How AlphaGo Inspired Human Go Players to Embrace Creativity

Earlier this year, an amateur Go player achieved a remarkable victory against one of the game's top-ranked AI systems. This win was facilitated by a strategy developed using a program designed to identify weaknesses in systems like KataGo. This victory is part of a larger Go renaissance, whereby human players have become increasingly innovative since AlphaGo's groundbreaking success in 2016.

A recent study published in PNAS by researchers from the City University of Hong Kong and Yale highlights this trend. By analyzing over 5.8 million Go moves made by professionals from 1950 to 2021, the researchers found that human players have become less predictable. They utilized a “superhuman” Go AI to evaluate the quality of individual moves, resulting in a new metric called the “decision quality index” (DQI).

The analysis revealed that prior to 2016, the quality of professional play showed minimal annual improvement, with a median DQI change of just 0.2 in some years—even declining in others. However, since the emergence of superhuman AIs in 2018, median DQI values now change at a rate exceeding 0.7. During this period, professional players have increasingly adopted novel strategies, with 88% of games in 2018 featuring unique combinations of moves, compared to only 63% in 2015.

According to the research team, “Our findings suggest that the development of superhuman AI programs may have prompted human players to break away from traditional strategies and explore novel moves, consequently enhancing their decision-making.” This observation aligns with insights from Professor Stuart Russell of the University of California, Berkeley, who noted that it is expected for players training against machines to adopt strategies favored by those machines.