INVESTIGATING EMOTIONAL CONTAGION DYNAMICS IN HUMAN-AI COLLABORATIVE TEAMS THROUGH MODELING AFFECTIVE FEEDBACK LOOPS

Abstract

This study examined the emotional contagion within human-AI team-based collaborators by simulating affective feedback mechanisms using a mixed-method design of experiment. The participants were divided into groups consisting of three people, two humans and one agent of AI with different affective responses abilities. Groups worked on joint problem-solving activities during four weeks in controlled laboratory conditions. The AI agents showed three different patterns of emotional responses, namely, neutral, positive-adaptive, and mirroring conditions. The pre and after task mood assessment, physiological measures, video recordings and sentiment analysis of communications were included in data collection. The findings also indicated that there was a profound emotional transfer in both directions between human team members and AI agents, with the positive-adaptive AI agents producing the most team cohesion and team performance outcomes. Mirroring condition teams were characterized by exaggerated emotional conditions both negative and positive whereas neutral condition teams were characterized by reduced emotional engagement with time. Multilevel modeling proved that AI emotional expressiveness was a major predictor of human affective states, which also determined collaborative efficiency and innovation. The results confirmed that emotional contagion in human-AI teams works in two-way feedback loops, which may be applied in designing emotionally intelligent AI systems to improve team behavior and team performance during organizational situations.