Amateur gamer beats “superhuman” AI thanks to AI in board game Go

Artificial intelligence has been considered an unbeatable opponent in the board game Go for several years. Even what was once the best Go player in the world ended his career as a result. But now an amateur player has managed to defeat a “superhuman” AI using all the rules of the art. Another AI, of all things, helped him.

South Korean Lee Sedol was once considered the best Go player in the world. But in March 2016 his career began to falter. Back then, Lee competed for the second time against the artificial intelligence “AlphaGo” from Google subsidiary Deepmind.

After losing four out of five scheduled games, he was shocked. Only about three years later, Lee Sedol announced the end of his career. Reason: The superiority of AI in the board game Go.

Amateur player beats AI in the board game Go

Since then, artificial intelligence has been considered an almost unbeatable opponent in Go. But now the scene draws hope again. Because the amateur player Kellin Pelrine managed to beat two artificial intelligences, Kata-Go and Leela Zero, that are equal to AlphaGO.

But the American didn’t just win. He outclassed the AI ​​kata-go according to all the rules of the art in 14 out of 15 games – even though the artificial intelligence played on the “superhuman” difficulty level. The story looks like a victory from a fairy tale book – especially since the hobby player Pelrine would probably have no chance against a professional like Lee Sedol.

Go is considered one of the most complex board games in the world

But the amateur player prepared extensively before his duel against the AI ​​Kata-Go and had the help of all things from another artificial intelligence. Background: While an AI for a successful chess game, for example, pure computing power is sufficient, the board game Go requires a certain understanding of the game.

The triumph of AlphaGo and Co. in one of the most complex games in the world was therefore all the more astonishing. But for his victory against kata-go, hobby player Pelrine received a crucial tip – from the research company Far AI.

Go: Other artificial intelligence helped Kellin Pelrine win

Together with several US universities, the scientists developed AI software that can detect weaknesses in other AIs. After over a million games against Kata-GO, Far AI hit a blind spot. With the help of a distraction strategy, artificial intelligence could therefore be beaten. Pelrine also used this tactic.

Adam Gleave, head of Far-AI, told the Financial Times about it: “It was surprisingly easy for us to exploit the system” and added: The tactic is “not completely trivial, but not super difficult either”. The strategy is simply to distract the AI ​​in one corner of the board while encircling your opponent’s pieces in another.

Also interesting: