machine and human fist bump

05 Nov 2019

New poker AI so good it’s not being released

Share this

Online poker continues to grow, with amateur and pro players alike using companies like Skrill to manage their bankroll.

The latest news in the online poker world comes from the best player in the world — a computer program.

When it comes to chess, researchers have developed an AI to beat the best in the world. But no AI has cracked the complex game of multiple-player poker, until now...

Open a Skrill account

Meet Pluribus, the superhuman poker bot built for $150

In 2017, a poker bot named Libratus was developed by researchers at Carnegie Melon University (CMU). The bot beat some of the best heads-up poker pros in the world in a game of Texas hold'em, including well-known players such as Chris Ferguson, Jason Les, Dong Kim, Daniel McAulay and Jimmy Chou. 

group poker of 5 people

The bot’s success was a milestone in the development of AI, but its applications were limited. It was only able to beat one opponent at a time — in heads-up play.

But now, Pluribus, the latest bot developed by the same researchers (in a joint project between Facebook AI and CMU) has achieved something that no other AI has managed. 

Pluribus defeated multiple strong players in a game of no-limit hold'em in a six-handed format. And it did all this more efficiently than any other documented poker bot before it. 

In fact, the algorithm was so successful that the researchers have decided not to release its code for fear it could be used to bankrupt online poker companies.

typing on laptop

The secret? Self-play

Pluribus' strategy was mostly computed by "self-play", meaning it developed its strategy based on playing copies of itself, rather than learning from hands played by humans or other AI units. 

Because of this, Pluribus has also made some unconventional plays. Many proved profitable against some of the best players in the world, including “donk betting” (ending one round with a call but then starting the next round with a bet). 

One player noted that Pluribus was exceptional at making itself unpredictable, changing strategy even when playing identical hands. The bot also bet — and bluffed — big. 

The AI played so well that it won about $1,000 (in chips) an hour against its opponents.

Simply put, Pluribus plays very differently from how humans play the game. The bot learned the timing of huge bets and bluffs and produced a combination of unpredictability and bold moves that humans would struggle to match.

This is obviously a big advancement in AI, and there’s an inevitability about it getting even better in the years to come. So, what does this mean for online poker?

AI milestone, or poker tombstone?

The code for this AI is being kept secret and not being released, which is good news for online poker. 

Although poker was the chosen experiment, this is much more about advancing AI in general. The scientists behind Pluribus say the victory is a significant milestone in AI research. 

Machine learning has already reached superhuman levels in board games like chess and Go. However, computer games like Starcraft II and six-person no-limit Texas Hold ‘em represents, by some measures, a much higher benchmark of difficulty. 

For these games, the information needed to win is hidden from players (making it what’s known as an “imperfect-information game”). They also involve multiple players and complex victory outcomes.

Applications beyond poker

The techniques — AI learning and the tactical algorithms — used to create Pluribus will be transferable to other situations. Cybersecurity, fraud prevention, and financial negotiations for example. Or even helping navigate traffic with self-driving cars.

The thing is, we shouldn’t be too worried as AI is already used by poker players for calculating optimal bankroll sizes. AI is currently, and will continue to be an incredibly useful tool, rather than a method for cheating. So, for now at least, the robots are on our side…

machine and human fist bump
Back to news