Scientists have created an artificial intelligence (AI), who managed to beat Texas hold’em (the most popular form of poker) five professional players. About the success of scientists from the University Carnegie Mellon (CPU) reported in the journal Science.
“In the past two decades, AI systems followed a more complex form of poker. However, all the previous breaks were limited to two players, “—said the publication.
Compared with the paired games, party with big number of players caused the AI difficulty. To overcome them, scientists have equipped a system of “abstract” and “abstract information”. These properties are necessary to help the system to reduce the number of options and to highlight those that are needed for decision-making.
To do this, experts have developed a program called Pluribus, which has learned to play Texas hold’em with five of his copies. Playing over and over again at random, Pluribus gradually determined what actions and with what probability give the best results. By analyzing these data, the system gradually improved its strategy.
The method of learning has previously been used in games for two with a zero amount, such as backgammon, Dota 2, Starcraft 2 and doubles poker. But according to experts, there is applied the algorithms differed significantly from the Pluribus.
How was the experiment
The computer game was evaluated in two formats: five players against Pluribus and one player against five copies of Pluribus.
Each participated in the experiment earned on the poker over $ 1 million and Participants were players Jimmy Chou, Seth Davis, Michael Gagliano, Anthony Gregg, Dong Kim, Jason Forest, Linus Seliger, Daniel of Macaulay, Greg Merson, Nicholas Pietrangelo, Shawn Rouen, Trevor savage, Jacob Toole.
Every day for the game were selected by five professionals. However, they did not know other participants. Each person was given a pseudonym, which remained unchanged throughout the experiment. It was possible to track the progress of players. Between people divided $ 50 thousand to motivate them for the game with full dedication. Players guaranteed at least $ 0.40 for the distribution. The amount could increase to $ 1.60 per hand depending on the game results.
In the match with five professional poker players or five copies of Pluribus playing vs single player AI for 12 days of the experiment showed the best results in more than 10 thousand hands.
As previously reported “FACTS”, artificial intelligence has learned to manipulate photos.
We read in the Telegram channel, Facebook and Twitter