Why health AI interfaces must adapt to user expertise

MIT researchers found that AI explainability tools can yield sharply different results depending on the user. In skin disease diagnosis, non-experts improved their accuracy with AI assistance, while primary care providers performed best when receiving AI predictions without explanations. The study, published in Nature Medicine, highlights how AI interfaces can influence diagnostic accuracy. The findings suggest that while good AI systems can enhance performance in health settings, they must be carefully balanced with algorithmic deference that can lead to errors.