[1J1-OS-21] 信頼されるAIのためのインタラクションデザイン

[1J1-OS-21-02] Estimation of the Internal State of Others from Behavioral Observations Toward Active Explanation

〇Yuya Hishiki¹, Tatsuya Sakai¹, Takato Horii¹, Takayuki Nagai^1,2 (1. Univ. of Osaka, 2. Univ. of Electro-Communications)

Keywords:Estimate Internal State, Bayesian Inference , Explainability, Explainable Autonomous Agents

It is important for autonomous robots to have explanatory ability in terms of reliability.In this paper, we first consider what kind of behavior of others makes us feel the need for explanations through simulation experiments.Then, we propose a method by which the autonomous robot can actively determine the need for explanations without the user explicitly requesting it.Through simulation experiments in a grid environment, we confirmed that the results of the proposed method and the results of human judgments of the necessity of explanations are similar.This suggests that the proposed method is capable of autonomously determining the necessity of explanations.