Photo: Shin Jin-seo (right) plays against KataGo. Credit: News1.
In 2016, the artificial intelligence engine AlphaGo, developed by Google Deepmind, caused a sensation by handily defeating Lee Se-dol 이세돌 in go, known as baduk 바둑 in Korean, a game that was considered to be too advanced for an AI to master. In the decade since then, AI go machines have vastly surpassed human players, to a point that they fundamentally changed the way in which professional go players study and play the game. (See previous coverage, “Lee Se-dol Looks Back on AI.”)
Ten years later, world champion go player Shin Jin-seo 신진서 scored a limited redemption for humans. On July 19, Shin defeated KataGo, the world’s leading AI go machine, in the first human victory over AI since Lee scraped a single win against AlphaGo in his 4-1 loss. Shin is playing a best-of-three match against KataGo, and won two games after dropping the first game, winning the series 2-1.
The win comes with a caveat: Shin was given the two-stone advantage at the start, equivalent to spotting the human champion 18 points. Shin was also given a five-hour total clock and 30-second countdown after the expiration of the initially allocated five hours, while KataGo had to play every move within 20 seconds. Nevertheless, even under these conditions, no human has beaten KataGo until Shin did, by 4.5 points.
Since AlphaGo, human go players are measured by their synchronization rate to AI. For pro go players, the average AI synchro rate is 28.5%. Shin holds the record for the highest overall AI synchro rate among humans, averaging 48.9% across 26 matches played from 2019 to 2020. Although the conventional wisdom for the past decade was that humans can only defend against AI, Shin went on an offensive and twisted the established plays to confuse AI.
