AI vs. Humans: Artificial Intelligence Beats 6 Professional Players at No-limit Texas Hold'em Poker
Scientists have developed an AI program which can beat professional poker players at the card game, in what they described as a milestone for artificial intelligence.
The AI named Pluribus beat top human professionals while playing six-player, no-limit Texas hold'em poker. According to the authors of the study published in the journal Science, this is the most popular form of the game where players attempt to win a pot of money by waging bets that they have the best hand.
As artificial intelligence has become more advanced in recent decades, experts have attempted to create machines which can overcome certain problems.
AI could already beat humans in the zero-sum two player games of chess, checkers, GO, and two-player limit and two-player no-limit poker, the authors explained. But scientists have found it difficult to scale up to six-player poker.
In 10,000 hands of poker, Pluribus played against five copies of itself versus one top-class professional player, as well as five top-class professional poker players versus one version of itself. The scientists found it performed significantly better than humans on average.
Study co-author Dr. Tuomas Sandholm, Professor of Computer Science, Carnegie Mellon University, told Newsweek: "Poker is the main benchmark problem for testing imperfect-information game solving capability—both in the AI community (at least since 1970) and the game theory community (at least since 1950)."
He explained: "All the AI gaming milestones in history have been for two-player zero-sum games. This is the first such milestone for multi-player games. Multi-player games present additional challenges not present in two-player zero-sum games."
It was hard for researchers to move beyond two-player poker because of the size of a six-player game. They are much bigger, and experts needed to find a way to move away from the Nash equilibrium, a game theory where no player gains from changing their strategy if the strategies of others remain unchanged.
"Strategies are much harder to compute outside of two-player zero-sum games," explained Sandholm. However, the newest algorithm was "strong and scalable," he said.
Asked to describe the potential uses of the research in the future, Sandholm said strategic reasoning technologies have a range of applications from poker and video games, to strategy optimization in investment banking, political campaigns, and even steering evolution and biological adaptation "such as for medical treatment planning and synthetic biology and so on."
The technology also has uses in electronic warfare, security and "optimizing world stability," he said.