The Generalist-Specialist Paradox of Medical AI

AbstractRapid advances in medical AI tools have enabled software tools to perform similarly to or better than many experts in tasks traditionally performed by medical specialists, such as the interpretation of electroencephalograms by neurologists. Conversely, medical AI tools have shown slower progress in achieving practical equivalence to generalist physicians outside highly controlled environments, for example, when faced with mock board examination questions and clinical vignettes. This divergence is recognized as the generalist-specialist paradox of medical AI and it has multifactorial causes and if it persists, then it will have broad implications for medicine, training, and policy.