Proposing an Inferential Model for Indebtedness Arising from Agents' Helping Behavior Using Bayesian Model

In the field of social psychology, a number of indebtedness models have been discussed. However, the indebtedness felt by the recipient for the aid received is based on a recipient's subjective evaluation, and the extent to which indebtedness affects the recipient's subsequent behavior has rarely been discussed. In addition, the quantitative relationship between the amount of help given and the corresponding amount of indebtedness has not been clarified. Therefore, it is difficult to directly apply the conventional qualitative indebtedness model to human-agent interactions (HAIs). In the design of interpersonal assistance agents, there is a need to clarify the quantitative index of how much indebtedness the agents' assistance imposes on users. This study developed a model that predicts the amount of help given by an agent to a specific recipient. In the experiment, we imposed a task that is difficult for the participants to accomplish independently, and the agent assisted them to reproduce the helping scene. The participants were then given the opportunity to respond to the agent, and the number of responses was observed. The results were analyzed using Bayesian statistics to estimate the magnitude of the indebtedness felt by the recipient and the amount of return for the assistance given. The results generally supported the conventional indebtedness model and provided a quantitative measure of the reciprocation for the help given. The results of this study are expected to contribute to the design of interpersonal assistance agents as a criterion for considering the magnitude of users' feelings of indebtedness.