Every couple of years someone asks this and every couple of years I provide them with the same answer. No specialty is going to get wiped out by AI any time soon, because medicine is much more complicated than every hack in Silicon Valley expects. There's endless variables and asking the right questions in conjunction with a physical exam (and often being able to interpret what is truthful and not truthful of the information given) is far beyond the capability of any AI we will be seeing in our working lifetimes. AI is bad at the one thing human brains are good at- pattern recognition.
Even the most advanced AI struggles to recognize simple concepts that a toddler can comprehend, and the patterns that they recognize can easily be thrown off by minor variables (think harsh shadows under a bridge being interpreted as a solid object by AI in cars, for instance). A child can roughly associate the concept of, for instance, a dog, despite the many differences in breeds and the differences in appearance between a puppy and an older pet. AIs struggle with this, and despite out best efforts still can't comprehend this sort of categorical association, often mistaking foxes, cats, and any manner of other creature with one another based upon animal age and image angle.
The only area where AI may be useful to some degree in its current form is radiology and pathology. These are high liability fields that will always require a human to give a second read due to both the fallability of AI and the function of the physician as a liability sponge for the hospital and AI company. In this way AI may improve care, as it can be good at some very specific focused diagnostic applications that do not involve interviewing or performing a physical on an actual patient. It will not, however, displace physicians within these fields.
And as the poster above noted, there's always the optics. "Hey, you know that thing that can't even solve a basic CAPTCHA? We're going to have it tell you whether you have cancer or not!" That won't play well with public opinion.
