Thanks for a very interesting read. Since you asked for a comment from a medical perspective, let me give you my 5 cents on automation. I've been a medical doctor in Sweden for 5 years now, and recently started my residency in radiology. Before choosing my current field, I was quite worried I might end up becoming a specialist in a completely automated field of medicine within 5 years. I'm sure if you set IBM's Watson on looking at x-rays, you could get a reasonably high accuracy already now. However the problems with automation in healthcare are manifold.
The biggest problem seems to be that it is done so incredibly inefficiently. You would think using modern software would reduce the need to fill out forms over and over again. Instead it turns out, making things quicker to document, gives you more time to document even more stuff! And because it is implemented in such a bad fashion, we're now filling out forms for several different authorities, that don't communicate with each other. The same data is entered over and over again. Swedish doctors and nurses spend most of their time filling forms and documenting stuff no one cares about any longer. Unless of course you're being reported for some mistake (perceived or real), the mass of text we document is not helping the patients, the medical staff or any researchers.
And yet, Sweden is one of the most advanced IT countries in Europe. This is what gives me most comfort when thinking of automation. Healthcare IT is extremely backwards; I'd say we're using software from the late 90's. Even the server solutions are abysmal, it takes us so long to get journal/medical record texts, and lab reports - simple text and numbers data. And not only is it slow, its insecure as well - I've lost count on how many times the servers are down (sometimes for days!), and healthcare suddenly stops, because we have no way to order new tests or x-rays. Hell, we don't even know whom the patient is without our IT!
Imagine then, an IT revolution in Swedish healthcare, would that make automation possible? I'd argue it would start, but that it'd take at least another 10 years until it starts to become reality. The reason? Human workers are extremely all-round capable resources. Our sensors are so exquisite that we don't even think about how we recognize the data from our finger-tips, our nostrils, vision and hearing, all simultaneously. Place a finger on wrist, you feel the patients pulse, heat, wetness among other things. Sure, there are apparatus that can do the same thing, but they cost money, and they break, and they require a technician for service, a technician to use it, and someone to sterilize it. We're a long way from a cheap and reliable all encompassing solution for these problems. Imagine Honda's ASIMO trying to do any of these tasks - 10+ years seems a reasonable time for technology to mature.
And what if we did have some sort of capable robot, that could help lift patients onto toilets, give patients their pills and take blood tests, would that mean the end of nurses? Again probably not in the near future, because the healthcare "industry" has an infinite demand and finite resources. You're not replacing workers, you're freeing them to do other tasks for other people.
So what about my field, radiology? Surely a learning AI could learn to interpret x-ray images, at least the easy ones? Yes, at least simple images of normal human beings, but things get tricky when you deal with sick people. Often, they are not normal, sometimes far from it. They are elderly, whom have received random injuries and surgery (which sometimes is random as well) that make them different from normal. Patients with disabilities might not be cooperative during examination, a person with dementia might try to run out of the lab while taking x-ray pictures. Or patients with syndromes, mutations which have a large varied effect on human anatomy.
And when AI does get good enough to examine x-rays and other things, I'm 100% sure legislation will be too slow handle growing privacy issues that arise with handling more and more data. There won't be any actual issues of course, I mean we've all taken an oath, but that doesn't mean that legislators will allow us to implement things quickly, even if it is for the good of our patients.
The problems outlined above are of course possible (and some cases easy) to handle, and believe me, I'd rather the IT problems were solved than going to work and not knowing whom I'm treating or examining. But I think Swedish healthcare simply can't afford to jump headlong into automation, simply because it can't even handle IT and other technology from the last 20 years. I'm (probably) safe in radiology for the future of my career anyway.
Let me know if you think I'm wrong, I might have to rethink my career in that case!