor that levels the playing field. "The poker problem is turning out to be very vexing," says Jonathan Schaeffer, a computer-science professor at Canada's University of Alberta and a researcher at the school's Computer Poker Research Group. "Chess was very easy for IBM's Deep Blue in the sense that you discovered there was an alpha-beta algorithm for playing chess. If you threw enough horsepower behind it, you could beat a human. The problem with poker, or any domain where you're working with unknown information, is that there's no one way to do it." Both Schaeffer and Lange see this week's tournament as a way to expose their artificial-intelligence work. While many people can relate to poker, the underlying technology has broader applications. "A poker bot is trying to learn its opponent," Lange says. "It will record how its opponents play their hands, creating a database of an opponent's moves." Like a seasoned poker pro, who essentially keeps his or her own database of opponents' moves and tendencies stored behind dark sunglasses, the computer uses this historic information to predict future outcomes. The success of poker programs such as those created by Poker Academy, the Computer Poker Research Group, and the six international entrants in this week's computer-poker tournament could extend into areas with great business, political, and even military implications, Schaeffer says. Leaders are frequently challenged to make decisions based on incomplete information. "The highest-stakes game of poker that I ever followed started with the first Gulf War," Schaeffer says. "Saddam Hussein anted by invading Kuwait." President George H.W. Bush raised by moving in troops. Hussein would later counter by introducing the possibility of Iraq having weapons of mass destruction, and it was up to Bush and subsequent presidents to decide whether he was bluffing. "That's the real world," says Schaeffer, who doesn't play poker. Poker Academy is providing the software and server technology for the World Series of Poker Robots tournament. In essence, Poker Academy is the tournament director, dealing digital cards, determining the winner of each hand, raising the blinds, and performing various other duties. Think of it as a virtual Matt Savage, director of the actual World Series of Poker tournament. Poker Academy is a division of biotech software developer BioTools Inc., a company University of Alberta researchers formed in 1995 to develop DNA- and protein-sequence analysis software. BioTools' computer scientists eventually found a more profitable outlet for their skills in online poker. The company earlier this month released Poker Academy Pro Texas Hold'em 2.0, a program for training people to play the popular poker game. The Computer Poker Research Group developed the format for the World Series of Poker Robots. Although organizers of the World Series of Poker Robots have kept details of the game's format close to the vest, six computer programmers from around the world, including China, Spain, and the United States, will pit their software against each other in a series of elimination rounds. Based upon the format agreed to as of earlier in the week, the tournament will consist of four-hour sessions, with a brief halftime during which programmers can tweak their software, Schaeffer says. The tournament winner will be decided by Thursday and is then scheduled to face off against the Computer Poker Research Group's Poki computer-software program. The winner of that duel is on Friday expected to then take on poker pro Phil Laak, known as the "Unabomber" because of his trademark hooded sweatshirt. One concern Schaeffer has about the "Robot" tournament's ability to advance artificial-intelligence research is whether the participating software programs will play enough poker hands to reduce the luck factor. Schaeffer projects that the programs will play a few hundred hands for each four-hour session, which will include commentary and analysis for the benefit of spectators. "When we do our experiments between two computer programs, we run them through 40,000 or 50,000 hands," he says. Schaeffer hopes that the University of Alberta's artificial-intelligence research will someday catch on with a larger audience. He says he's pitched his technology to several "major IT vendors," but that they see the associations with his poker model as too much of a gamble.
About the Author(s)
You May Also Like