Human-machine collaboration under pressure: Trust dynamics in ai-guided trauma imaging in emergency departments

The integration of Artificial Intelligence (AI) into Point-of-Care Ultrasound (POCUS) has the potential to revolutionize rapid diagnostic decision-making in Emergency Departments (EDs). However, the socio-psychological dynamics of human-machine collaboration in life-or-death scenarios remain poorly understood. This study investigates how extreme acute stress affects the trust and reliance of ED physicians and paramedics on AI-guided imaging during the Extended Focused Assessment with Sonography for Trauma (eFAST) protocol. Utilizing a mixed-methods approach within high-fidelity simulated multi-organ trauma scenarios, we measured cognitive load, physiological stress markers, and algorithmic adherence among 45 emergency medical professionals. The findings reveal a bimodal distribution of trust under pressure: high cognitive load precipitated either severe "automation bias" (blindly accepting AI diagnostic overlays despite clinical discrepancies) or profound "algorithmic aversion" (completely ignoring the AI to rely solely on manual heuristics). The study highlights that trust in AI is not a static attribute but a highly dynamic cognitive resource depleted by stress. These results emphasize the urgent need for socio-technical redesigns in medical AI interfaces and the implementation of stress-inoculation training to ensure safe human-AI synergy in time-critical medical environments.