Hallucinations and the Dreams of AI: How Delightful, Generative AI

Abstract

This paper explores hallucinations generated by AI—not as mere errors, but as expressions of creativity and the unconscious. Based on the hypothesis that the poetic deviations and uncanny associations found in generative AI output share structural similarities with human dreams, we re-examine the meaning of hallucination through perspectives from psychology, literature, and philosophy. Introducing a new interactive narrative format called “tuning fiction,” in which readers can modify the storyline themselves, we envision a future of collaborative storytelling between humans and AI. The paper also discusses the potential applications of this approach in multilingual communication, education, and business, highlighting the strategic role of generative AI in the global dissemination of Japanese-language content. This is both an experimental and practical attempt to reconstruct the contours of the human mind through poetic and philosophical dialogue with AI.