Quantifying Unperceived Shifts in Aesthetic Evaluation under Human-AI Interaction: A Testbed for AI-Induced Value Change

Even if AI causes neither extinction nor suffering, it may still erode what makes life worth living. This risk to ikigai is hard to study directly: it resists quantification, unfolds over decades, and raises ethical problems if manipulated. We propose aesthetic value as a tractable testbed, and present an experimental paradigm for measuring unperceived shifts in it. Ratings from the participant and from the AI are placed in a common space, so that how much a person's evaluations shifted through AI interaction, how far they moved toward the AI's bias, and how much of the shift the person perceived can all be computed on the same scale. The form of AI interaction can be varied independently, allowing systematic study of which kinds of AI use shift evaluations without the person perceiving it.