Why is machine learning not more widely used for medical diagnosis?

In my understanding, diagnosis is a relatively straightforward machine learning problem. Technology exists for coding both symptoms and conditions, and there are large datasets of training data available. The literature on the topic is pretty thorough and some advanced algorithms have already been tested (one common approach is to use Bayesian networks).Why, then, does clinical diagnosis still rely mostly on doctors' expertise and intuition? There are three possible reasons I can think of, but I have no medical experience and cannot come to a conclusion.

Technological: No one has yet created an algorithm that is as effective as doctors are. Most of the literature I've seen contradicts this.Legal: There could be regulations or issues of liability that prevent hospitals from relying too heavily on computers for diagnosis.Cultural: Doctors don't trust computational diagnosis, and resent the idea that computers could replace them.I'm wrong: Machine learning tools are often used in clinical diagnosis.

‍