Facebook built an AI poker-bot capable of beating some of the world's best poker players

REUTERS/Suzanne Plunkett

• Facebook built a poker bot capable of beating human pros.
• It's a major milestone for artificial intelligence (AI) research, which has been working toward this for decades.
• The breakthrough will have implications for AI in fields with limited access to knowledge, from self-driving cars to negotiations.

Facebook has built a poker-playing artificial intelligence (AI) bot capable of beating some of the world's best players of the card game, in a significant step forward for AI research.

Working with Carnegie Mellon, researchers at the California tech giant developed a piece of software called Pluribus that was able to handily defeat an array of A-list poker stars in games of six-player no-limit Texas Hold'em poker, the company announced on Thursday.

Complimentary Tech Event

Transform talent with learning that works

Capability development is critical for businesses who want to push the envelope of innovation.Discover how business leaders are strategizing around building talent capabilities and empowering employee transformation.Know More

Achieving this milestone has long been a target for AI researchers, and it has now been achieved - with major implications for both the field of AI and the game of poker itself. It now joins the ranks of games like Go and Chess before it where the world's best human players have been superseded by computer agents.

Poker offers some unique challenges not seem in more "simple" games: There are multiple players, and each player only has limited information, allowing for bluffing and other advanced strategies (unlike chess where a bot could, with enough computing power, theoretically calculate every possible outcome of every move their opponent makes with absolute certainty). AI agents have previously been able to defeat human opponents in two-player poker, but not six-player, which adds an additional layer of complexity.

"Poker has served as a challenge problem for the fields of artificial intelligence (AI) and game theory for decades," wrote researchers Noam Brown and Tuomas Sandholm in a paper to be published in Science. "No other popular recreational game captures the challenges of hidden information as effectively and as elegantly as poker."

Pluribus played in two formats: Against five other humans, and against one human and four other versions of itself (the bots could not communicate and didn't know who they played against, preventing any collusion). All the human opponents had won at least $1 million in professional poker over their careers, including major tournament winners Chris Ferguson, Greg Merson, and Darren Elias.

Of course, this isn't just about poker. It's a major achievement - but it is significant because it has implications far beyond the card-playing space. It's a demonstration that AI can operate at "superhuman" levels in situations with multiple actors and limited access to information, and could be applied anywhere from self-driving car technology to negotiations, Brown said in an interview.

Got a tip? Contact this reporter via encrypted messaging app Signal at +1 (650) 636-6268 using a non-work phone, email at rprice@businessinsider.com, Telegram or WeChat at robaeprice, or Twitter DM at @robaeprice. (PR pitches by email only, please.) You can also contact Business Insider securely via SecureDrop.

Read more:

• Mark Zuckerberg's personal security chief accused of sexual harassment and making racist remarks about Priscilla Chan by 2 former staffers
• Facebook says it 'unintentionally uploaded' 1.5 million people's email contacts without their consent
• Years of Mark Zuckerberg's old Facebook posts have vanished. The company says it 'mistakenly deleted' them.
• Car-bomb fears and stolen prototypes: Inside Facebook's efforts to protect its 80,000 workers around the globe