What Is Pluribus
Pluribus is an AI poker bot developed by Facebook AI Research and Carnegie Mellon University that defeated professional human players in six-player no-limit Texas Hold'em. It was announced in July 2019 after a landmark experiment where it played against 12 professional poker players and won with statistical significance. The system represents a breakthrough in multi-agent artificial intelligence because poker involves hidden information, bluffing, and multiple opponents, making it far more complex than perfect-information games like chess or Go. More details about the technical architecture are available on the official Facebook AI Research blog Pluribus: A Superhuman AI Poker Bot.
The development team included researchers Tuomas Sandholm and Noam Brown, who previously created Libratus, an AI that defeated top humans in heads-up no-limit Texas Hold'em. Pluribus extended that work to handle six or more players simultaneously, a scenario that introduces new strategic complexity because players must reason about alliances, collusion, and shifting dynamics. The AI ran on a single server with 128 GB of RAM and no specialized hardware, demonstrating that superhuman poker performance does not require massive supercomputing clusters. The experiment was conducted at a casino in Pittsburgh and involved 10,000 hands per human participant.
How Pluribus Works
Pluribus uses a technique called blueprint strategy computation combined with real-time search during live play. It first computes a simplified abstracted version of poker using a technique called counterfactual regret minimization, which helps the AI learn which actions yield the best long-term results. During actual games, Pluribus performs real-time search to refine its decisions based on the specific cards and betting actions it observes. This hybrid approach allows it to balance computational efficiency with strategic depth, making it practical to run on commodity hardware.
Key Technical Innovations
One key innovation is the use of a new algorithm for computing approximate Nash equilibria in large imperfect-information games, which allows Pluribus to find strategies that are difficult to exploit. Another innovation is its ability to handle multiple players simultaneously without relying on assumptions about how opponents will coordinate or behave. The system also uses a technique called subgame solving, where it focuses computational resources on the most relevant parts of the decision tree during a hand. These methods together allow Pluribus to make decisions in milliseconds that would be difficult for even elite human players to counter consistently.
Impact on AI and Poker
Pluribus marked the first time an AI achieved superhuman performance in a multiplayer poker format, a milestone that had remained elusive for years because of the added complexity of multiple opponents and hidden information. The result was published in the journal Science in July 2019, with the paper titled "Superhuman AI for Multiplayer Poker." The achievement demonstrated that AI systems can handle environments where deception, incomplete information, and strategic interaction are central, which has implications beyond poker into areas like cybersecurity, business negotiations, and autonomous driving.
The poker community responded with a mix of admiration and concern, as professional players noted that Pluribus played a style that was difficult to read and exploited common human tendencies such as overfolding and calling too much. Some players described the AI's betting patterns as alien and highly effective, particularly in situations involving bluffing and pot control. The experiment also sparked broader discussions about the future of AI in competitive environments and the potential for similar systems to be applied in real-world strategic scenarios. The research was funded in part by the National Science Foundation and the U.S. Army Research Office, with additional support from Facebook AI Research.