Facebook has teamed up with a US university to release a revolutionary poker program using artificial intelligence.
The social media giant has been working with Carnegie Mellon University in Pennsylvania to make the bot, called Pluribus, which has won against some of the world’s best poker players time and time again.
It follows AI bot Libratus which was developed at the same university in 2017, specializing in Texas Hold ‘Em.
Libratus won $1,766,250 in chips in a marathon 20-day poker competition.
This time, however, the new and improved bot has a better and faster online search algorithm which stays just a couple of moves ahead at all times.
Its specialty is in six-handed no-limit hold ’em, one of the most popular poker games which requires players to successfully combine theory, gambling, and strategy to stand a chance of winning.
Known as an ‘imperfect game’ with hidden cards and no betting restrictions, the game is particularly hard to train an AI bot to win. The new software had to be programmed to deal with misleading information – otherwise known as bluffs.
During a testing period of 12 days and around 10,000 hands, Pluribus played against 13 professional poker players.
One test had five bots present alongside the professional while the other had five humans.
These trials involved high-end players such as Jimmy Chou, Trevor Savage, Greg Merson, and Darren Elias.
Each one of these players has previously won at least $1M during professional poker games.
A share of a $50,000 prize was offered to the players which would be split based on the performance of each.
In a statement on the release, Facebook said: “If each chip was worth a dollar, Pluribus would have won an average of about $5 per hand and would have made about $1,000/hour playing against five human players. These results are considered a decisive margin of victory by poker professionals.”
In the human trials, the bot reported an average win rate of five big blinds per 100 hands, with each player losing 2.3 bb per 100.
In the AI tests, results varied, however individual player results ranged from -0.5 to -4 big blinds per 100.
Throughout the trials, no consistent weakness was found with the bot.
Adrian Sireca from Online Casino Gems is very excited about the future of AI in the casino space. “Probably the most impressive thing about the artificial intelligence bot is that it took just eight days and $150 to program. It uses 512GB of RAM, 128 GB of memory and runs on 2 CPUs. The bot also plays much faster than the typical human – taking just 20 seconds per hand.
Even professionals like myself are impressed with the product”.
Mr Sireca is correct. Two-time WSOP runner-up Jason Les said: “I probably have more experience battling against best-in-class poker AI systems than any other poker professionals in the world.
“I know all the spots to look for weaknesses, all the tricks to try to take advantage of a computer’s shortcomings.
“In this competition, the AI played a sound, game-theory optimal strategy that you only really see from top human professionals and, despite my best efforts, I was not successful in finding a way to exploit it.
“I would not want to play in a game of poker where this AI poker bot was at the table.”
This inexpensive yet effective bot has also proven to be record-breaking: it is the first time any AI bot has beaten professionals in any major game with more than two players or teams.
Research scientist at Facebook AI Research and co-creator of Pluribus Noam Brown told The Verge: “It’s safe to say we’re at a superhuman level, and that’s not going to change.”
Pluribus was first trained by playing against copies of its coding in order to allow the bot to develop through trial and error. The bot was then tweaked to watch the game just a few steps ahead.
It can play exactly like a professional would – able to successfully bluff its opponents without using any facial recognition technology.
The code plays with every possible hand then computes a strategy based on those possibilities to avoid playing the best hand every time, which would make it predictable and easy to win against.
Even professional player Sean Ruane pointed out the bot’s “relentless consistency” when playing.
Noam Brown and his colleague Tuomas Sandholm hope that Pluribus won’t just be successful in this one field, but that the technology can be rolled out into areas such as cybersecurity and fraud prevention.
After all, the variables remain the same: a win or lose situation, two or more players, and hidden information.
Professional player Darren Elias said in a CMU statement: “Its major strength is its ability to use mixed strategies.
“That’s the same thing that humans try to do. It’s a matter of execution for humans — to do this in a perfectly random way and to do so consistently. Most people just can’t.
“The bot wasn’t just playing against some middle-of-the-road pros. It was playing some of the best players in the world.”
Tuomas Sandholm has lead a research team at CMU who have been working on strategic reasoning technologies for over 16 years.
From this, he developed two companies, Strategic Machine Inc and Strategy Robot Inc, which license the technologies developed in the university laboratory.
Strategic Machine is currently working to tie the newly-found technology to poker, gaming, business, and medicine, while Strategy Robot is working to see how the coding could fit in the defense and intelligence sector.
Brown has said Facebook has no plans to use the bot’s techniques developed in six-player poker games, but has said the coding could be useful when it comes to developing more advanced computer games.
Details of the team’s research and testing have been published in ScienceMag.
