Bjorn-Jostein Singstad

The increasing use of artificial intelligence (AI) in medicine raises fundamental epistemological questions about the nature and status of scientific knowledge. Many contemporary AI systems operate as opaque ``black boxes'', producing highly accurate predictions without offering transparent causal explanations. This challenges established biomedical epistemic frameworks, particularly the evidence hierarchy that privileges randomized controlled trials and systematic reviews as the gold standard of medical knowledge. At the same time, AI systems such as AlphaFold suggest that machine learning can contribute meaningfully to scientific discovery, despite being primarily correlational rather than mechanistic in nature. This paper examines the tension between predictive accuracy and epistemic understanding in AI-driven medicine. Drawing on epistemological distinctions between empiricism, pragmatism, and theory-driven science, it argues that the growing reliance on opaque AI systems complicates traditional accounts of what counts as scientific knowledge. While AI-generated outputs may function effectively as tools for hypothesis generation and downstream experimental validation, their opacity raises concerns about whether such outputs can themselves be considered epistemically authoritative. The paper further considers the implications of large-scale AI systems for the structure of medical knowledge production. It argues that although AI may accelerate discovery, an overreliance on opaque systems risks weakening mechanistic understanding and fragmenting collective scientific knowledge. The discussion suggests that AI should be integrated as a hypothesis-generating tool within existing experimental frameworks, rather than replacing the epistemic foundations of medical science.