[toc]The first round went to the humans, but this time around artificial intelligence (AI) prevailed in the poker battle between heads-up no limit hold’em specialists and the new iteration of Carnegie Mellon University’s poker AI, Libratus.
The bot beat four top poker pros to the tune of $1,766,250 over 120,000 hands. The feat makes it the most successful tested poker artificial intelligence.
Libratus mastered the art of bluffing
The previous artificial intelligence from the Pittsburgh university lost to the pros in 2015. Though the people won, the margin of victory was not considered statistically significant.
The main criticism the pros had for the bot called Claudico was an inability to bluff credibly. This has historically been the biggest issue for no-limit AI. While limit hold’em bots can battle with human players, bet sizing was difficult to conquer.
“The computer can’t win at poker if it can’t bluff,” Frank Pfenning, head of the CMU computer science department said in a university press release. “Developing an AI that can do that successfully is a tremendous step forward scientifically and has numerous applications. Imagine that your smartphone will someday be able to negotiate the best price on a new car for you. That’s just the beginning.”
The other big difference in Libratus and other AI is how the algorithms are utilized. Rather than trying to use them to exploit weaknesses in opponents, Libratus had an algorithm designed to identify and repair the biggest error the bot detected in its own game at the end of each day of play.
Rivers Casino hosted 20-day AI experiment
The quartet was comprised of online poker players Jason Les, Dong Kim, Jimmy Chou, and Daniel McAulay. The foursome lost, but they will split a $200,000 purse for their efforts.
According to the pros, the experiment also proved to be a valuable learning experience.
Usually, you have to lose a lot and pay a lot of money for the experience. Here, at least I’m not losing any money,” Les explained.
Carnegie Mellon has helped develop artificial intelligence for chess and the game show Jeopardy! over the past few years. The big difference between these games and poker is that poker is a game of imperfect information. Being able to overcome that obstacle is a huge breakthrough for the program.
University of Alberta claims to have a poker bot too
Carnegie Mellon University had to share the artificial intelligence spotlight earlier this month when researchers at University of Alberta released a paper about their own no-limit hold’em AI, DeepStack.
The paper has not been peer reviewed. It claims the bot is beating real poker players in 45,000-hand matches.
The game of no-limit hold’em will probably never be completely solved. These advances in artificial intelligence are a huge step in the advancement of the field though.