Human Go Player Defeated Completely by AI... First Incident in 7 Years Since Lee Sedol

Approval Granted for Exploiting AI Vulnerabilities
Expert Says "Humans Overestimate AI"

[Asia Economy Reporter Choi Seung-woo] An amateur Go player has won a match against artificial intelligence (AI). This is the first time a human has defeated AI since 9-dan Lee Sedol.

On the 18th (local time), the Financial Times reported that Kellin Perlin, ranked 2nd among American amateurs, recorded 14 wins out of 15 matches against the AI Kata GO.

This victory marks the first time a human Go player has beaten AI since 2016, when 9-dan Lee Sedol won one game against Alpha GO, developed by Google's DeepMind. Kata GO, which Perlin faced this time, is known to use the same system as Alpha GO.

However, Perlin's victory was largely credited to an "AI that revealed the weaknesses of AI." Perlin exploited Kata GO's vulnerabilities based on tactics taught by a program from FAR AI, an AI startup in California. Adam Gliebe, CEO of FAR AI, explained, "To identify AI's blind spots, we had AI play over one million games against each other," adding, "Finding Kata GO's weaknesses was surprisingly easy."

Human Go Player Defeated Completely by AI... First Incident in 7 Years Since Lee Sedol

[Image source=Yonhap News]

Regarding this, Perlin stated, "Learning the tactics suggested by AI was not very difficult, but it was not very easy either. It is a level that intermediate amateurs can learn." He also added, "I placed stones large and slowly to distract the AI," and "The AI failed to recognize dangers that a human would easily notice."

On this matter, Stuart Russell, a computer science professor at UC Berkeley, explained, "The flaws observed in some AI Go players may suggest fundamental defects in AI's deep learning systems." Professor Russell of the Computer Science Department at the University of California, Berkeley, said, "AI only understands special situations from past data. Generalization, which humans can easily do, is impossible for AI." He further added, "This victory is a case that reminds us once again that humans are prematurely overestimating AI."