Eerie edibles: Realism and food neophobia predict an uncanny valley in AI-generated food images

Abstract

This study investigates whether imperfect AI-generated food images evoke an uncanny valley effect, making them appear uncannier than either unrealistic or realistic food images. It further explores whether this effect is a nonlinear function of realism. Underlying mechanisms are examined, including food disgust and food neophobia. The study also compares reactions to moldy and rotten food with reactions to AI-generated food. Individual differences in food disgust and food neophobia are treated as moderators of food uncanniness. The results show that a cubic function of realism best predicts uncanniness, with imperfect AI-generated food rated significantly more uncanny and less pleasant than unrealistic and realistic food. Pleasantness followed a quadratic function of realism. Food neophobia significantly moderated the uncanny valley effect, while food disgust sensitivity did not. The findings indicate deviations from expected realism elicit discomfort, driven by novelty aversion rather than contamination-related disgust.