HHpoker created its own AI
April 23, 2022
HHPOKER has created an “artificial intelligence game system” that can play Texas, chess, go, poker. Depmind, the artificial intelligence lab of Google’s parent company Alphabet, has long invested in artificial intelligence game systems.The concept of the lab is that while games have no obvious commercial application, they are unique challenges to cognitive and logical skills.This makes them a useful reference point for the development of artificial intelligence.
Unlike previously developed game systems,HHPOKER has officially created a system called Playerofgames, which is the first artificial intelligence algorithm to achieve powerful performance in full information games and incomplete information games.Unlike other game systems developed previously by Deepmind, such as Alphazero, National Chess Champion and IIAlphastar, players can perform well in full information games (such as Chinese Go and National Chess) and incomplete information games (such as poker).
Whether it’s road planning to solve traffic congestion problems, negotiating contracts, communicating with customers and other interactive tasks, we have to consider and balance people’s preferences, which is very similar to game strategy.Artificial intelligence systems can benefit from coordination, cooperation and collaboration between groups or organizations.Systems like Playerofgames can identify the goals and motivations of others, allowing them to successfully collaborate with others.
Not quite right.
During the game, information about the incomplete information game is hidden from the player. unlike the full information game, all information will be displayed at the beginning.
Playing a full information game requires a great deal of predictability and planning.Players must deal with what they see on the board, decide what their opponents can do, and strive for the ultimate goal of winning.Games with incomplete information require players to consider hidden information and consider how to win next, including possible bluffs or teams against opponents.
Hhpoker said Playerofgames is the first general and reliable search algorithm and has achieved high performance on both official sites and incomplete information games.
Playerofgames is very versatile, but not all games can be played.Martin Schmid, a senior researcher at Depmind who participated in the study, said that Alphazero is more powerful than Playerofgames in full information games, but it is not as strong in incomplete information games.The system must consider all possible perspectives of each player in the game.While there is only one perspective in a full information game, there can be many in an incomplete information game. In poker, for example, there are about 2,000 perspectives.Furthermore, unlike Depmind’s successorAlphazero Muzero, Playerofgames must also understand the rules of the game he is playing, and Muzero can immediately master the rules of a full information game.
Depmind’s performance in the national discipline chess, Go, Texas and strategy board game “Scotland Yard” evaluated Playerofgames’ use of Google’s accelerated TPUV4 chipset for learning. For go it set 200 games between Alphazero and Playerofgames, while for national chess Depmind allows Playerofgames to compete with top-notch systems such as GnuGo, Pachi, Stockfish and Alphazero.Texas game Playerofgames uses publicly available Slumbot, and the algorithm also competes with Pimbot, developed by Josephantonin.
In Kokuji chess and Go Playerofgames was more powerful than Stockfish and Pachi in certain configurations, winning 0.5% of games against the strongest system Alphazero.Despite heavy losses in the game against Alphazero, Depmind believes that Playerofgames performance has reached first-class human amateur levels and may even reach professional levels.
The results show that Playerofgames is a more interesting home for Texas and Scotland Yard.When fighting Slumbot, the algorithm wins an average of 7 million big blinds (mbb/hand) per hand, and mbb/hand is the average number of big blinds won per 1,000 hands.
At the same time in Scotland, Deepmind said that although Pimbot had a better chance of finding winning tricks, Playerofgames beat him significantly.
Material was prepared by the POKERBOTAI team