Why Is AI Hallucination Asymmetrically Problematized? Structural Conditions of Asymmetric Problematization

Erroneous outputs produced by AI systems are commonly called “hallucinations.” Yet humans also produce outputs that diverge from reference points. Despite this structural similarity, only AI outputs are treated as constituting a distinctive problem category under the name “hallucination.” This paper asks under what structural conditions this problematization is established. What this paper shows is the repositioning of the coordinate of inquiry concerning this problematization. The analysis decomposes the phenomenon into three layers: divergence between output and reference point (first layer), the external inaccessibility of the generative process (second layer), and the interpretive structure on the side of the evaluator (third layer). Through elimination, the paper shows that no structural ground for the evaluative asymmetry is established in the first or second layer. It further shows that intervention in the first layer does not structurally guarantee change in the third layer. The paper’s analysis repositions the coordinate of inquiry concerning the problematization of AI hallucination from inquiry centered on the relation between outputs and reference points to structural analysis that distinguishes the output layer from the evaluative-structure layer. This is the conclusion derived from the identification of the structural locus of problematization on the side of the evaluator, together with the fact that intervention in the first layer does not structurally guarantee change in the third layer.