Measuring the spread of trust and distrust through influence in human-autonomy teams

Trust in human-autonomy teams is dynamic and shaped by ongoing interaction, yet most measures rely on retrospective self-report, limiting detection of how trust and distrust propagate in real time. This study introduces and validates influence-an interaction-based metric computed as the average mutual information between joint and team-state time series-to track the spread of trust and distrust. Forty-five participants completed five simulated Remotely Piloted Aircraft System missions in three-member teams, with trust manipulated through communicative and behavioural spreading across congruent, mismatched, and human-versus-agent conditions. Influence predicted subjective team trust and objective performance during active spreading but not in a no-spreading control and showed strong reliability and construct validity. As validation checks, influence recovered known effects: credibility violations damaged trust most, and human sources outweighed agents. Critically, the influence-trust relationship reversed direction across contexts, revealing a context-dependent mechanism. Influence thus offers a real-time, unobtrusive measure linking interaction structure to team outcomes.