March 9, 2016 at 13:22

AlphaGo wins Lee Sedol's first game

    Google AI turned out to be stronger than one of the best players in go


    Artificial intelligence can demonstrate its superiority over humans in the form of a win in a logical game. Such games are divided into those in which AI already plays stronger than the best of people, and those that have not yet succumbed. Today, as a result of the first of five games, the Asian go game is one step closer to falling into the first category.

    The British artificial intelligence company DeepMind was acquired by Google in 2014. DeepMind has already attracted media attention with the development of the DQN systemwho independently learned to play 49 old Atari games. For the past two years, a team of company employees has been working on AlphaGo. This is an artificial intelligence system for the game of go, which shows excellent results. She achieves this with a combination of the Monte Carlo method and the neural networks of politics and value. The neural networks of the program were taught on 160 thousand games from the KGS server.

    Go rules may vary, but their basis is about the same. Two players place stones of two colors on a board of a certain size, the standard field is 19 × 19 lines. The goal of the game is simple: you need to fence off a territory larger than the opponent’s size on the board with stones of your color.

    Go is considered a difficult game for computer systems. This is due to the huge search space and the difficulty of choosing moves. The possible positions of stones on a standard board are more than a googol (10 100 ) times more than in chess. The number of possible positions is greater than the atoms in the universe. Existing programs can play go at the amateur level. For a while, it was believed that before the advent of systems, a decade is stronger than a person.

    AlphaGo has demonstrated that this is not so. The program easily beats not only other products, but also people. Three-time European champion Fan Hui played against AI. The result of five games was striking: in all of them, the player of the second professional dana lost to the computer system . Google’s choice was later criticized: Go's level of ownership in Europe is not very high, and the mistakes made by a representative of a team of people raised questions. But Google immediately announced that AlphaGo will play against the legendary Korean go-pro Lee Sedol - the best go player in the last decade, the second in the number of titles. Obviously, his level of play should be higher.


    Today in Seoul was the first of five games. She attracted attention, which is not surprising for Korea, where tens of millions are playing go. The game was attended by two members of the Korean parliament and former head of Google Eric Schmidt. Opposite Lee Sedola is AlphaGo operator Aja Huan. By the way, the researchers did not sit still after the match against Fan Hui in October last year and improved AlphaGo. For the past five months, AI has studied in parties against itself.


    Man played black and walked first, AI - white. The beginning played was not very unexpected, it even looked a bit like Fan played in Britain. The gray-haired man showed his typical style and made aggressive moves. But AlphaGo did not hold back - the system began to advance in about 12 moves. According to commentators, AlphaGo plays like a human: the system searches for and creates vulnerable groups of stones to make strong moves. Although the AI ​​made a few mistakes, Black had serious reasons for concern. The game of 186 moves for 3.5 hours ended with the ninth dan master admitting defeat.


    It's too early to talk about the historical moment. The last fifth game will take place on March 15th. Then the fate of the prize million dollars will be determined - either he will go to charity, or he will be received by a human player. Next Tuesday, it will become clear whether AI has conquered a new game or is too early to talk about it. The gray-haired man is full of determination: he believes that the opening moves will make it possible to achieve greater chances of victory.

    All games are broadcast on YouTube .
    Tags:

    Also popular now: