AI melanoma-diagnosis systems perform better on white skin than black skin, largely because their training data contains more examples of white skin.
Notes on verification
Confirmed by multiple independent peer-reviewed and institutional sources (JAAD review, Science Advances/Groh et al., arXiv 2025 study, Melanoma Research Alliance-backed research), all converging on the same finding: AI skin-cancer diagnostic tools underperform on darker skin due to underrepresentation in training datasets like ISIC and HAM10000.
Sources
- These new AI benchmarks could help make models less biased (technologyreview.com)
- https://www.curemelanoma.org/blog/making-ai-work-for-people-of-color-diagnosing-melanoma-and-other-skin-cancers (curemelanoma.org)
- https://www.jaad.org/article/S0190-9622(21)02086-7/fulltext (jaad.org)
- https://www.science.org/doi/10.1126/sciadv.abq6147 (science.org)
- https://arxiv.org/pdf/2507.14176 (arxiv.org)