I'm building and selling Healthcare AI right now - and these are snippets from what I've seen so far.

1. Diagnosis should be a machine problem. There are way too many data points and factors that are clinically relevant than is reasonable for a doctor to fit into her head.

2. However, to get to machine diagnosis using AI, the data needs to be available in the first place - which means that your entire medical history (correctly and extensively documented in a standard format), all hardware/vendors, and other ancillary systems must be able to communicate with each other. That API-fication of healthcare is largely missing; and given that the vendors of most systems are archaic tech giants, it is unlikely to change very fast.

3. Having said that, AI-based diagnosis can certainly do "screenings" - which are large, public health programs looking for diseases like TB or Diabetic Retinopathy. It can also do mechanical tasks such as preparing reports specific to a blood sample. Note that both of these applications are extremely specific, and take out some tasks, rather than entire jobs.

4. Finally, diagnosis is at best 20% of a doctor's job. She also has to counsel the patient, or do surgeries. While there are advancements, they are not robust enough to replace doctors entirely.

To sum up:
Specifically for the next 10 years, there wouldn't be any significant increase or decrease of healthcare jobs - AI is not yet big enough to eat healthcare. However, as systems slowly settle into the ecosystem, the way doctors work will start changing - they will become more and more reliant on systems, and for the patient, we can hope to achieve better diagnosis, faster turn-around times and overall better experience. The cost for the patient though ... that's an entirely separate question altogether. Will a patient absorb the millions of dollars of cost sunk into building a healthcare AI? What's the implication for the hospital, the patient and the insurance provider?
I love discussing these; please do let me know if you want a follow-up conversation!