Poker news from http://money.canoe.ca/
If the crew of 2001: A Space Odyssey needed to defeat evil supercomputer HAL, they should have challenged it to a game of poker.
Unlike IBM's Deep Blue, a computer that was able to beat world-champion chess player Garry Kasparov in 1997, even the world's best poker-playing computers would flop against the top human players.
That's because computer scientists have not yet figured out how to write programs that can make informed decisions based on incomplete or inaccurate information, said Jonathan Schaeffer, chair of the University of Alberta's computer science department and Canada Research Chair in artificial intelligence.
"The skills that make human poker players really good are skills that don't seem to match well with what computers can do," said Schaeffer. "Computers aren't particularly good at learning, for example, or reasoning by analogy."
Schaeffer was part of the team that designed Hyperborean, a poker-playing computer that recently went undefeated at two tournaments hosted by the American Association of Artificial Intelligence.
In the first tournament, four computers played 40,000 hands of limit Texas hold 'em against each of their competitors. Each program was given seven seconds to make its next move and the computer that won the most money won the tournament.
In the second competition, the computers played 12,000 hands of poker against each of their opponents but were given a total of 60 seconds to make their decisions to encourage a higher level of play.
To ensure that no one computer got lucky, each side was given the opportunity to play its opponent's hand after each deal.
Although more than 250,000 hands of poker were played between the two tournaments, Hyperborean wasn't designed simply to amuse poker lovers, said Michael Bowling, head of the research group that created the computer.
"Poker has what are currently some of the biggest challenges to (artificial intelligence) systems, and uncertainty is the primary hurdle that we're facing," said Bowling, adding that the University of Alberta program was able to use its opponents' actions to infer certain things about their hands.
"The same techniques, the same principles that we're developing to build poker systems are the same principles that can be applied to many other problems."
The University of Alberta was one of the first institutions in the world to attempt to develop a poker-playing computer, said Bowling. The research was initiated in 1997 by a graduate student named Darse Billings.
"He was actually an ex-professional poker player, and also being somewhat of a math geek, he said, 'Can we actually build programs that can do as well as humans can at this very psychological game?' Then it sort of steamrolled because it's an exciting and fun topic to work on," said Bowling.
The programs used to create Hyperborean have been licensed to a University of Alberta spinoff company called BioTools, he added, and have been turned into an online practice tool called Poker Academy.
"We're not at the point where we can beat the world's elite professional players, but we can certainly give them a reasonable competition so that they're not playing with something that makes horrible mistakes," he said.
But researchers are still a long way from creating a Deep Blue-quality poker program, said Schaeffer.
"The nice thing about chess as a property of the game is what we call perfect information. You look at the board, you know where all the pieces are, you know whose turn it is - you have complete knowledge of the game," he said.
"But in the real world, knowing everything is just so rare. Everything we do all day long is all about partial information. So poker's much more representative of what the real world's like, and in that sense it becomes a much harder problem.
"The end result is that we're going to learn more in terms of research outcomes from poker than we ever did from chess."
|
|
