The advancement of Artificial Intelligence (AI) has been broadly explored the past few decades. However, the greatest breakthrough in AI experiments came through recently.
AI and its application to the game of poker are the focus of research for two Carnegie Mellon University scholars. Computer Science professor, Tuomas Sandholdm and Ph.D. student Noam Brown, developed an AI program that is capable of beating out top professional poker players.
To put the invention to the test, Libratus, the name of the program is known by, took on four season poker players in a 20 day No Limit Hold’em heads-up match. He took on Jimmy Chou, Dong Kim, Jason Les, and Daniel McAulay. Libratus managed to beat them all out by ending the tournament with $1.7 million worth of chips.
At first, Libratus had difficulty sorting through imperfect information, while trying to detect whether his opponents were bluffing. This is because Libratus has been developed to deploy a different kind of approach to the game of poker, in comparison to previous AI products.
After each day Libratus would analyse his strategy used throughout the day with the use of a complex meta-algorithm. He also explored the holes his opponents had spotted in his behavior during a round of poker. From here, he would prioritize these behaviours and improve on the three most obvious behaviours.
Libratus was able to self-improve his gameplay each day and his ability to identify the presence and essence of holes in his own strategy are what has made Libratus’s ability ground-breaking.
This has left Libratus for the time being the most proficient program in heads-up No-Limit Hold’em games. However, the more players at one poker table does mean more possible outcomes. Effectively, this does provide more of a challenge for the AI program to go for the best possible moves.