pluribus

Pluribus is a Latin term meaning “many,” most widely known from the motto E pluribus unum (Out of many, one), which appears on the Great Seal of the United States. Beyond its linguistic use, the name has been adopted in technology to refer to a pioneering artificial intelligence system designed for multi-player poker.

In 2019, Pluribus was developed by researchers from Carnegie Mellon University and Facebook AI Research (FAIR). It plays five-player no-limit Texas Hold'em and represents a significant advance in solving imperfect-information, multi-agent games. The system combines deep neural networks with a game-search framework to propose actions and evaluate positions, and it relies on self-play to learn strategies that approximate equilibria in environments with multiple opponents.

In extensive experiments, Pluribus defeated top human professionals in five-player poker, demonstrating a level of play comparable to expert human performance in a complex multi-agent setting. The results were published in Science and highlighted the system’s ability to cope with hidden information and strategic interactions among more than two players. The work emphasized the potential of combining learning with planning to address large, real-world decision problems.

Limitations include substantial computational requirements and the reliance on domain-specific abstractions to manage the game's complexity. While Pluribus marks a milestone in AI for multi-agent games, its approach is not a general solution to artificial intelligence and does not imply broad competence beyond curated, high-constraint domains like poker.