A study reported on Aug. 4, 2026, by MIT researchers and colleagues found that AI assistance generally improved the accuracy of non-experts and clinicians in diagnosing skin diseases, but explainability methods affected the two groups differently. The work was described in Nature Medicine.

Non-experts judging whether images of skin moles were cancerous improved with every explainable AI approach tested, largely because they trusted the system. According to the researchers, they trusted large language model explanations whether those were right or wrong, and found vague or generic explanations more convincing. Clinicians, given the harder task of differential diagnosis, were resilient to incorrect AI explanations and performed best with only a model's prediction and no explanation.

The authors said the results point to designing AI with users in mind and to methods that encourage critical thinking rather than overreliance. The research was funded in part by the National Science Foundation, Schmidt Sciences, the National Bureau of Economic Research and Columbia University.