Superhuman AI for strategic reasoning: Beating top pros in heads-up no-limit Texas holdem

Tuomas Sandholm offers an overview of Libratusan AI that beat a team of four top specialist pros in heads-up no-limit Texas holdem, which has 10^161 decision pointsand explains how Strategic Machine is applying the domain-independent algorithms behind Libratus to a variety of imperfect-information games.

Poker has been a challenging problem in AI and game theory for decades. As a game of imperfect information, it involves obstacles not present in games like chess and Go and requires totally different techniques. No program had been able to beat top players in large poker games. . .until now. In January 2017, an AI, Libratus, beat a team of four top specialist pros in heads-up no-limit Texas hold’em, which has 10^161 decision points. Libratus is powered by new algorithms in each of its three modules: computing approximate Nash equilibrium strategies before the event, endgame solving during play, and fixing its own strategy to play even closer to equilibrium based on what holes the opponents have been able to exploit. Tuomas Sandholm offers an overview of Libratus and explains how Strategic Machine is applying the domain-independent algorithms behind it to a variety of imperfect-information games, such as business strategy, negotiation, strategic pricing, product portfolio planning, finance, cybersecurity, military applications, political campaigns, auctions, and steering biological adaptation and evolution for medical treatment planning. (This is joint work with Tuomas’s PhD student Noam Brown.)