The Epistemology of AI-driven Science: The Case of AlphaFold

The paper examines the epistemological implications of AlphaFold, an AI system that predicts protein structures, challenging traditional notions of scientific knowledge by showing that its opaque yet trustworthy predictions can function as scientific knowledge when integrated into scientific practice. It argues that scientific knowledge can be strongly opaque to humans, provided it meets functional and social criteria within the collective scientific enterprise.