Technology and computers are improving each year, with modern machines capable of performing complex tasks and incredibly difficult calculations extremely quickly. While computers are mainly used for serious things in medical, engineering, and other industries, they can also be used for fun. Computers allow us to access a world of information and play games like poker online. Modern poker software can even be used to give players an advantage over their opponents, showing them the best course of action to take.
Poker software has been around for some time, but it mostly helps already experienced players to improve their game. It can't win the game for you on its own. However, it might not be long before it can. In 2019, an AI created by Carnegie Mellon University defeated some of the world's best professional players in a six-player game of Texas hold'em. It may be a long time before you're playing poker online against a computer, but it's clear that AI is now capable of beating the best players.
Pluribus AI
The Carnegie Mellon University researchers who created this poker AI named it Pluribus. Just like chess grandmasters have now been beaten at their own game by an AI, so too have professional poker players. Poker player Darren Elias, who holds the record for the most World Poker Tour victories, and Chris "Jesus" Ferguson, a six-time World Series of Poker champion, were both defeated by computer software. Five copies of Pluribus were used in the 5,000 hands of poker played by each professional.
Pluribus played five pros at a time for a total of 10,000 hands in another trial with 13 professionals, all of whom have won more than $1 million playing poker, and again prevailed. Its statistically significant victory was made all the more impressive given its opposition. The AI wasn't just competing against average professionals. It was playing against some of the most players in the world.
Previously, two of the researchers who created Pluribus had developed Libratus, another AI capable of beating pros at heads-up Texas hold 'em. Together, the research team has been exploring the possibilities of AI poker for 16 years, and their work isn't done yet.
How Does Pluribus Work?
Pluribus is a type of AI or artificial intelligence. As a machine, it doesn't have real intelligence, but it's able to learn by spotting and predicting patterns in behavior. It also makes multiple calculations using probability, working out potential outcomes of the game. AI research has traditionally used games like Go and Chess as benchmarks. Every player in those games is aware of the current state of the playing field and every piece.
However, poker is far more challenging for an AI as there's no way to be certain which cards are in play. In addition, bluffing makes it even more difficult to read the intentions of opponents. As a result, it is both a more difficult AI challenge and more applicable to a variety of real-world issues involving multiple factors and incomplete data.
All AIs that are good at two-player games approximate what is known as a Nash equilibrium. This is a set of two strategies (one for each player), neither of which may be altered while the other player's strategy is held constant. The AI wins if its opponent makes errors and is unable to preserve equilibrium, even though its strategy only ensures a result no worse than a tie.
For multi-opponent games like poker, using Nash equilibrium is more likely to lead to losses. Instead of trying to avoid losing, Pluribus makes use of strategies that help it outplay its opponents. It seeks to be as unpredictable as possible, as only betting when holding the best hand quickly reveals to opponents when it has a strong hand and when it has a weaker hand.
Pluribus also limits how far ahead it calculates, aiming to be as efficient as possible. It looks ahead several moves rather than to the end of the game, as this would require too much computer power and wouldn't necessarily produce the best results.
