[2D5-GS-9] Human interface:

[2D5-GS-9-04] Bayesian Model of the Uncanny Valley Phenomenon in Humanoid Robots

Impact of Predictive Uncertainty about Robots on Affinity

〇Shimon Honda¹, Rin Shibano¹, Hideyoshi Yanagisawa¹ (1. The University of Tokyo)

Keywords:uncanny valley, Bayesian model, uncertainty

This study investigated how prediction uncertainty affects feelings of affinity for humanoid robots by mathematically modeling the uncanny valley phenomenon. The uncanny valley describes the anxiety caused by artificial entities that are highly human-like but not fully human, with categorization ambiguity proposed as a key cause. We modeled affinity as negative surprise, grounded in the free energy principle, where the brain minimizes prediction error. In the experiment, participants rated the familiarity of morphed images, with prediction uncertainty manipulated through varying levels of blurring in pre-stimuli. Results showed increased familiarity under low uncertainty conditions, partially aligning with the model's predictions. These findings suggest that enhancing robot recognition may reduce uncanniness. This study advances quantitative understanding of the uncanny valley and offers design guidelines for improving affinity. Future research should extend the model to other sensory modalities.