The natural apotheosis of this argument is a hospital devoid of doctors, run by administrators and accountants behind computers watching the clankers provide care. This is not a criticism of the authors of this piece, or anyone else interested in such things—people should study whatever they like. Leaders of academic medical centers, however, have a choice to make about how they evaluate the efficacy of AI. And that choice will have a tectonic impact on the future of medicine. No doctor I know told me they went into medicine to remove humans from patient care. But the purpose of a system is what it does. If the firepower behind AI removes humans from medicine and biomedical research, that was its purpose. Incidentally, humans are the biggest liability in delivering care. How much would malpractice insurance for a computer cost a health system? Someone is doing this calculation. How will a generation of scientists be inspired to pursue research or doctors be inspired to complete medical training? The answer is not by having a computer do the most interesting parts. I am not a Luddite and believe that AI can enable human doctors and scientists to accomplish remarkable new things. But only if the humans judge outcomes accordingly. Someone is going to image an EHR and let superintelligent agents run wild on it—the ‘CRISPR baby’ moment for AI. In the meantime, the first objective for AI in an academic medical center should be quite pedestrian: reduce administrative burden on faculty and students by transferring these tasks to computers, not to more administrators. This is not a zero-sum proposition: jazzier research into transforming healthcare with AI should of course continue in parallel. Electronic health records and ICD codes were created for billing but are being repurposed for a great revolution in computational medicine. Rather than finding ways to remove doctors and scientists from the process, academic medical centers should direct AI implementation towards: (1) replacement of any faculty activities not related to teaching, research and medical care and (2) a reduction (say, by 90%) in the personnel who are neither faculty nor providers (i.e. doctors, nurses, and other trained healthcare professionals). If these goals are achieved, we may conclude that the use of AI in the administration of academic medical centers has served its purpose. @ZekeEmanuel @JAMA_current @JAMANetwork