Interpersonal influence matters: Trust contagion and repair in human-human-AI team

As human-AI teams (HATs) become prevalent to enhance team performance, the interaction of multi-human-AI teams have been understudied, particularly how human interactions affect trust in AI teammates. This study investigated whether trust in AI can be contagious from human to human and whether this effect, named trust contagion , can be served as a trust repair strategy in multi-human-AI teams. Using a 2 (AI reliability: high and low, within-participants factor) × 3 (confederate trusting: trusting, neutral, distrusting, between-participants factor) mixed design, participants teamed up with a confederate and an AI teammate in a cooperative trust-based resource allocation game. Self-reported, behavioral, and conversational data were collected. We found that trust is contagious, yet positive and negative trust contagion effects were asymmetrical. While participants teamed with the trusting confederate used more positive words and showed high reliance and self-reported trust in the AI despite its errors, those teamed with the distrusting confederate showed only a significant decrease in reliance. Our results further show positive trust contagion can be used as a trust repair mechanism to mitigate trust drop after trust violations. Additionally, negative trust contagion showed modality-dependent effects, specifically in behavior. Positive trust contagion was advantageous when the AI is unreliable, while negative trust contagion was effective in decreasing reliance when the AI was performing well. Trust contagion was explained through interpersonal trust between participant and confederate mediated by confederate-trusting levels and trust in AI. Our research extends trust beyond dyadic interactions to convey trust is contagious from humans and can repair trust.