Human player beats go-AI KataGo

A human player has defeated an artificial intelligence in the board game Go. Although he was supported by a computer himself, it now rarely happens that people can win the game against an AI.

The American Kellin Pelrine defeated KataGo bot JBXKata005, an artificial intelligence designed to play the board game go, in 14 out of 15 games. Go is an age-old game in which players try to surround each other on a 19 by 19 square board with black and white stones. In recent years, several artificial intelligences have been set up aimed at winning Go, with the best-known example being AlphaGo from Google. In 2017, it won a match against a human for the first time. Later, a global player quit the game when he felt that the AI ​​would always be better than a human.

Since then, the Go models have only gotten better, but it now appears that they are not completely unbeatable. Pelrine was able to beat KataGo with the help of another AI. The American research company FAR AI had its own program play a million games against KataGo to investigate whether KataGo had known vulnerabilities in the game. This resulted in a strategy that Pelrine could then apply in the game. That strategy consisted of Pelrine placing a large loop around the AI’s groups while distracting the AI ​​by placing stones in the corners of the board.

According to the makers, an AI did not necessarily have to be used to find that vulnerability. That was ‘difficult, but not inhumane’, the makers say. A top-level human player such as Pelrine is therefore not supported by the helping AI during the game itself. Later, Pelrine also managed to beat Leela Zero, an alternative go AI.