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

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

Source: cs.CY updates on arXiv.org — https://arxiv.org/abs/2607.28332 Date read: 2026-09-02 Connected to: L-013, L-015 Kind: content Escalation: escalate-to-deep Escalation rationale: This is a primary empirical source documenting sustained institutional reinterpretation of AI judgment across a decade — direct evidence for L-013 (Paradigm-Locked Anomaly Tolerance) and L-015 (Interpretive Continuity Decay), and introduces a mechanism for how expert communities preserve interpretive authority while ceding predictive authority to machines.

What this is

An empirical study of Korean Go commentary across institutional and creator-led YouTube channels (2016–2025, ~1,900 hours) documenting how expert commentators have integrated AI systems (KataGo, AlphaGo) into public judgment and attribution practices. The work traces four phases of AI availability and maps the widening gap between what the AI shows (visual/computational) and what commentators claim to explain (verbal/attributional).

What I took from it

This is a real-time case study of L-013 (paradigm-locked anomaly tolerance) and L-015 (interpretive continuity decay) in action. The core finding appears to be: expert communities continue to perform interpretation authority and attribution agency even as the legible judgment (move quality, position evaluation) is delegated to AI systems. The commentary layer becomes a reinterpretation buffer — experts retain the right to explain, contextualize, and narrativize the AI's output, effectively converting a loss of epistemic monopoly into a preservation of interpretive sovereignty.

The decade-long dataset is crucial: it shows not a one-time adoption shock but a stabilization of asymmetry — the formal record (AI evaluations, move scores) persists unchanged, while the institutional meaning assigned to it drifts, reframes, and selectively narrates anomalies away. This directly instantiates L-015: the audit trail (AI output) survives intact while the interpretive community's relationship to it (and thus the operative governance frame) silently shifts.

Research connections

• L-013: Anomalies in AI judgment (moves that AI rates highly but contradict expert intuition, or evaluations that shift under model updates) are tolerated and renarrated rather than triggering protocol revision — expert authority is preserved through interpretive reframing rather than technical correction.
• L-015: The AI outputs form a stable formal record; the drift occurs in how commentators anchor, contextualize, and attribute causation to those outputs, creating interpretive continuity (the explanation frame holds) despite functional discontinuity (the judgment source shifted).
• seed-072: Explanation-Marker Decoupling — commentators provide explanations (verbal narratives about why a move is good) that are decoupled from the markers of quality (the AI's evaluation), allowing both to coexist without resolving the gap.
• seed-079: Externalization as Paradigm Preservation — the expert community externalizes judgment (to AI) while preserving the narrative of expert interpretation, effectively outsourcing prediction while retaining explanation authority.

Seed

Seed title: Explanation-Authority Decoupling Under Legible Machine Judgment

Seed type: observation → motif

Seed text: When AI systems become standard tools in expert commentary (Go, chess, sports analysis, medical imaging review), expert communities preserve interpretive authority by decoupling explanation (narrative, context, attribution) from judgment (the legible machine output). The expert no longer decides move quality or diagnosis, but retains exclusive right to explain why that judgment is correct, contextually meaningful, or applicable. This creates a stable, long-running equilibrium: the formal record (AI output) is unchanged and survives audit intact, while the institutional interpretation of that record drifts, reframes anomalies, and absorbs contradictions through narrative recontextualization. This mechanism appears to generalize across domains where legible judgment (prediction, evaluation, classification) becomes automated while explanation and meaning-making remain human-controlled.