An artificial intelligence program called Pluribus has successfully beaten five professional human players, after a lengthy 5 day poker session. Previous versions of the technology have only ever managed to beat a single opponent in a one-on-one scenario. Pluribus is the first to win in a multi-player setup.
The humans players were all elite professionals, who had previously won at least $1 million. Pluribus played over 10,000 hands of no-limit Texas hold’em poker (the most popular version of the game) and won a virtual $48,000.
It’s relatively easy to train an AI to play a game like chess, where all the pieces are visible and all the information is known. What makes poker so challenging is the amount of hidden information, bluffing, and complex human behaviour involved.
Thea achievement marks the “first time AI has achieved superhuman performance in a multiplayer game,” according to Tuomas Sandholm, who developed Pluribus with his PhD student Noam Brown at Carnegie Mellon University in Pittsburgh. He hopes that the sophisticated algorithms needed to beat five players in such a complex game can have other real-world applications.
How the algorithm was trained
Like many AI algorithms, Pluribus learned through a rapid process trial and error. Playing against a copy of itself to begin with, it started out playing randomly and made slight changes to its tactics after every game. After eight days of training, it was ready to test against human opposition.
Interestingly, Pluribus has found success using some bizarre strategies that regular players usually avoid. For example, it would wildly vary the size of each bet. It also frequently used ‘donk betting’ – the practice of ending the first round of betting with a call and opening the next with a bet. Professional players usually advise against doing this, but Pluribus managed to pull it off successfully.
Sean Ruane, one of the pros who took on Pluribus, found it to be a tough opponent. “In a game that will, more often than not, reward you when you exhibit mental discipline, focus, and consistency, and certainly punish you when you lack any of the three, competing for hours on end against an AI bot that obviously doesn’t have to worry about these shortcomings is a gruelling task.”
The developers who created Pluribus hope that the technology can be transferred to other areas of research – from investment banking and negotiation strategies to helping political candidates decide how much they should spend on advertising. Tuomas Sandholm and his laboratory team are already working on medical applications with partial funding from US Army Research Office.
You can find more information about the project on the Facebook artificial intelligence blog.