A human player has defeated a highly ranked AI system in the board game of Go.
The amateur player, Kellin Pelrine, beat the AI program KataGo in 14 out of 15 games with help from the research company FAR AI.
To aid Pelrine, FAR AI developed its own computer program that played over a million games with KataGo in order to identify any of the AI's weaknesses.
- The company shared its findings with Pelrine, who used a method involving distraction to defeat the AI in most games.
- KataGo developer Lightvector said it's aware of the exploit, which other players have taken advantage of, and is working to fix it.
- In 2016, DeepMind's AlphaGo AI system beat Go world champion Lee Sedol, who attributed his retirement in 2019 to AI being "an entity that cannot be defeated."
