Artificial intelligence and global health equity

Artificial intelligence (AI) could help achieve global health equity by extending efficient and cost effective healthcare to historically underserved patients.123 However, biases ingrained in AI tools can reduce equity and cause harm unless their creation, deployment, and assessment take account of existing health inequities and are tailored to the settings and populations in which they are being used.45678

AI algorithms are trained using large datasets. Unconscious judgments of researchers and AI experts about which issues to work on, which populations to study, and which training datasets to use can perpetuate prejudice.9 Large language models (LLMs), artificial neural networks that encode knowledge, are trained on clinical content that is rife with societal prejudices. This plays out in, for example, LLMs offering differential diagnoses and treatment recommendations that stereotype certain demographic groups, furthering health inequities.10