Modeling Forms of Surprise in Artificial Agents: Empirical and Theoretical Study of Surprise Functions

This paper addresses the issue of how to compute the intensity of surprise in an artificial agent. Resolution of this issue is important for the further specification of the computational model of surprise proposed by Macedo and Cardoso (2001) that was implemented in artificial agents "living" in a multi-agent environment. This model of surprise is mainly rooted in the cognitive-psychoevolutionary model of surprise proposed by the research group of the University of Bielefeld (Meyer, Reisenzein, & Schutzwohl, 1997) and in proposals by Ortony and Partridge. We propose several possible functions to compute the intensity of surprise. To assess their accuracy, they were evaluated in an experimental test that focused on the comparison of surprise intensity values generated by artificial agents with ratings by humans under similar circumstances.