What You Say vs What You Do: Utilizing Positive Emotional Expressions to Relay AI Teammate Intent within Human-AI Teams

Abstract

With the expansive growth of AI's capabilities in recent years, researchers have been tasked with developing and improving human-centered AI collaborations, necessitating the creation of human-AI teams (HATs). However, the differences in communication styles between humans and AI often prevent human teammates from fully understanding the intent and needs of AI teammates. One core difference is that humans naturally leverage a positive emotional tone during communication to convey their confidence or lack thereof to convey doubt in their ability to complete a task. Yet, this communication strategy must be explicitly designed in order for an AI teammate to be human-centered. In this mixed-methods study, 45 participants completed a study examining how human teammates interpret the behaviors of their AI teammates when they express different positive emotions via specific words/phrases. Quantitative results show that, based on corresponding behaviors, AI teammates were able to use displays of emotion to increase trust in AI teammates and the positive mood of the human teammate. Additionally, our qualitative findings indicate that participants preferred their AI teammates to increase the intensity of their displayed emotions to help reduce the perceived risk of their AI teammate's behavior. When taken in sum, these findings describe the relevance of AI teammates expressing intensities of emotion while performing various behavioral decisions as a continued means to provide social support to the wider team and better task performance.