Recently, at the River Casino in Pittsburgh, Pennsylvania, four of the world's top Texas Hold'em players competed fiercely with the Libratus artificial intelligence system developed by Carnegie Mellon University.
Eventually Libratus defeated the human player and won.
They are compared to "Unlimited Texas Hold'em", and the betting of this complex poker game often goes through many hands. The game lasted for 20 days. Every day before 11am, four poker players started sitting in front of the computer screen and started a "duel" with the Libratus-controlled computer system. At least 1,500 hands were played every day until after 10 pm. At the end of the game, they played a total of 120,000 hand cards.
In the rules of Texas Hold'em, each player has 2 cards as a "bottom card" and 5 community cards. The player combines his 2 cards and 5 community cards to select 5 cards, regardless of the number of cards in his hand (you can use the cards in your hand) to make the biggest card, which is the size of other players. The winner is determined by the size of the final card.
Unlike Go, in the Texas Hold'em game, both players have hidden cards; and people will use irrational strategies such as “deception, speculationâ€; but in Go, all information on both sides is open and symmetrical. This allows AI to no longer adopt the same learning strategy as AlphaGo - using deep learning to analyze the 30 million pieces of human players to learn the skills of Go, and then improve the skills by playing chess with yourself.
What the computer needs to deal with in Texas Hold'em is the "game of incomplete information." According to Wired, Carnegie Mellon University has adopted a set of algorithms called Counterfactual regret minimizaTIon (counterfactual regret minimization). It will let Libratus repeatedly play self-games, playing hundreds of millions of hands of poker at random, to the challenge of top poker players.
But in the end, what makes Lirapus really better than human players is that it can increase the range of betting and randomness to the level that human players can't reach by the absolute advantage of calculation and statistics, which makes it difficult for human players. Guess what kind of cards are in the hands of the computer.
Before Carnegie Mellon University's Libratus. Several scientists in Canada and the Czech Republic have published DeepStack, a algorithm that can defeat human players. Its principle is similar to that of Libratus, and both sets of artificial intelligence systems focus on allowing computers to reason about specific situations in the game, unlike before. Just need to run through all possible situations.
After Go was captured by artificial intelligence, why did the top Texas Hold'em players not rival artificial intelligence? This makes many players feel skeptical.
Because the advantage of artificial intelligence lies in computing power. But there will be many human factors in the Texas Hold'em competition, such as "luck", "mutual fraud" and even "competitive" elements.
In the past, computers rely on the absolute advantages of calculation and statistics to obtain the victory of competitive games that are highly dependent on reasoning and computing. But the process of measuring people's minds and understanding human emotions is the hardest ability of computers to learn.
Therefore, in the poker competition held in Pittsburgh, artificial intelligence will have an absolute advantage when it is necessary to play more than a certain number of hands. After the probability of human factors is diluted, the advantages of artificial intelligence are highlighted. It can record the patterns and routines of every hand of human beings. When he collects the data of human understanding of poker, human beings can't fight at all.
But in the end, it's still the result of learning and training based on a lot of data, not that the machine really understands your emotions and psychology.
Electronic Cable,Electronic Wire,Red Yellow Black Electron Wire,Connector Cable For Controller
Dongguan ZhiChuangXing Electronics Co., LTD , https://www.zcxelectronics.com