Friction Theory: The Structural Repositioning of Hallucination

The term “hallucination” originates in human cognition and expression. The fact that this term is applied to AI outputs indicates that AI outputs are already being understood within a human interpretive framework. This is not an assumption introduced by this paper; it is a condition that precedes it. However, this understanding is not symmetric. While humans also generate outputs that diverge from their reference points, only AI outputs are problematized as a distinct kind of problem termed “hallucination.” This paper asks: what structural conditions give rise to this asymmetry? This paper decomposes the phenomenon into three analytically distinct components: (1) the relation between an output (T_s) and its reference point, (2) the structural independence of internal judgment formation (R_f) within the output-generating subject from external determination, and (3) the prior internalization of social convergence (T_o) on the side of the evaluator. By examining each component in isolation, the paper shows that (1) and (2) do not vary with the type of generating agent. The asymmetry therefore cannot be located at the level of output accuracy or generative process. Instead, it arises at the level of evaluation, where pre-existing convergence structures shape how outputs are interpreted and judged. On this basis, the paper identifies a structural mislocation in current discussions: the tendency to locate the problem of hallucination in the properties of outputs or their generation, rather than in the evaluative structures that render such outputs problematic. The analysis provides a structural account of why improvements in output accuracy alone do not eliminate the perception of hallucination as a problem. Version 2: Corrected a reference notation in Section 1.4 and Table 1. The citation "Proposition 14" has been corrected to "Definition Section 9" in Hideppa (2026a).