본문 바로가기
  • Home

Hallucinatory ‘AI Hallucination’

Hyeongjoo Kim 1

1중앙대학교

Accredited

ABSTRACT

This study aims to reveal that the phenomenon we call "AI hallucination" is nothing more than just a hallucination. Moreover, It argue that the epistemological concept of hallucination is a fiction if we re-examine it by expanding the scope of thought to the world of common sense, thereby it will be strongly supported again the fact that the "AI hallucination" is just a hallucination. Even if the discussion becomes philosophically deeper as the research progresses, I will always maintain the perspective of our everyday world, that is, the perspective of common sense, in which the term 'AI hallucination' is used very broadly, and return to this perspective to continuously adjust the level of discussion. From this perspective, in this study I will analyze first the everyday meaning of "AI hallucination" and then examines the meaning of "AI hallucination" defined by researchers. then connect the two from an integrated perspective. As a result, it will be revealed that "AI hallucination" is not a term that is quantitatively differentiated and qualitatively defined in a single way. Secondly, I will attempt to infer the core theses of "hallucination philosophy" based on the arguments of naive realism and declarativeism, which have systematically explored "hallucination". I will then compare these theses with other philosophical positions. As a result, it will be inferred that philosophical hallucinations are conceptually impossible. Finally, I will argue that the various explanations of AI hallucinations will inevitably collapse conceptually and ultimately be nullified if the second result is applied again to the first.

Citation status

* References for papers published after 2023 are currently being built.