Sat. Jun 3rd, 2023
Online poker pro Dong Kim plays poker against an AI program in 2015. He lost to an updated program in this year's rematch event.
enlarge / Online poker pro Dong Kim plays poker against an AI program in 2015. He lost to an updated program in this year’s rematch event.

Carnegie Mellon University

Artificial intelligence has had some notable victories in the field of online gaming in recent years, albeit mostly in slower games with clear – albeit incredibly multi-threaded – control systems. On Monday, the robots continued with a slightly more remarkable online gaming victory over their puny human masters when an AI program won big at Texas Hold ‘Em poker.

A long tournament of Hold ‘Em, especially the heads-up, no-limit variant, ended with a win for Libratus, an AI program developed by a professor and PhD student at Carnegie Mellon University. Libratus emerged victorious after 120,000 combined poker hands were played against four human online poker professionals. Libratus’ $1.7 million margin of victory, combined with so many hands, clears the primary bar of the “Brains Vs. Artificial Intelligence” tournament: victory with statistical significance.

Despite Texas Hold ‘Em’s intriguing bluff-related wrinkles, Libratus seems to get the bluffs. Carnegie Mellon’s Explained AI Program Design Describes Emphasis on “Algorithms” [used] to analyze the rules of poker and determine its own strategy”—with, you know, 15 million supercomputer hours for that purpose. Libratus’ primary computer, called Bridges, performed additional routines both while playing live and in the nightly hours thereafter every match day.

As IEEE Spectrum reports, this strategy differs significantly from other poker bots that have more consciously focused on investigating and exploiting human “errors” during any given hand. “When you exploit opponents, you open yourself up to exploitation,” Carnegie Mellon researcher Tuomas Sandholm told IEEE Spectrum.

IEEE also confirms that the four human players hatched a plan together to come up with an initial strategy: play with the bet sizes to see if that would uncover any quirks in Libratus’ betting and thinking. While the folks were able to eat into Libratus’ chip lead to some degree, the program’s lead eventually exploded, tipping the “statistical significance” scales.

Researchers at the Carnegie Mellon project point out that the amount of supercomputing required to perform this high-level calculation is extreme – meaning average online poker rooms shouldn’t be overrun with genius robots handling countless gaming options in real time. (By the way, Bridges’ supercomputer did this all at once, processing and playing hands against all four human players at once.) For their troubles, the four suckers — er, I mean, poker pros — shared a $200,000 purse according to their gig. against Libratus.

Carnegie Mellon’s team also emphasizes that AI’s success in games with “incomplete” and “misleading” information has notable implications for other AI breakthroughs, particularly in the fight against biological, rapidly evolving viruses. (The same researchers failed to estimate the possibility that the same AI programs could trick all of humanity into doing their bidding — which may or may not mean those robots have already proved successful.)

By akfire1

Leave a Reply

Your email address will not be published.