Name: Emanuel Rojas
Master’s Thesis Defense Meeting
Date: Wednesday, September 4th, 2024
Time: 12PM EST
Location: JS Coon Building Room 217
Thesis Chair/Advisor: Mengyao Li, Ph.D. (Georgia Tech)
Thesis Committee Members:
Mengyao Li, Ph.D. (Georgia Tech)
Richard Catrambone, Ph.D. (Georgia Tech)
Rick Thomas, Ph.D. (Georgia Tech)
Title: Trust Contagion: Repairing Trust in Human-Human-AI Teaming
Abstract: Human teams are integrating autonomous agents as team members to increase efficiency in human-agent teams (HAT). Autonomous agents are systems, usually powered through AI, with some degree of autonomy that can communicate and collaborate with humans to achieve team mutual goals. Autonomous agents, just like humans, can make errors during the task. These errors decrease people’s trust and potentially lead to failed cooperation. For this reason, repairing trust is essential for people to calibrate their trust and properly rely on the autonomous agent for team cooperation and performance. Past literature mainly emphasized dyadic human-autonomy teams for trust repair, yet real-life contexts often involve multiple humans and automation. In multi-human teams, individuals can consciously or subconsciously influence others’ teammates trust and behaviors when cooperating with autonomous agents, defined as trust contagion. This study investigated the effects of trust contagion as a medium of repairing trust towards the agent. This study was a 2 (agent reliability: high vs. low, within-subjects factor) x 3 (confederate trusting: high, neutral, and low, between-subjects factor) mixed-subject design. A team of three—participant, confederate, and autonomous agent—engaged in a ten-round trust-based cooperative game, involving 42 participants recruited. Trust was measured through subjective ratings, behavioral responses in the game, and in-game conversations. Results supporting our hypotheses indicated trust contagion occurred in the positive direction and served as a mechanism to repair trust. Trust contagion was mediated by the interpersonal trust of the two human teammates. This suggests evaluating teammates’ dispositional trust levels in human-human-AI teams is essential for optimal team composition.