An artificial intelligence-powered, poker-playing bot destroyed professional card slingers in six-player Texas Hold ’em, in the latest example of A.I. showing its dominance over humans in games.
The A.I. system, named Pluribus and created in collaboration between Facebook’s AI lab and Carnegie Mellon University, played 10,000 hands over the course of 12 days against 12 professional poker players. There were two scenarios: Pluribus against five human players, or five versions of the A.I. against one human player.
In a paper published in Science, the researchers revealed that Pluribus won an average of $5 per hand with winnings of $1,000 per hour, which they described as a “decisive margin of victory.”
“It’s safe to say we’re at a superhuman level and that’s not going to change,” Facebook A.I. Research scientist and Pluribus co-creator Noam Brown told The Verge.
This is not the first time that A.I. bested humans in a game of poker. In January 2017, an earlier version of Pluribus, named Libratus, defeated four professional players in heads-up, no-limit Texas Hold ’em. The difference, however, is that Libratus went up against humans in one-on-one games, while Pluribus rose to the challenge in six-player tables.
The difference between two-player and multiplayer games, of course, was huge. In two-player poker, the bots may easily figure out a winning strategy. Things are not as simple in six-player poker games, due to the much higher number of variables and the hidden information, unlike games such as chess where all the positions of the pieces are known at any given time.
Pluribus, however, still came out on top, as it taught itself six-player Texas Hold ’em while formulating strategies that were very different from how humans played the game. The bot learned the timing of huge bets and bluffs, and produced a combination of unpredictability and bold moves that humans simply could not match.
The development of A.I. has seen it move into multiplayer gaming as a learning environment, such as when Google’s DeepMind just over a month ago tackled into Quake III’s Capture the Flag. The mode required A.I. agents to work together to beat teams made up of human opponents, and they were very successful at it.
Pluribus takes things a step further, and Brown believes that the bot’s ability to handle multiple players, hidden information, and numerous possible outcomes may find real-world applications for the benefit of mankind.
- Facebook taps Minecraft as training ground for next stage of A.I.
- Here’s how Facebook taught its Portal A.I. to think like a Hollywood filmmaker