For game theorists, poker is a great frontier because of the variables created by the human mind. Texas Hold ‘em is much more difficult to map out than, say, deciding which checker to move first.
In poker, an opponent’s $10 bet could imply that he has a pair of aces. Or maybe he has two rags and is trying to bluff you.
How to put that into a computer program? And why does it matter?
That’s a piece of the quest Florida International University professor Sam Ganzfried is on. Ganzfried, who received a doctorate in computer science from Carnegie Mellon after studying math at Harvard, has already cleared one hurdle: creating a poker program that can defeat players in a head-to-head match.
Never miss a local story.
Now he’s weighing whether to delve more deeply into poker, or to take a fork in the road and use his powers for greater good. That’s because the strategies used in head-to-head poker can be similar to binary choices made in disease treatment, virus dynamics and mutations, he notes.
But let’s go back to poker for a minute. Ganzfried helped create a two-player, limit Texas hold ‘em agent named Claudico (Latin for “I limp”), that competed in the inaugural 2015 Brain vs. Artificial Intelligence Competition against four of the world’s top human online players (not the guys you see on TV). The players prevailed by a whisker, then lost a rematch early this year against Claudico’s successor, named Libratus. But there’s still a long way to go. …
“Limit poker has variables of about 10 to the 17th power,” said Ganzfried, who is on staff at the College of Engineering and Computing. “No-limit increases that to about 10 to the 65th power.
“In these games, there’s what’s called a Nash equilibrium, an optimal solution,” he said, and yes, the “Nash” refers to John Forbes Nash Jr., the subject of the movie “A Beautiful Mind.”
Ganzfried himself enjoys high-stakes poker in local casinos. He played in a November 2015 filming of “Poker Night in America” in Pittsburgh, with poker stars Jen Harman and Phil Laak and his girlfriend, actress Jennifer Tilly, at the table. Minimum buy-in was $5,000. Ganzfried won a hand with a pair of aces, and busted Laak later in the show.
“Ganzfried is absolutely crushing it,” one commentator said.
“Are you a bot?” Harman asked. “Can you peel your skin off like in ‘Terminator?’ ”
But Ganzfried (Twitter handle: Sam_B0t) says, sorry, there are no tried-and-true AI tips transferable to help players with their games.
“But one thing worth noting was there were a lot of unconventional bet sizes,” he said. Players trying to win a pot will often bet two-thirds of the amount already wagered. The program sometimes bet four times the pot, or sometimes only 10 percent.
But now with more poker frontiers, though, Ganzfried, is waffling.
“The idea of applying this to something outside of poker is interesting,” he said. Part of his decision will be based on which students latch on to the project.
“Ideally you can do both. It’s going to depend on the students I can get,” he said. “If I can get a lot of funding and multiple strong students I can get one working on poker and another applying it to medicine.”
He said Carnegie Mellon’s rival school in artificial intelligence, the University of Alberta, published a paper looking at a problem in managing diabetes, which involves patient cooperation.
“It’s a robust problem with parameters, and only one agent,” he said. “But if you view the parameters as being chosen adversarially, then the game becomes a model as a two-player zero sum game.”
Those problems are where artificial intelligence and game theory can bring about progress in the world, including cybersecurity, said Ganzfried, who has spoken in the Netherlands and at Stanford University.
“I didn’t go into my Ph.D. program planning to work on poker,” he said.
Get daily gambling news at SouthFloridaGambling.com