(ORDO NEWS) — The Ars technica portal reports that the American player Kellin Pelrin was able to defeat the artificial intelligence system in the Go board game.
This came as a complete surprise, since since 2016, computer programs like AlphaGo, developed by Google’s DeepMind division, have been considered invincible.
True, Pelrin did not play with AlphaGo, but with another system that is “considered equal” in strength to the DeepMind program.
Kellyn Pelrin won 14 out of 15 games by exploiting a previously unknown bug discovered by another program, however the player reportedly did without direct computer support during his winning streak.
Pelrin’s tactic was to neatly encircle one of the enemy groups while distracting the AI with moves in other corners of the board.
According to Pelrin, the Go-playing bot didn’t notice the vulnerability, even when the environment was nearly full.
In general, this news is doubtful: firstly, the name of the program with which Pelrin played is not mentioned, but only it is said that it is as strong as AlphaGo.
Secondly, if we analyze the games of AlphaGo against Lee Sedol or Ke Jie (professionals from the world of Go), which are posted on the go4go platform, it is not difficult to notice that the AI is not distracted by the player’s moves in another part of the board and does not leave weak groups that can be easily surrounded, and also plays out fuseki (combinations of the initial stage of the game).
Thirdly, even if Pelrin could “quietly” surround one of the enemy groups, this does not mean that he would be able to “kill” it, or win the game, since the AI is very good at calculating victory points and may well donate to one of your groups.
Therefore, it would be interesting to look at the kifu (recording of games) of these games – however, they were not attached to the news.
Go is a board game that is included in the program of basic disciplines of the World Mind Games. In the classical variant, Go players must sequentially “capture” the majority of the field (goban), drawn into 19 × 19 parallel lines, with the help of white or black stones.
Due to the large number of possible moves and the high level of abstraction, for a long time it was believed that a computer could not defeat a professional Go player, but AlphaGo managed to do this in 2016 and 2017 during matches with Li Sedol or Ke Jie.
—
Online:
Contact us: [email protected]
Our Standards, Terms of Use: Standard Terms And Conditions.