From Hallucination to Confabulation: A Philosophical Reassessment of AI Error Terminology

Abstract

The term hallucination, which is frequently discussed in the field of artificial intelligence, does not fully correspond to its established usage in psychology. This situation is not merely a terminological problem but also a consequence of implicit philosophical assumptions concerning meaning, mentality, and intentionality. Definitions of AI hallucination vary considerably in the literature, ranging from “plausible-sounding errors” to formulations broad enough to risk encompassing all AI-generated outputs. Some studies argue that these phenomena have been misnamed and that the term would be more appropriately drawn from philosophy than from psychology. Conceptual clarity is therefore essential. Focusing on the relationship between artificial intelligence and psychology, this study aims to shed light on the origins of the naming, definitional, and taxonomic problems surrounding the notion of AI hallucination. To this end, foundational texts in the philosophy of artificial intelligence and the literature on AI hallucination are examined through a comparative and conceptual analysis. The analysis suggests that many researchers employ a language implicitly grounded in dualistic assumptions. At the same time, despite the tensions generated by such assumptions, the influence of psychology remains evident in the selection of concepts used to describe AI errors.