Divergent impacts of explainable AI for dermatological diagnosis on clinicians versus lay people

The study examined how explainable artificial intelligence (XAI) affects diagnostic accuracy in skin disease diagnosis among 623 lay people and 153 primary care physicians. An AI model with balanced performance across different skin tones improved diagnostic accuracy and reduced diagnostic disparities across skin tones in both groups. However, explanations based on large language models (LLMs) had divergent effects: lay users showed higher automation bias, while experienced physicians remained resilient to model errors. Presenting the AI diagnosis before human decision-making may lead to stronger anchoring bias. The results demonstrate that XAI has varying impacts depending on human expertise and the timing of AI prediction presentation.