12.07.2019, 11:19
Artificial Intelligence Beat Five People in Poker at the Same Time
Source: OREANDA-NEWS
OREANDA-NEWS. Scientists have created artificial intelligence (AI), which managed to beat in Texas Hold'em (the most popular form of poker) five professional players. The success of scientists from Carnegie Mellon University (CPU) is reported in the journal Science.
"In the past two decades, AI systems mastered more and more complex forms of poker. However, all previous breakthroughs were limited to the participation of only two players", the publication said. Compared to the paired games, games with a large number of players caused difficulties for the AI. To overcome their scientists have equipped the system with "abstract action" and "abstract information". This technology helps the system to reduce the number of possible solutions and narrow down the options for selection, with the help of this innovation, training is accelerated.
To do this, experts have developed a program called Pluribus, which learned the game of the most popular form of gambling. Mishina played randomly at all times, and Plurib gradually determined which moves brought the greatest gains, and which ones led to the loss. Analyzing this data, the system gradually improved its strategy. A similar method was previously used to develop smart bots for disciplines such as Dota 2, Starcraft 2, and doubles poker. But now used a new way of learning.
The experiment was carried out in several ways, in first one several cars competed with one person, in the other case 1 car confronted five people. It's worth noting that all the players had great experience in this game and earned a large fortune in their careers. Among them were famous Chinese, American and Korean players, as well as some other famous poker professors.
In both cases, the AI showed excellent results, the car was able to beat several people at the same time, and several cars easily beat one person.
"In the past two decades, AI systems mastered more and more complex forms of poker. However, all previous breakthroughs were limited to the participation of only two players", the publication said. Compared to the paired games, games with a large number of players caused difficulties for the AI. To overcome their scientists have equipped the system with "abstract action" and "abstract information". This technology helps the system to reduce the number of possible solutions and narrow down the options for selection, with the help of this innovation, training is accelerated.
To do this, experts have developed a program called Pluribus, which learned the game of the most popular form of gambling. Mishina played randomly at all times, and Plurib gradually determined which moves brought the greatest gains, and which ones led to the loss. Analyzing this data, the system gradually improved its strategy. A similar method was previously used to develop smart bots for disciplines such as Dota 2, Starcraft 2, and doubles poker. But now used a new way of learning.
The experiment was carried out in several ways, in first one several cars competed with one person, in the other case 1 car confronted five people. It's worth noting that all the players had great experience in this game and earned a large fortune in their careers. Among them were famous Chinese, American and Korean players, as well as some other famous poker professors.
In both cases, the AI showed excellent results, the car was able to beat several people at the same time, and several cars easily beat one person.
Комментарии