AI Helps Users Understand Skin Diseases: New Google Research

Google researchers Rory Sayres and Yun Liu published data on how AI tools can help ordinary people in dermatology. In a large quantitative study published in JAMA Dermatology, 2345 participants were shown de-identified skin condition cases. Those who used an AI assistant attempted to name the condition 62% of the time (versus 41% in the control group), and their accuracy was nearly three times higher (23% vs. 8%). In a Wizard of Oz setting (simulated perfect AI), accuracy reached 36%. However, accuracy in choosing next steps (e.g., home remedies or seeing a doctor) increased only slightly. In another study conducted jointly with Stanford University and the Santa Clara Family Health Plan, 110 participants with real skin issues used an AI-powered app in four languages. Using AI improved the ability to name the condition by 260%, although overall accuracy remained low. Physicians noted that AI predictions matched their assessments in 86% of cases, and the app was deemed a useful tool in 92% of cases. The researchers emphasize that to improve these tools, more training images with different skin tones, severity levels, and body parts need to be added, along with more personalized information about next steps.