Tech communities hold a common opinion regarding ML and medicine: "bring machines to the clinics! Research shows them to be superior". Everyone always takes great care in forgetting that the ML research for medicine and diagnosis assesses those systems in a well-defined setting. I.e, you have to feed the system the appropriate info for it to work. From personal experience, I can say that obtaining this info can be exceptionally difficult for a multitude of reasons, and trying to fully automate that is currently not economically doable. The medical system currently does not have the information-retrieval capability to make ML really useful.

I am sorry to say that anyone thinking that we could "ML all the things" in a hospital or office practice on this here day clearly has no idea what a mess our hospitals really are. Some things are amenable to ML though, and most of us welcome any help we can get. Even from machines.