It took scientists more than two years to achieve this achievement.
In January 2017, it became known that the Libratus computer program beat four professional poker players and won two million dollars.
However, its creators decided not to stop there and have now presented the world with a new version of the program - Pluribus, which has already beaten five Texas Hold'em champions in this game, "sitting" with them all at the same table, just like in traditional casinos. The journal Science writes about the development of scientists.
According to Noam Brown of Carnegie Mellon University in Pittsburgh (USA), one of the creators of Pluribus, artificial intelligence, developing its strategy, has created several innovative techniques that can further influence the game of poker professionals.
The authors of Pluribus equipped it with "abstract action" and "abstract information" - so the program was able to discard unnecessary options for further moves and leave only those that are required to make a decision. During the Pluribus tests, using a self-learning method and a system of "minimizing regrets" (when the program minimizes the difference between the already made decision and the optimal strategy), I fought poker with five copies of itself, at first playing at random. Subsequently, the AI learned to analyze the moves and understood what actions and with what probability lead to victory in the game.
The main part of the experiment took place over 12 days in two stages: one player against five copies of Pluribus, or five professionals against AI.
Five different poker champions participated in the game every day, each with over a million dollars in the past. As a result, with the help of its unique and somewhat “paradoxical” strategy, the AI beat all the best players in more than ten thousand hands and earned half a million dollars.
“The main advantage of Pluribus is the ability to combine different strategies. This is where professional poker players go. Few people manage to change tactics in a random way and stay in the black. Nearly all humans fail to do this,”said World Poker Tour record holder Darren Elias, who also suffered a loss in an AI game.
Promotional video:
Maria Azarova