AI vs Poker Legends


Libratus is the name of a new AI developed by Carnegie Mellon University, to challenge the poker pros Jason Les, Dong Kim, Daniel McAulay and Jimmy Chou. The event is called “Brains Vs Artificial Intelligence: Upping the Ante” and is scheduled to take place from January 11 at Pittsburgh’s Rivers Casino. The decision will give us an idea on how the players perform against the machine.

The event will continue for 20 days and will have 120,000 hands-up no-limit action between the  poker pros and the new AI software for a share of $ 200,000 in prize money.

The pros will pair up and play duplicate matches (with each pair receiving the same cards as the computer in each scenario) on the casino floor and in an isolated separate room. The increased number of days and the “two-table” play (playing two hands simultaneously) will increase the chance of reaching statistical significance. The duel will be open for general public who can watch the proceedings daily during playing hours which will stretch from 1 am to around 7 pm.

The Carnegie Mellon University took part in a similar event last year as well. The Artificial Intelligence program Claudico, however failed to win the event as it finished below 3 of the 4 human competitors. It played 80,000 hands and collected the 2nd lowest number of chips.

This year, Libratus has been created to set a new benchmark for artificial intelligence.

“Since the earliest days of AI research, beating top human players has been a powerful measure of progress in the field,” said Tuomas Sandholm, the professor of computer science at Carnegie Mellon who created the bot with his PHD student Noam Brown.

“That was achieved with chess in 1997, with Jeopardy! In 2009 and with the board game Go just last year. Poker poses a far more difficult challenge than these games, as it requires a machine to make extremely complicated decisions based on incomplete information while contending with bluffs, slow play and other ploys,” he added.

But Sandholm has learned from last years’ experience. “We knew Claudico was the strongest computer poker programme in the world, but we had no idea before this competition how it would fare against four Top 10 poker players,” he had said then.

Libratus is not merely an improvement on Claudico as it was built from the start. It was helped by the Pittsburgh Supercomputing Centre’s Bridge supercomputer to calculate the potential winning strategy for imperfect information games.

A massive improvement over Claudico, 15 million core hours of computation has been used for Libratus versus the two to three million core hours for Claudico. Other improvements include some “weird moves” like limping (a favoured strategy of the old bot which was exploited by the players) and new technology to achieve Nash equilibrium – a strategy that neither player can benefit from changing strategy if the other player’s strategy remains the same.

The Latin meaning of Libratus is balanced and powerful, and it will have to show exactly these two traits it will need to beat the human competitors. It has a new endgame-solving approach that has uses a method to find equilibrium faster, identifying hands that are not promising and ignoring these paths the next time, it uses the Bridge’s computer to do live computations.

The results from this game are likely to have implications for other sectors that are characterised by incomplete and misleading information, like business, military, cyber security and medicine.