Mistake patterns among human Go players: Insights from AI

Carlos G Urzúa-Traslaviña, Quentin Rendu /May 19, 2025

Abstract
      Artificial Intelligence (AI) is now routinely used by Go players to review their games. Analyzing individual mistakes helps players identify their weaknesses. However, deriving generalizable insights requires a broader analysis of mistake patterns.

In this study, 100,682 AI-scored amateur and professional Go games are studied to investigate mistake patterns. Three different ranks are examined, ranging from low-level amateurs to top professionals. Various move features such as height, distance to previous move, and adjacent stones are analyzed to gain a deeper understanding of these mistake patterns.

The key findings are as follows: (1) a noticeable improvement in opening performance among professional players since 2017; (2) a significant performance gap in the endgame between professionals (who exhibit near-optimal play) and amateurs; and (3) areas for improvement in tactical skills among amateurs, particularly in first-line and sacrificial moves.

Keywords: Go, Baduk, Weiqi, AI, Katago, Mistake distribution