I remember former fellow students telling how they corrected medical students' practical physics internship assignments. Apparently the statistical rigour is much lower. On the other hand, I guess in pure natural sciences you work with much more indirection and have zero results nowadays unless you use a lot of tools. Probably there aren't that many medical conditions that can be treated in-depth based on first principles but I think concluding that whole medical science can be replaced by an AI assistant might be a bit far-fetched. Compassion is probably under-estimated indeed. Some people with serious conditions prefer to be treated at objectively less sophisticated yet less industrialized facilities.