AI loses to human being at Go after seven years of victories

A human beat a top-ranked AI system in the board game Go, proving that the rise of machines may not be as imminent as previously believed.

The player was Kellin Pelrine, an American one level below the top amateur ranking. He achieved this victory by taking advantage of a previously unknown weakness that another computer had identified. 

This is according to a report by the Financial Times published on Sunday.

Pelrine won 14 of 15 games but did not do it alone.

The tactics that secured his victory were suggested by a computer program investigating the AI system looking for flaws in judgment. 

“It was surprisingly easy for us to exploit this system,” said Adam Gleave, chief executive of FAR AI, the Californian research firm that designed the program. He added that to find potential weaknesses that humans could use, the software played more than 1 million games against KataGo.

Pelrine also explained that the software’s unearthed winning strategy “is not completely trivial, but it’s not super-difficult” for a human to learn. He claimed he used the method to win against Leela Zero, another top Go system.

An unassailable lead over humans

The human triumph comes seven years after what seemed like an unassailable lead over humans by AI in the game.

In 2019, South Korean Go champion Lee Se-dol retired after being beaten by DeepMind's AlphaGo software. At the time, he told Yonhap news agency that his decision was influenced by the fact that artificial intelligence "cannot be defeated."

Pelrine added that a human would have likely spotted his winning strategy as it consisted of slowly stringing together a large loop of stones to surround one of his opponent’s groups while distracting the AI with moves in other corners of the board.

“As a human, it would be quite easy to spot,” he explained to the Financial Times.

The victory illustrates that AI systems are limited as they can only make moves under specific conditions they have been exposed to. They can not innovate and generalize in a way that comes easily to humans.

The tactic exploited by Pelrine is very rare, indicating the AI systems had not been trained on enough similar games to tackle the situation successfully. The same tactic would not have beaten a human as they would have been able to adapt to an unfamiliar circumstance.