health π§ Health AI tools need to match user knowledge
MIT researchers found that AI explainability tools yield different results based on user expertise. When testing skin disease diagnosis, non-experts improved accuracy by leaning on the AI model's suggestions. Primary care providers performed best when they just received an AI prediction without any accompanying explanation. These findings stress that interface design must carefully balance AI assistance with automation bias risks. The study suggests that explanations suited for patients might mislead novices more than they help clinicians.