While AI had some success at beating humans at other games such as chess and Go (games that follow predefined rules and aren’t random), winning at poker proved to be more challenging because it requires strategy, intuition, and reasoning based on hidden information. Despite the challenges, artificial intelligence can now play—and win—poker.

Artificial intelligence systems including DeepStack and Libratus paved the way for Pluribus, the AI that beat five other players in six-player Texas Hold ’em, the most popular version of poker. This feat goes beyond games. This achievement means that artificial intelligence can now expand to help solve some of the world’s most challenging issues.

Tuomas Sandholm, a CMU professor who helped create Pluribus, stated in a press release: “The ability to beat five other players in such a complicated game opens up new opportunities to use AI to solve a wide variety of real-world problems.”

DeepStack: Scalable Approach to Win at Poker

The DeepStack team, from the University of Alberta in Edmonton, Canada, combined deep machine learning and algorithms to create AI capable of winning at two-player, “no-limit” Texas Hold ’em, a game more complex for AI to master than others because of its random nature, hidden cards and players’ bluffs. DeepStack’s neural networks were trained by solving more than 10 million poker game situations. The AI relies on its neural networks to determine the best moves. DeepStack played two-player Texas Hold ’em against professional poker players from the International Federation of Poker. After playing 44,852 games, DeepStack’s results were ten times what a professional poker player considers a sizable margin.

Libratus: Masters Two-Player Texas Hold ’Em

Libratus is an AI, built by Noam Brown and Tuomas Sandholm of Carnegie Mellon University in 2017, that was ultimately unbeatable at two-person poker. This system required 100 central processing units (CPUs) to run. Libratus played 120,000 hands in a 20-day poker competition against four top-ranked Texas Hold ’em players. It won by a staggering amount and walked away with $1.8 million in chips. 

Pluribus: Superhuman Poker-Playing Bot

A very significant milestone was achieved by Pluribus, a robot that was able to beat some of the best poker players in the world in a game of six-player Texas Hold ’em. The scientists at Carnegie Mellon University along with Facebook AI collaborated on the project—the first where AI competed in a game against more than one person and where it couldn’t just rely on game strategy to win. Now that artificial intelligence can beat multiple players in such a complicated game, it’s the gateway to solve some of the world’s most vexing issues such as automated negotiations, drug development, security and cyber-security, self-driving cars and better fraud detection.

Pluribus’ results were impressive. It played 10,000 hands of poker against five others from a pool of million-dollar earners in poker. On average, Pluribus won $480 from its human competitors for every 100 hands-on par with what professional poker players aim to achieve.

The research team created Pluribus by building on what it learned from Libratus. It radically overhauled the search algorithm. Typically, part of the success formula for AI when playing strategic games against an opponent is to process through decision trees until the end of the game before making a move. However, in a multi-player game, this process wasn’t feasible because there was too much-hidden information and the possibilities to process are much greater. The solution for Pluribus was that it only looks ahead a few moves to determine the action it would take, rather than evaluate all moves until the end of the game. The AI teaches itself through reinforcement learning, continually looking back at plays and evaluating the success based on the circumstances. If it determines the outcome would have been better with a different move, it learns to apply that to future play.

Before competing against humans, Pluribus played trillions of hands of poker against itself. Then it faced off against one professional poker player; when it made a mistake, the player alerted the team. Soon, the bot was improving very rapidly given the new information moving quickly from being mediocre to a world-class poker player. Ultimately, it determined its own style of play, even adopting mixed strategies based on the situation to beat five other players.