Robot Libratus Won 1.8 Million Dollars at Texas Hold’em
Researchers from Carnegie Mellon University (Pennsylvania) developed an algorithm that defeated four of the strongest players in the so-called Texas Hold’em (one of the most popular varieties of poker).
Poker players plan actions based on information about cards, as well as studying the mimicry and behavior of rivals. For the program, this approach is too complicated, so earlier developers of algorithms for playing poker used various ways to facilitate its task.
However, the system of artificial intelligence developed by American scientists called Libratus can think “abstractly”. The algorithm is able to find common features in various combinations of cards, bets and other elements of the game. If an opponent chooses an unexpected way out, then Libratus processes its consequences separately with the help of a supercomputer. Successful versions are recorded in a special “strategy log”, which Libratus uses in subsequent rounds.
In the course of working on the system, scientists first tested it in games with previously created algorithms for poker, and then invited four leading players to Texas Hold’em and organized a tournament with a prize fund of 200 thousand dollars, 10% of which was guaranteed to each of the famous players. As a result, the tournament ended in a complete defeat of people – at the end of the game, there were chips for 1.8 million dollars on the hands of Libratus. If the tournament was held by all rules, then the entire prize fund would have gone to artificial intelligence.
The developers emphasized that techniques they developed when creating this artificial intelligence system are not specific to poker. They surely can be used to solve any other problems when there is a lack of information for making a decision.
Earlier great success in this area has been made by Google-owned company DeepMind. The AlphaGo algorithm easily defeated the world’s strongest Go players. Recently the company introduced an improved version of the program that defeated AlphaGo and mastered chess and shogi.