Carnegie Mellon University (CMU) researchers developed an artificial intelligence (AI) program that beat elite professional poker players at six-player no-limit Texas hold'em poker.
CMU's Noam Brown and Tuomas Sandholm created the Pluribus AI by updating an earlier program, Libratus, which only plays two-player matches.
The researchers revamped Libratus' search algorithm, which searches to the end of a game before selecting an action. Adding more players negated the practicality of this approach, so Brown and Sandholm invented a technique that permitted Pluribus to make good choices after looking ahead only a few moves.
Pluribus trained itself by initially playing poker randomly, and improved as it ascertained which actions won more money; after each hand, it reevaluated its moves, and checked whether it would have won more with different actions, which it will be more likely to utilize later on.
View Full Article
Abstracts Copyright © 2019 SmithBucklin, Washington, DC, USA