Poker-playing AI 'bot' carries long-range impact
For Sam Ganzfried, No Limit Hold 'Em means more than big pots, timely bluffs and the thrill of an all-in bet.
He also sees the game as a path to using science to solve an array of complex issues, such as figuring out the best blend of treatments for an HIV patient or identifying the most effective methods for protecting airports.
Ganzfried, who just moved from Pittsburgh to Miami, where he is beginning a stint as assistant professor of computer science at Florida International University, approaches poker as both a player and scientist. He finished in the money at a 2010 World Series of Poker event and cashed last year in the main events of the Pittsburgh Poker Open and the Three Rivers Poker Challenge, both at Rivers Casino.
In addition, he was one of three Carnegie Mellon University researchers who developed Claudico, which used artificial intelligence to play 80,000 hands of No Limit Hold 'Em against four of the world's best heads-up players in 2015.
Hold 'Em is a good research tool because it requires players, whether human or AI, to make decisions based on incomplete information. Each player has two cards that others at the table don't see. Processes that help AI pick the best strategy in Hold 'Em can be adapted to solve complex problems fraught with what scientists call “imperfect information.”
“The computer poker research area is just a little over a decade old,” says Ganzfried, who has a doctorate in computer science from CMU and a bachelor's degree in math from Harvard. “The ideas are starting to have applications in medicine, security. It's exciting to see these applications develop.”
The trick is to view an issue, say medical treatment, as a zero-sum game, like heads-up poker: If the patient wins, the disease loses, or vice versa. AI's goal is to devise a strategy that will do as well as possible in the long run.
Gaznfried says a Heads-up No Limit Hold 'Em player can face an almost incomprehensible number of situations: 10 to the 161st power. According to universetoday.com, that's far greater than the number of atoms in the universe.
“There are a lot of (nonpoker) situations where multiple agents have private info only they know,” Ganzfried says. “(AI) agents have to act strategically. It's only matter of time before agents make an impact in other areas.” For example, Ganzfried says a process developed at Alberta University, which used AI in No Limit Hold 'Em, helped compute individualized diabetes treatments.
On a pure poker note, Ganzfried worries about the potential effect of computer “bots” in online games. Pennsylvania may soon join Nevada, New Jersey and Delaware in regulating online poker. Millions of Americans still play online poker through offshore websites, and complaints about bots giving their creators an unfair advantage are common. Bots are illegal on authorized U.S. poker sites, according to pokerupdate.com.
“Even though I've worked on the computer program for poker, I also play poker,” Ganzfried says. “I certainly don't want these bots to have any adverse impact on a game. Certainly none of the academic researchers are making their bot code available or using it online.”
Still, some people have developed computer programs to play online poker better than many human opponents. Ganzfried says that puts the host websites in a tough spot. If bots play lots of hands, the sites make more money from the rake taken on each hand. But if human players leave because of bots, the sites lose money.
One approach might be to have sites open only to bots or to identify which players are bots and which are human. “That could lead to some interesting data to be used for research,” Ganzfried says.
Whatever happens with online poker, he aims to keep studying the game as a player and as a scientist.
“The goal for me is not to just to build a poker program,” Ganzfried says. “The goal is to study fundamental scientific questions that have broader applicability.”
Mark Gruetze is the Tribune-Review's gambling columnist. Reach him at PlayersAdv@outlook.com