| Recent decades have seen increased interest on the role of emotional expression in human-computer interaction. However, despite a growing number of empirical findings reported in the literature, the results are still inconclusive about the mechanism for the social effects of emotion. This dissertation studies how emotion displays in computers impact people's decision-making and proposes a mechanism for such effects based on appraisal theories of emotion. In appraisal theories, emotion displays arise from cognitive appraisal of events with respect to one's goals (e.g., is this event congruent with my goals? Who is responsible for this event?). According to the pattern of appraisals that occurs, different emotions are experienced and displayed. Since displays reflect the agent's intentions through the appraisal process, we argue people infer, from emotion displays, how computer agents are appraising the ongoing interaction and, from this information, make inferences about the agents' intentions. We refer to this theory as reverse appraisal. To support it, several empirical studies are presented where participants engage in the iterated prisoner's dilemma with virtual agents that, though following the same strategy to choose their actions, display emotions in the face that are consistent with either cooperative or competitive goals. The results confirm that emotion displays impact people's decision making and people cooperate more with cooperative agents. Consistent with reverse appraisal's prediction that what is critical for the effects is not the emotion but the underlying appraisals, the results also show that the same display (e.g., a smile) can lead to opposite effects on cooperation depending on the context in which it is shown. A final study shows that people can successfully retrieve, from emotion displays, information about how agents are appraising the ongoing interaction; moreover, these perceptions of appraisal are shown to mediate the effect of emotion displays on perceptions of the agents' likelihood of cooperation. Having established the theoretical foundations of reverse appraisal, the dissertation presents several computational models of decision-making in the prisoner's dilemma. These models were developed by applying statistical and machine learning techniques on the data collected in the studies. Experimental results show that, as expected, computer models can better replicate human behavior in the original studies if they take into consideration the counterpart's appraisals and emotion displays. In addition, the dissertation also contributes a platform for creating realistic embodied agents that can display emotion. This platform was used in all studies to implement the experimental manipulations and can be used as a research tool to study human-agent and human-human interaction. Finally, the dissertation discusses the implications of reverse appraisal and the reported social effects of emotion for human-computer interaction, artificial intelligence, decision and emotion theory. |
