Upon Further Consideration, Carnegie Mellon Pokerbot Sweeps Contest

Monday, August 29, 2016 - by Byron Spice

Baby Tartanian8, created by Noam Brown and Tuomas Sandholm, not only took first place in the Annual Computer Poker Competition total bankroll category of the Heads-Up, No-Limit Texas Hold'em game, but it also won the game's bankroll instant run-off category.

Everyone knew Carnegie Mellon's latest computer poker program, Baby Tartanian8, was good. But it turns out its performance in the Annual Computer Poker Competition this year was even better than people thought.

Not only did Baby Tartanian8 take first place in the competition's total bankroll category of the Heads-Up, No-Limit Texas Hold'em game, as announced in February at the Association for the Advancement of Artificial Intelligence meeting in Phoenix, but organizers recently concluded that it also won the game's bankroll instant run-off category.

This is the second consecutive sweep of the game's categories by a Carnegie Mellon pokerbot from the lab of Tuomas Sandholm, professor of computer science.

In the total bankroll category, the winner is determined by total winnings. In bankroll instant run-off, the pokerbot with the lowest total bankroll is eliminated in each round.

Baby Tartanian8, created by Sandholm and Noam Brown, a Ph.D. student in the Computer Science Department, originally was placed third in the bankroll instant run-off category. But it later emerged that an error occurred when additional matches were run to separate the four top-scoring pokerbots for Heads-Up, No-Limit Texas Hold'em. The additional matches erroneously used the same sequence of cards, which biased the results.

When new additional matches were run by the competition organizers, the CMU poker program outscored the other competitors by a statistically significant amount.

During the previous computer poker competition, in July 2014, the CMU team's Tartanian7 won both categories in Heads-Up, No-Limit Texas Hold'em, prompting the School of Computer Science to create a "Brains Vs. Artificial Intelligence" competition in the spring of 2015 that pitted Sandholm's Claudico pokerbot against four of the top 10 Heads-Up No-Limit Texas Hold'em professional players. That two-week, 80,000-hand contest at Pittsburgh's Rivers Casino ended with three of the four pros with higher winnings than the pokerbot, but not quite enough to attain statistical significance.

Tartanian8 was a completely new bot whose strategy was created by algorithms from scratch. The algorithms were more advanced than those used to create Tartanian7 and Claudico.

The CMU team competed with a "baby" version of Tartanian8 that was scaled down significantly to fit within the competition's 200 gigabyte data storage limit. It also could not do sophisticated real-time deliberation because of the competition's  processing limit. The original Tartanian8 strategy was computed in late fall 2015 by Brown and Sandholm on the Comet supercomputer at the San Diego Supercomputer Center.

Poker has become a major test of artificial intelligence, Sandholm explained, because it is an incomplete information game. Players don't know what cards their opponents hold and all players try to mislead their opponents by bluffing, slow play and other devices.

The same sort of algorithms used to play poker could also be used to create strategies for applications involving cybersecurity, business transactions, and medicine — all applications that Sandholm's research group works on. For instance, an AI might help doctors develop sequential treatment plans for a patient, or design drugs that are less prone to resistance. Or, such an AI might help people negotiate their best deal when purchasing a house or a car.

For more information, Contact:
Byron Spice | 412-268-9068 | bspice@cs.cmu.edu