When an AI system produces false information, who bears responsibility — the machine or the human who accepts it without verification? This essay argues that the concept of "AI hallucination" is a misattribution: large language models do not hallucinate in any meaningful sense. They generate probabilistic text. The failure occurs when humans treat machine output as verified truth. Drawing on epistemology, cognitive bias research, and real-world case studies, the essay examines seven forms of hallucination-like output, proposes a responsibility framework, and argues that the real crisis is not artificial intelligence but artificial trust.
Halit Cengiz Uzuner (Mon,) studied this question.