Using a computer to beat Poker

• By Michael Stetz
• UNION-TRIBUNE STAFF WRITER

Using a computer to beat chess masters is so 1997. Today - programmers are gunning for the world’s best Texas Hold ’em players. Forget Deep Blue’s mastery of bishops, queens and knights. Think pocket aces, open-ended straight draws, the flop – all calculated by increasingly popular poker bots. Last month - for the first time, a computer program beat top online poker players in a contest called *Man vs. Machine.* About that time, a Texas Hold ’em program created by a Hillcrest resident finished in a three-way tie for second place in an annual contest that pits artificial intelligence poker programs against each other. The program is called Fell Omen 2 and its play has no weaknesses, said creator Ian Fellows, a researcher and statistician with the UC San Diego Department of Psychiatry, who developed the program in his spare time. Far from the Texas Hold ’em video games you find in stores, these artificial intelligence creations can figure out the many angles and odds of the game and make bets and decisions accordingly. They even bluff. It’s a challenge to create such programs because, unlike chess, in which the action is out in the open, cards are hidden in poker. The computer can’t get all the information. Fellows has made Fell Omen 2 available for free on his Web site. With a few clicks, average players can sharpen their skills by going head to head with a virtual expert versus, say, their buddies in the garage. For years - many thought humans were too nimble and creative to be beaten by a machine. Then in a landmark achievement, IBM’s Deep Blue beat world champion Garry Kasparov in chess in 1997. Now your cell phone can probably beat Kasparov! *Hyerborean07,* the program that topped Fellows’ work in last month’s competition, was created by the University of Alberta Computer Poker Research Group, which has been working on Hold ’em programs for more than a decade and making steady improvements. The group also created *Polaris II,* which humbled top players with its aggressive and unpredictable play in the Man vs. Machine contest. Poker players aren’t ready to fold yet, though. Jerry Yang was a relative newcomer to the game when he took home $8.25 million for his stunning win in the 2007 World Series of Poker. While the Fresno resident said he’s not too familiar with the new technology, he doubts that a computer can beat professional players in tournaments. The game has too many variables, he said, and it takes human senses to play well. Yang studies other players’ moves at the table, for instance, looking for clues as to how they play. *Body language is so important. I look to see how players talk, how they breathe,* Yang said. I don’t think a computer can do that. Programmers argue that the game comes down to numbers, not facial tics. And they believe it’s only a matter of time before the programs become more sophisticated. Still - this is Hold ’em, not chess. Computers taking on chess masters made sense. Chess players, while eccentric – Bobby Fischer, anyone? – seem to calculate and forecast moves with almost technological precision. Hold ’em, on the other hand, consists of zany, unpredictable characters. It’s got a Wild West mentality. Players wear sunglasses and hoodies. They have nicknames such as Jesus and The Dragon. More than a few talk trash. *The best players in the world say poker is a game of people played with cards and chips, and if so, then computers should lose more often because they can’t read people,* poker expert Steve Rosenbloom said via e-mail. His syndicated column on the game appears in The San Diego Union-Tribune as well as other publications. *Poker is a game of making fewer mistakes,* he said. Computers are better at that than humans. But I can’t see computers working at a nine-handed table. Right now - the computer programs do have limits. The most competitive programs play only heads-up, meaning one against one. And they don’t play the all-in tournament style, where players can bet their entire stack of chips. The betting is limited. The all-in programs aren’t that strong yet, experts say. Fellows, 27, dabbles in Hold ’em, but he says he’s not that good. Being a math guy, he was intrigued by the idea of creating a program that could do what he couldn’t, which was to win consistently. Now, he said he won’t go against his creation: *It beats me and I get discouraged.* Fell Omen 2 is his second effort, following a program called INOT, which took second place in last year’s Computer Poker Competition, sponsored by the Association for the Advancement of Artificial Intelligence. Fellows’ concept goes something like this: He introduced the computer program to Hold ’em at its most simple level and then instructed it to exploit itself. Fellows likens it to the way a person would learn to play the game. Even bluffing is a learned tactic, he said. So ultimately, Fell Omen 2 became a master, one that will grind its way to a win. This reporter was able to win several hands, but, over the long run – 10,000 hands, perhaps – the program would eventually dominate, Fellows said. Other programs do even more. Polaris II, for instance, can figure out patterns in its opponents’ play and make adjustments, said Michael Bowling of the Alberta Computer Poker Research Group. If you’re prone to only playing strong hands, Polaris will pick up on that and exploit it, without displaying any discernible pattern itself. This work isn’t being done just to beat poker players. The advances could be applied to other fields in which artificial intelligence makes calculations based on limited information, such as economics. But don’t be shocked if you see a computer wearing sunglasses, of course – at the 2020 World Series of Poker. By Michael Stetz: (619) 293-1720: [email protected]