Roxana Daneshjou, MD, PhD, outlines steps to close AI equity gaps in dermatology, from diverse data sets to camera testing for skin of color.
Roxana Daneshjou, MD, PhD, assistant professor of biomedical data science and
Her discussion of these steps continues a conversation with Morayo Adisa, MD, medical director of Dermatology Physicians SC in Chicago and Kenilworth, Illinois, which took place during the
Daneshjou said progress starts with diversifying who is involved in developing artificial intelligence technology, noting that representation is essential to recognizing when data sets or algorithms fail to reflect diverse populations. She credited organizations such as SOCS for prioritizing this work and said expanding diverse data sets remains a key priority. She also pointed to growing reliance on chatbots for medical information, stressing that researchers must identify and publicize flaws in how these tools respond to health questions relevant to patients with skin of color.
Daneshjou referenced her own prior research documenting racially biased chatbot responses to medical questions, noting that public attention after its publication contributed to later improvements from technology companies. She said she would prefer equity testing built into development rather than addressed only after products reach market.
Asked whether any camera systems or algorithms currently outperform others for imaging skin of color, Daneshjou said she was unaware of empirical data supporting specific claims and was unwilling to endorse any product without direct testing. She noted that both Apple and Google have taken steps toward improving how their technology captures diverse skin tones, based on conversations with research teams at each company.
Daneshjou also referenced her past criticism of an early Google dermatology algorithm with limited representation of Fitzpatrick skin types V and VI, saying she remains willing to challenge large tech companies when warranted, while acknowledging both are investing in improvement.
Editor’s note: This episode was summarized with the help of AI tools.
References
