Validating Three Properties of Trust Dynamics in Pharmacist-AI Interaction With Professional Pharmacists (Extended Abstract)

This study examined whether established properties of trust dynamics: negativity bias, stabilization, and continuity, generalize from controlled laboratory settings to real-world professional contexts. Thirty licensed pharmacists completed an AI-assisted medication verification task, during which they evaluated prescription accuracy with support from an AI system. Trust was measured using a trial-by-trial scale ranging from 0 to 100. Results confirmed negativity bias, with pharmacists showing significantly larger decreases in trust following incorrect AI recommendations than increases after correct ones. Evidence for stabilization was partial. Trust adjustments following incorrect AI outputs diminished over time, but no similar reduction occurred for correct outputs. The continuity property was supported, as trust ratings were strongly influenced by prior trust levels and closely tracked recent AI performance, with temporal autocorrelation decreasing over longer intervals. Overall, the findings suggest that fundamental properties of trust dynamics generalize beyond student samples and abstract tasks, supporting the ecological validity.