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

AI systems become "invisible" even as they become ubiquitous: In Korean Go commentary, AI-generated winrate graphs appear in approximately 98% of late-period broadcasts, yet commentators explicitly mention "AI" in only 2.63% of sentences. This suggests that as AI tools become routine, they fade from public discussion even while their outputs remain visually prominent.

Visual presence without verbal accountability creates a "domestication" problem: The paper documents a shift from explicitly naming AI sources to simply displaying AI metrics on screen. This compositional change means audiences see AI judgments without hearing explanations of where they come from or how they work, potentially obscuring the machine's role in shaping expert analysis.

Different mediation approaches preserve different opportunities for public contestation: The researchers distinguish between "source-foregrounding" (explicitly naming AI) and "source-receding" (just showing the numbers). Source-foregrounding preserves "discursive anchors" that allow audiences to recognize and question AI judgments, while source-receding approaches remove these hooks for scrutiny.

The stakes of invisibility are higher outside of Go: The authors note that Go is a domain where AI is highly reliable and its judgments are relatively uncontroversial. In other high-stakes domains (healthcare, criminal justice, hiring), this same pattern of AI becoming routine and unnamed could allow flawed or biased machine judgments to circulate without adequate public scrutiny or contestability.