Human player defeats AI at the board game Go by finding and exploiting a weakness in the system

In a series of matches held in February 2023 Kellin Pelrine, an American research scientist intern at FAR AI, defeated an AI system at the board game Go. The win by Pelrine was a surprise reversal of the 2016 computer victory by “AlphaGo” against world champion Lee Sedol. AlphaGo was the first computer program to defeat a professional human Go player, the first to defeat a Go world champion, and is (was?) the strongest Go player in history. How was this triumph achieved?

In this game, which originated in China over 3,000 years ago, two players, using either white or black stones, take turns placing their stones on a board. The goal is to surround and capture their opponent’s stones or strategically create spaces of territory. Once all possible moves have been played, both the stones on the board and the empty points are tallied. The highest number wins.

A group of researchers, including Kellin Pelrine, trained their own AI “adversaries” to search for weaknesses in KataGo. They have published a preprint of their research, where they explain the process. The “trick” the algorithm found was to attempt to create a large loop of stones around the AI victim’s stones, but then “distract” the AI by placing pieces in other areas of the board. The computer fails to pick up on the strategy, and loses 97-99% of the time, depending on which version of KataGo is used. The strategy it developed was then used by Kellin Pelrine to beat the computer itself consistently (14 out of 15 matches).

“Our results demonstrate that even superhuman AI systems may harbor surprising failure modes” the researchers say, but they warn that whilst failures in Go AI systems may be entertaining “similar failures in safety-critical systems like automated financial trading or autonomous vehicles could have dire consequences.” Indeed.

Sources

“Human Beats AI In 14 Out Of 15 Go Games By Tricking It Into Serious Blunder”, James Felton, ISLScience, February 21, 2023

“Man beats machine at Go in human victory over AI”, Richard Waters, from Financial Times in Ars Technica, February 19, 2023

“Adversarial Policies Beat Professional-Level Go AIs“, Kellin Pelrine, Tony Tong Wang, Adam Gleave, Nora Belrose, Tom Tseng, Joseph Miller, Michael D Dennis, Yawen Duan, Viktor Pogrebniak, Sergey Levine, Stuart Russell, 2022/11/

“Adversarial Policies Beat Superhuman Go AIs”

Photo credit

Shutterstock