AI in Go commentary: How KataGo reshaped Korean broadcasts
KataGo's silent dominance in Korean Go broadcasts reveals how AI becomes invisible yet inescapable
A decade-long study by Haewoon Kwak on Korean Go commentary reveals how AI systems like KataGo have become silently embedded in professional broadcasts. Analyzing 1,900 hours of footage from 2016-2025, the research tracks four phases of AI adoption across institutional and creator-led channels.
The paper documents a striking asymmetry: AI winrate graphs are visible in 98% of late-period institutional broadcasts, yet explicit AI mentions account for only 2.63% of sentences. This 'communicative signature of domestication' shows how AI's presence recedes from discourse while metrics persist. Creator-led commentary leans further into interface rendering than institutional sources, with the study developing a typology distinguishing source-foregrounding from source-receding mediation. The work was accepted at AIES 2026 and argues that this shift affects how audiences can question machine sources in domains where AI is less reliable than Go.
- KataGo appears in 98% of late-period Korean Go broadcasts but is explicitly mentioned in only 2.63% of sentences
- Creator-led commentary relies more on interface rendering (winrate graphs) than explicit AI attribution
- Study analyzes 1,900 hours of footage across 10 years with four distinct phases of AI adoption
Why It Matters
Shows how AI's silent integration into expert domains can hide its role from audiences, raising transparency concerns