Later this week, the "Libratus" artificial intelligence (AI) system developed by Carnegie Mellon University (CMU) will begin a new challenge: trying to "no limit Texas Hold'em" in one-on-one, unlimited bets ( No-Limit Texas Hold'em) Challenges top human pros in the game, and the winner will receive a $200,000 prize.
Carnegie Mellon University researchers said that they hope that Libratus will eventually win the game, so they can announce to the world for the first time: artificial intelligence has captured Texas Hold'em and defeated the human champion for the first time. But at the same time, Carnegie Mellon’s competitors expressed dissatisfaction.
Researchers at the University of Alberta in Canada, Charles University in Prague, Czech Republic, and the Czech University of Technology recently published a paper that the artificial intelligence system they developed has reached an expert level in Texas Hold'em. Therefore, their artificial intelligence system is the first to defeat humanity.
It should be pointed out that the paper has not yet been "peer-reviewed", so it is still unbelievable. But based on the historical achievements of this team, the results they claim are not unfounded.
In 2008, a team of almost the same team of researchers developed a system that defeated human masters for the first time in the Limit Texas Hold'em game with betting restrictions. In 2015, the team completed an upgrade to the system and said it was near perfect.
Researchers at the University of Alberta said in the paper that the system they developed uses a generic algorithm called "DeepStack," recently in "Heads-up No-Limit Texas Hold". 'em) defeated humanity in the game.
From November 7th to December 12th, 2016, DeepStack conducted a total of 44,852 contests with 33 human players on the Internet. The results of the game show that the average win rate of DeepStack is 492mbb/g. The DeepStack team said that the average human player is considered to have a big advantage at 50mbb/g, while the 750mbb/g is the equity that the opponent folds every game.
But Professor Tuomas Sandholm, who helped Carnegie Mellon University design the Libratus artificial intelligence system, said: "DeepStack has not beaten top players, so it can't be said that it has already defeated humans."
Sandholm also said: "We have reached their level two years ago, but that is not our goal. Our goal has never been 'expert level', but 'superman level', like 'dark blue' and 'Watson'."
So far, the artificial intelligence system can only deal with the "one-on-one" game in the "Limited Note Hold'em" or "No Limit Hold'em" competition. In the simultaneous competition of multiple people (five or more), the complexity is still difficult for artificial intelligence to cope with. But in any case, the results of these competitions show that at least two sets of artificial intelligence systems have to create new milestones.
DeepStack researchers say Texas Hold'em is a typical non-complete information game because you don't know the cards of your competitors or what the next hand is. Therefore, for artificial intelligence, its complexity is more than chess. DeepStack is a general-purpose algorithm that can be used for a large class of sequence incomplete information games. Its achievements show that artificial intelligence systems are proficient in incomplete information games.
Like Google's AlphaGo, Libratus and DeepStack don't try to solve all the problems of the entire game because it takes into account too many possibilities. Instead, the two artificial intelligence systems use deep learning to develop an intuition. DeepStack researchers said: "Before the game, DeepStack does not calculate and store the complete strategy, but uses a fast 'approximate estimation' method, which we call 'intuition'."
If artificial intelligence can beat human masters in the Texas Hold'em game this time, it means that researchers can use this technology to solve some practical problems in daily life, especially those involving incomplete information, such as business negotiations. . In the negotiations, both parties will hide some information.
Noam Brown, a Ph.D. student at Carnegie Mellon University who is involved in Libratus research and development, said: "In the real world, information cannot be as neat and orderly as the chess pieces on the board. There are always some important information lost or hidden. Artificial intelligence needs to be able to handle this problem."
Grain Screw Conveyors,Belt Conveyor For Rice Mill,Hopper Belt Conveyor,Pvc Blet Conveyor
HuangShi Stee Metal Material Technology Exploitation Co.,Ltd , https://www.wosenstechnology.com