Firstly, before anyone builds an AI that can look after people, maybe they can fix the software for looking at Xrays WHICH ALWAYS STOPS WORKING IN THE MIDDLE OF A BUSY CLINIC.
Secondly, I challenge someone here to put a number on how much it would cost to produce an AI that can operate robustly in a clinical environment. This includes the costs of developing such a thing, deploying it as well as testing it. Testing it would arguably be one of the most difficult and complex software testing problems ever. Now whatever that number may be, I hazard to guess that it's a lot. So what you are proposing is to spend a lot of money, so that we can practice medicine... slightly better... MAYBE slightly better.
The greatest improvements in healthcare come from the development of novel therapeutic technologies (antibiotics, vaccines, anaesthesia) or public health measures that change behaviour (sanitation, seat belts, quit smoking campaigns etc).
All of this frankly makes me sceptical of the wisdom of spending zillions on building medical AI, even if such a thing were possible.
I think in some ways we're thinking of automation in the wrong way. That doesn't have to mean having having an AI doctor, with, you know, a lovely bedside manner. Reducing the cost of blood tests so that you can do a sweeping diagnosis of 50 problems at once rather than just 10 is a sort of automation - the doctor no longer has to carefully navigate their way through "there's a chance that it is this, but the blood test is expensive, so we'll try something else first". It's automation to analyse blood tests in combination, and look for profiles for diagnosis. It's automation for a patient to have their own smartphone-connected blood pressure monitor, which will alert the surgery if the read-out is dangerous. It's certainly automation to use software image analysis to pick up the presence of cancer cells in tissue biopsies. It's a step-by-step process, it doesn't have to be all at once.
As other people have commented, augmentation is a very good idea, and automation is part of that. The title of this submission indicates something entirely different to that however.
As an aside, the really interesting part of image analysis is not that we can automatically detect tumor cells, but that machine learning can extract features from pathology specimens that a human could never see:
Firstly, before anyone builds an AI that can look after people, maybe they can fix the software for looking at Xrays WHICH ALWAYS STOPS WORKING IN THE MIDDLE OF A BUSY CLINIC.
Secondly, I challenge someone here to put a number on how much it would cost to produce an AI that can operate robustly in a clinical environment. This includes the costs of developing such a thing, deploying it as well as testing it. Testing it would arguably be one of the most difficult and complex software testing problems ever. Now whatever that number may be, I hazard to guess that it's a lot. So what you are proposing is to spend a lot of money, so that we can practice medicine... slightly better... MAYBE slightly better.
The greatest improvements in healthcare come from the development of novel therapeutic technologies (antibiotics, vaccines, anaesthesia) or public health measures that change behaviour (sanitation, seat belts, quit smoking campaigns etc).
All of this frankly makes me sceptical of the wisdom of spending zillions on building medical AI, even if such a thing were possible.