How Generative AI Engages with Tacit Knowledge: A theoretical examination through mediation, structure, and representation

Abstract

This paper theoretically examines the extent to which generative AI handles tacit knowledge by categorizing it into three layers: mediation (process of formalization), structure (social and cultural conditions), and representation (methodological observation). The authors argue that while current models like LLMs and multimodal AI can remarkably reproduce the external patterns of human judgment and action, they primarily capture "outer forms" while disregarding the internal processes and social contexts that allow tacit knowledge to function. Specifically, generative AI lacks the capacity to handle the subjective mediation of meaning, the communal norms of the "structure" layer, and the analytical observer-perspective found in the "representation" layer. Ultimately, the study suggests that identifying these "discarded" dimensions is crucial for designing future AI-human collaborations, emphasizing that qualitative approaches like ethnography remain essential for clarifying the social conditions that AI cannot inherently grasp.