[1K4-OS-11a] 人間とAIの共存のあるべき姿を考える

[1K4-OS-11a-02] Proposal of a hypothetical model for the persistent formation of negative impressions toward humanlike agents

〇Yoshimasa Tawatsuji¹, Keiichi Muramatsu¹, Tatsunori Matsui¹ (1. Waseda University)

Keywords:Uncanny Valley, Dynamic System, Temporal Cortex

Extraordinarily humanlike agents can elicit negative feelings from humans, a phenomenon known as ``uncanny valley''. Many researches has been developed for its mechanism in terms of predictive coding and category ambiguity. These, however, have the power to explain short-term responses to detecting anomalies such as whether the perceived entity is human or not, but it lacks explain medium-term and sustained emotion formation of eerie feeling. In this article, we propose a hypothetical model that the uncanny valley phenomenon is a state of convergence to some kind of unstable periodic solution that orbits around an attractor formed as categorical memory in the brain. Based on this model, it can be interpreted that states that do not converge to attractors, which are stable solutions, adversely affect humans, and as a result form a persistent negative impression on agents.