Algorithmic second opinions in primary care dermatology: Trust dynamics and automation bias across skin lesions and eruptions

Dermatological complaints constitute a significant portion of primary care consultations, encompassing a broad spectrum from infectious exanthems and inflammatory rashes to malignant neoplasms. The recent integration of Artificial Intelligence (AI) into primary care settings promises to bridge the diagnostic accuracy gap between General Practitioners (GPs) and board-certified dermatologists. However, the socio-technical dynamics of how GPs calibrate trust in AI diagnostic suggestions remain poorly understood. This study investigates human-computer interaction, focusing on diagnostic confidence and automation bias, when physicians are provided with an AI-generated "second opinion." Utilizing an explanatory mixed-methods design, 50 clinicians—comprising 25 GPs (Novices in specialized dermatology) and 25 Dermatologists (Experts)—evaluated a curated dataset of 40 diverse macroscopic and dermoscopic clinical cases, including viral rashes, fungal infections, inflammatory dermatoses, and skin tumors. To assess automation bias, the AI assistant was programmed to provide deliberately incorrect diagnostic prompts in 20% of the cases. Quantitative results demonstrate a profound experience-based divergence. GPs, operating under higher diagnostic uncertainty, exhibited severe automation bias, frequently overturning correct initial diagnoses to align with erroneous AI prompts. Conversely, dermatologists demonstrated resilient diagnostic confidence but exhibited instances of algorithmic aversion, often rejecting the AI entirely due to its "black box" nature. The findings emphasize that deploying AI in primary care requires shifting focus from algorithmic accuracy to human-centered interface design, specifically integrating Explainable AI (XAI) paradigms to ensure safe and synergistic clinical decision-making.