Predicting Trust Contagion via Gaze Allocation in Multi-Human-AI Teams

Abstract

In multi-human–AI teams, trust in AI is shaped not only by direct interaction but also by social influence among teammates, a process known as trust contagion. While social signals facilitate this process, prior research have overlooked specific mechanisms such as gaze. This study examined whether gaze allocation, attention dispersion across teammates, task interfaces, and the environment, serves as a behavioral indicator of trust contagion. Results showed that gaze allocation varied with both AI reliability and a confederate’s trust level. When AI performed poorly, participants paired with a high-trusting confederate showed more concentrated gaze, while those paired with a low-trusting confederate displayed greater dispersion, reflecting more frequent scanning for information and verification. Gaze predicted AI trust non-linearly, with moderate gaze associated with higher trust, suggesting both overly narrow and scattered gaze limit effective information processing. These findings highlight gaze allocation as a key nonverbal mechanism through which trust contagion unfolds.