Lee Sedol won the fourth game with AlphaGo

Past three games of historic match won by AI

A huge plus for the victory of man: we saw how the recognition of defeat looks in the interface of AlphaGo.

Today was the fourth installment of the match Lee Sedol - AlphaGo. They play the famous 33-year-old owner of the ninth professional dan and the computer go system from Google’s DeepMind division. Today Sedol won.

AlphaGo is a DeepMind product that combines the Monte Carlo method with neural networks of policy and value. The system playing in Seoul is the result of two years of work, including machine learning of neural networks on 160 thousand games from the KGS server and in games against itself. Last October, AlphaGo had already won a strong player . It was a three-time European champion Fan Hui, who lost to artificial intelligence in five out of five games.

It is believed that in Europe the level of go is lower than in the game’s homeland in Asia. Show a clear superiority must match against Lee Sedol. The professional has already lost in three games out of five. Now the outcome of the series is clear, and the fate of the prize million dollars - he will go to charity and go organizations. In the last two games, the depth of the defeat of a team of people will be determined.
')
Yesterday, after the third defeat of a person , one of the topics of the press conference was the question of whether AlphaGo has any weak points. Sedol mentioned how he felt they were. He apologized to the Korean nation and the go community for showing the result.

Easy to understand his stress. Go - a popular game in Korea, according to some estimates, it is played by about 8 million people. In South Korea, news about the match goes to the front pages of newspapers. The severity of what is happening for Google show the figures of observers. The games are visited by the former head of Google, Eric Schmidt, co-founder of the search giant Sergey Brin, and one of the important engineers of the company Jeff Dean. The latter helped Google build the optical channel to the Four Seasons Hotel, where the match is being played. The channel is needed to communicate with the company's data centers, where AlphaGo is launched on hundreds of processor cores and video accelerators.



The man played white, that is, the AI ​​had an advantage in the form of the move first. As in previous games, Cedol has run out of time before his opponent. That is, at some point a person was forced to spend only a minute on the course. In turn, the AlphaGo system is well-disposed of time - this is the result of adding a neural network that helps to manage it.

78 White’s move was excellent. Perhaps Sedol really felt the weak point of the computer system. AlphaGo responded weakly. According to the head of DeepMind Demis Hassabis, on the 79th move the system made a mistake, but understood it only on the 87th move. After that, the AI ​​got confused.


AlphaGo began to make frankly weak and horrible moves. The system acknowledges defeat when the assessment of a chance of victory drops below 20 percent. About this, Google engineer David Silver recalled during the break, which took Lee Cedol. Silver declined to comment on a series of obscure moves of the system. The game continued, and Sedol still had to work in the conditions of lack of time. AI made another mistake, and later admitted defeat. The result of the confrontation for 4.5 hours and 180 moves was the victory of the person. Sedol said that he had never been congratulated so much for just one win.



Five games will be held, and 3-1 has already put an end to the question of the winner. But to assess the strength of the system is important overall result of the match. The last, fifth installment will take place on Tuesday. Like the last four, it will be broadcast on DeepMind on YouTube . Interestingly, in October, Fan Hui also won against AlphaGo. The European champion won twice, but in informal games.

Go is an ancient East Asian puzzle game. Each of the two players need to isolate the territory of the largest possible size with stones of their own color. The researchers of artificial intelligence in go are attracted by the complexity: there are too many positions to sort through them. Computer systems have long dominated both checkers and chess. For example, since 2005, the best of people have been losing the best of chess computer systems. Computer go systems are there, but they play at the amateur level. Before the advent of AlphaGo, experts believed that the game would remain impregnable for another ten years.

Fourth game moves