Promoting transparency in AI for biomedical and behavioral research

Abstract

Recent advancements in artificial intelligence (AI) in healthcare have highlighted the need for transparency, including explainability, interpretability, and accountability across the AI lifecycle^1,2. Transparency ensures stakeholders can make informed decisions about data and model reuse, fostering trust and fairness while aligning with regulatory frameworks. However, the concept of transparency lacks a clear definition for both biomedical research and clinical care, resulting in inconsistent practices.

This Correspondence focuses on transparency within the realm of AI-driven biomedical and behavioral research. Although the effect of AI on clinical care is crucial, this discussion centers on its implications for research, addressing gaps in data reuse, model generalization and fairness. The National Institutes of Health (NIH) Office of Data Science Strategy (ODSS) convened a workshop that brought together leading experts in AI, healthcare and ethics to examine transparency in this context³. The workshop findings highlight practical solutions tailored to research contexts, addressing documentation standards, patient and community co-design, and oversight mechanisms to achieve equitable outcomes.