Understanding the human-agent interaction in the mixed-initiative environment

Abstract

The rapidly increasing pervasiveness and integration of computers in human society accelerate an intelligent agent's expanding role. Numerous researchers are working on this domain, and the majority of researchers have focused on decision making (cognitive intelligence) the optimum probabilistic result when discussing `intelligence.' Only a limited perspective intelligence exists, and the intelligent agent's behavioral aspect is rarely concerned. As a part of the result, the intelligent agent is often considered as an extension of the human body like a tool. A behavioral intelligent agent is important, which can directly impact user experience especially for the hedonic experience such as in the entertainment domain. This dissertation attempts to understand the `behavioral intelligent' agent, exploring the human-agent interaction in the mixed-initiative environment. Moreover, this dissertation speculates the intelligent agent's role in the hedonic perspective by examing the user experience. For this purpose, a Wizard of Oz based empirical study is carried out. The dissertation speculates the interaction to capture or foresee possible actions taken by the human operator in human-agent interaction. The dissertation explores visitor-initiative, co-initiative, and agent-initiative-based interactive installation. This dissertation captures visitors' emotional changes and the human operator's intentions through various methods (e.g., post-experimental interview, questionnaire) and a modified method (emotional journey map). Based on the empirical study cases, this dissertation defines five basic behavioral archetypes in the mixed-initiative interaction. This dissertation introduces three dominant attributes of the intelligent agent behavior and its behavioral spectrum in the interaction. Last, the dissertation proposes the behavioral intelligent agent's role as an interaction party instead of a tool, especially for entertainment.