Lengpudashi - an improved version of the Libratus poker artificial intelligence machine

Alan Du , a high-risk venture capitalist and participant in the World Series of Poker - an annual world poker tournament held in Las Vegas, USA - is on the fifth day of a brainstorming match with one The player seems to be inadequate. However, Alan Du's losses are gradually increasing.

Alan Du's opponent is actually not human! That's because Alan Du played against " Lengpudashi " - an improved version of the poker machine Libratus artificial intelligence , defeated four of the world's best poker players in January. AI - Lengpudashi is located at a supercomputing center near Carnegie Mellon University in Pittsburgh, its strange name translated into English is " cold poker master - cold poker expert ".

1. Libratus - artificial intelligence has just defeated 4 players in poker games

Lengpudashi - an improved version of the Libratus poker artificial intelligence machine Picture 1

Alan Du and a group of 5 members have totaled 36,000 cards in 5 consecutive days of competition. On Monday, April 10 at a conference center in Hainan Island, China, the final score was announced and many of us are no longer startled by the human intelligence system. create AI to win the game.

Poker is a popular game favored by a lot of venture capitalists, because " every time a game is a risky one, you try to count the risk and rate of return (Return On Investment - ROI ) ", Alan Du - an investor and the first Chinese to win the prestigious World Series of Poker tournament held in Las Vegas last year, said. " We have kept ourselves very good in front of this world-class competitor, " Alan Du added.

Lengpudashi - an improved version of the Libratus poker artificial intelligence machine Picture 2

Poker itself is a betting game full of computation and complexity, in which the factor " trick opponents " is an aspect that makes AI AI researchers particularly interested. A normal player can bet, " rogue opponent " ( bluff or raise when their hand is not too beautiful ) or punched ( fold ) without knowing how bad the opponent's card is. This is what makes a poker difficult to guess, unlike chess or Go, when every chess piece and every move is on the table.

Lengpudashi - an improved version of the Libratus poker artificial intelligence machine Picture 3

These balanced poker players are different from the team that tried to beat Libratus in February this year. Alan Du's team consists of engineers, computer science experts and venture capitalists who try to use the knowledge gained about artificial intelligence as well as the investment sector to deal with. every path the machine offers. However, the defeat of the human team showed that each effort was not enough.

During the premiere and demonstration of the most recent artificial intelligence held by Sinovation Ventures and authorities in Hainan Island, China showed that these breakthroughs in the field of AI do not create attraction. strong like the brainstorming game between AlphaGo of Google DeepMind and top player Lee Sedol. Perhaps, we humans are so used to AI artificial intelligence software that can defeat people in these intellectual fields. Recently Google also announced that its DeepMind artificial intelligence software is ready to face in the rematch between people and machines, this time the opponent is the grandmaster Ke Jie from China.

Lengpudashi - an improved version of the Libratus poker artificial intelligence machine Picture 4

Professor Tuomas Sandholm of Carnegie Mellon University in computer science has cultivated Libratus's ability - to bring down four poker players earlier this year in 2004, coaching him how to make decisions when No in-hand full information on the problem. The goal of training AI can play chess, poker is not for people to study more deeply about this game or create an environment for players to practice, but thanks to the controlled training environment This tight, computer decision-making ability will be honed. Later, these skills can be applied in practice, in business, finance, cyber security, .

" Many people often confuse areas where people and machines are capable of doing well. They think that 'fooling their opponents' is a very human behavior but - it turned out not to be. A machine It is possible to learn from the experiences it experiences, it will know that if it goes bad and can deceive the opponent, it will earn more money, "Noam Brown is part of the intellectual development team. Artificial Libratus said.

The AI ​​system does not learn to deceive players from mimicking other veteran players, but learns from the theory of poker games. " Computer tactics are calculated from the rules of the game , not from the analysis of the data included ," said Professor Tuomas Sandholm.

Another venture capitalist and chairman of Sinovation as well as the above AI event organizer, Kai-Fu Lee said that the rapid development of AI in the past 5 years? it can be true if there is no big data analysis technology. His fund invested a total of $ 120 million in artificial intelligence research companies in China.

One thing to mention at this AI exhibition is that advanced AI artificial intelligence systems are becoming much cheaper. Noam Brown said: " The AI ​​power that Lengpudashi has shown can be assembled for less than $ 20,000. This system is surprisingly cheap. Within 5 years, this system will be able to be used to run smartphones ".

Refer to some more articles:

1. Stephen Hawking revealed the AI ​​artificial intelligence solution that dominates people
2. Microsoft Research is creating an AI capable of writing code in seconds
3. Artificial intelligence learns to create another artificial intelligence, replacing people in the future

Having fun!