Signal When AI Becomes Routine: A Decade of Public AI Mediation in Korean Go Commentary

Summary

A study examined a decade (2016-2025) of Korean Go commentary on YouTube, covering approximately 1,900 hours of footage from institutional broadcasters and creator-led channels, to understand how AI systems such as KataGo, which became standard analytic tools after AlphaGo, are made publicly intelligible in commentary. The research documents a widening asymmetry between visual and verbal AI presence: AI win-rate graphs are visible in about 98% of late-period institutional broadcast time, yet sentences that explicitly reference AI account for only 2.63% of commentary. The study finds that what recedes is not the metric itself but the source label — win-rate and point-gap talk persists even as the word 'AI' goes largely unsaid. The strongest evidence identified is a compositional shift in verbal mediation, where explicit naming gives way to interface rendering, with creator-led commentary leaning further into this shift than institutional commentary. The authors develop a typology distinguishing 'source-foregrounding' from 'source-receding' mediation and note that the stakes of this difference rise in domains where AI is less reliable than in Go. The paper has been accepted at the AIES 2026 conference.