Poker Bot Pluribus Becomes First AI to Beat Humans in Six-Max NL Hold’em
Researchers at the Carnegie Melon University (CMU) are celebrating after the poker bot they created managed to do something that no other poker artificial intelligence (AI) has managed to do: beat human opponents in a multiplayer no-limit hold’em cash game.
Ph.D student Noam Brown and Professor Tuomas Sandholm created a poker bot named Libratus in 2017 that defeated some of the best heads-up no-limit hold’em professionals in a one-on-one scenario. Libratus was designed to take on one opponent at a time by running complex algorithms that allowed it to adjust its strategy in real-time.
A Poker Bot To Play Multiple Opponents
The number of variables for an AI to calculate when it faces multiple opponents had proven to be a stumbling block for researches, until now. Brown and Sandholm set about creating a brand new poker bot that they would then test against some strong no-limit hold’em cash game players in a six-handed format. Pluribus was born.
Brown and Sandholm devised two experiments. One where five humans took on Pluribus, and another where five of the AI were pitted against a single human opponent. To make this second experiment fair, the AIs were unable to communicate with one another.
Over the course of the two experiments, during which time hundreds of thousands of hands were played, Pluribus managed to wipe the floor with the humans to the tune of $5 per hand and almost $1,000 per hour.
A Self Taught Poker Bot
Pluribus actually taught itself how to play and adopted new strategies as it tested what plays worked and which were not profitable. It is common knowledge that limping from any position at the table other than the small blind is considered a suboptimal play and Pluribus agreed, dropping this move early into the experiment.
As Pluribus developed new strategies, it made some unconventional plays that seemed to catch the humans off guard. Professional poker players will often make plays outside of the norm in an attempt to keep their opponents second-guessing and Pluribus followed suit. The poker bot made some unusually large bets when it was value betting and semi-bluffing and was paid off handsomely on more than one occasion.
Former World Series of Poker Main Event champion Chris “Jesus” Ferguson said of Pluribus, “It’s really hard to pin him down on any kind of hand. He’s also very good at making thin value bets on the river. He’s very good at extracting value out of his good hands.”
Jason Les mirrored the rest of the humans in stating the unconventional bet sizing caught them off guard and that the poker bot was by far a better bluffer than a human opponent.
“It is an absolute monster bluffer. I would say it’s a much more efficient bluffer than most humans. And that’s what makes it so difficult to play against. You’re always in a situation with a ton of pressure that the AI is putting on you and you know it’s very likely it could be bluffing here.”
Poker Bots in the Online Poker World
Quite worrying from an online poker perspective is how little computing power this latest poker bot needed compared to the heads-up bot Libratus. Libratus needed 100 CPUs during its matches in 2017 but Pluribus ran on only two CPUs and equipment readily available to the general public.
Poker bots have long been an issue for online poker sites. Wherever there is money to be made, unscrupulous people will attempt to find a short cut.
One of the most high profile cases happened in 2015 when members of the Two Plus Two forums uncovered a “bot-ring” playing at PokerStars’ Pot-Limit Omaha cash game tables, mostly $0.50/$1 an $1/$2 stakes, where they had allegedly won a staggering $1.5 million. PokerStars shut down the poker bot accounts and redistributed as much of the ill gotten funds as it could.
partypoker as recently clamped down on third-party software as part of what it calls FairPlay. The second-biggest online poker site in the world forced all of its customers to change their alias in an attempt to render obsolete any databases of compiled hand histories. Other changes included the banning of heads-up displays, removing the ability to download hand histories and making it impossible to spectate cash games that the player is not seated at.
For the most part, online poker sites employ sophisticated techniques to detect poker bots. While some of the poker bots do get through, customers who are vigilant ca report suspect accounts to the sites’ customer support teams and any claims will be investigated fully.