- Terminator Salvation
A few days ago, it looked like AI was about to conquer No Limit Head’s Up Texas Hold’em poker. Suddenly, the human poker pros have started making a comeback.
Poker bot Libratus, a bot made at Carnegie Mellon University, is facing four human pros in a 20-day, 120,000 hand competition in Pittsburgh. After day one, Libratus was up $82,000. “AI is crushing humanity at poker,” declared The Verge. After day two, the bot’s lead was up to $150,000; after day three, it was at $193,000.
But then the humans started winning. After day four, Libratus’ lead fell to $151,000. After day six it plunged to $51,000. (You can see the latest here).
“We got off to a bad start, which I believe is kind of expected,” one of the players, Jason Les, said by email. “Those first hands we play without any idea how our opponent plays and it took us a while to study and get an understanding of what was going on.”
Les also represented humanity in a 2015 competition against CMU’s Claudico bot, which the humans narrowly won. He says that the latest bot is a major step forward.
“I believe Libratus would destroy Claudico by a good margin,” he wrote.
Yet the humans appear to be finding weaknesses to exploit. While Les wouldn’t elaborate on what they were, he said that the humans were getting together for a few hours every night to develop strategies.
Creativity and adaptivity are, for now, areas where humans outperform robots.
Libratus, created at CMU by Tuomas Sandholm and PhD student Noam Brown, aims to be the first bot that can beat top human players at Head’s Up No-Limit Hold’em. Head’s up means a game with two players: it’s easier to solve than group poker games. No-limit means there’s no limit to bets: it’s harder to solve than limit hold’em, which a bot from the University of Alberta conquered in 2015.
Why should we care about the rise of poker AI? Aside from how it is making humans much better at poker, it could help solve real-world problems that, like poker, involve incomplete information. For instance: negotiations.
“In the real world, all of the relevant information is not typically laid out neatly like pieces on a chessboard,” Brown writes in an email. “There will be important information that is missing or hidden, and the AI needs to be able to handle that.”