If a doctor and an AI app both have 95% accuracy, what’s the difference?
Dr. Bruce Darrow, interim chief digital and information officer and chief medical information officer at Mount Sinai Health System in New York
Hospitals, health systems and other healthcare organizations are flocking to generative AI, standing up a wide array of projects and use cases to see where the emerging technology can best help them find clinical and financial ROI.
Most are thankfully being cautious to make sure not to go too far, too fast.
But as AI technology gets ever more powerful, it's worth asking a common question: Could artificial intelligence ever replace doctors?
The answer to this question almost always is no. AI will help doctors, not replace them, say most experts.
Or, as Mayo Clinic platform President Dr. John Halamka has said, "If your doctor could be replaced by AI, your doctor should be replaced by AI."
But Dr. Bruce Darrow, interim chief digital and information officer and chief medical information officer at Mount Sinai Health System in New York, takes a more nuanced view.
In some cases, he says, where the clinical accuracy of doctors and AI are nearly the same, some clinical care in the future could indeed migrate over to AI.
In this ninth installment of our Top Voices in AI series, Darrow has plenty to say about AI overall (today's part one) and at his own health system (tomorrow's part two).
Q. Why do you think artificial intelligence is having such a big moment in healthcare today?
A. A lot of it has to do with this "A-ha" moment that happened when ChatGPT was first made available to the general public. AI has been around for a while, especially the predictive models. But the generative models really caught people's attention.
Especially when they could use it not in the setting of, "Well, I'm going to the doctor, and the doctor is going to have some Ninja Logic behind the scenes." But they could use it at home, and they could say, "Tell me about my cholesterol medication."
It would spit out all of this information about cholesterol medication. It would summarize things for them, and it would really create information in a way that was really tangible to a lot of people. I think since then, we've had a lot of interest in AI.
Also, in my experience provider organizations are always looking for something that makes them attractive to patients. So, saying something along the lines of, "We use AI to make sure you get the best possible care, that we make the most accurate diagnosis," becomes potentially a differentiator for an organization.
Q. When AI and healthcare come to mind, some people think AI will replace doctors. Will that ever happen? And why or why not?
A. When I talk to people about artificial intelligence, I've heard some suggest we should instead talk about augmented intelligence because a lot of the AI use cases these days are not to replace the doctor, but specifically to make the doctor more successful, more accurate, faster, to make their job easier.
In many cases, it's not a replacement. But I think the question of replacing the doctor, in some cases, may be just a matter of how far in advance you're looking. Today, most AI applications are not to replace the doctor, but to make the doctor better at his or her job.
For example, if you have a radiologist who uses AI tools to do the same job they always do, to look at radiology images and point out areas of disease or lack of disease, to be able to make a diagnosis, that job still exists.
But the AI may make that job easier, faster, may help them prioritize out of a list of 50 different head CTs to read, which one to read first because there's the greatest impact potentially for a time-sensitive diagnosis. These are things that don't replace the doctor.
Having said that, over time, I could potentially see ways in which a lot of the functions of doctors could potentially be replaced by AI, especially in areas of pattern recognition. Let's say you had a rash on your hand, and I gave you two options.
I say, "I can make you an appointment with a dermatologist." You'll probably need to wait five to 10 days to get that appointment. By the time you get there, maybe your rash is better, maybe it's worse, maybe it's gone. But the doctor will look at it and they will probably give you a correct diagnosis and treatment plan with, say, 95% accuracy.
Let's say I say, instead of waiting those five to 10 days to see the dermatologist, I can give you an app that you could download on your phone and you take a picture of that rash on your hand, and it would give you information with a certain amount of accuracy.
Then you could potentially treat that with over-the-counter hydrocortisone, or you could leave it alone and it would resolve on its own, or where you would know this is something that really would benefit from medical attention and the intervention of a doctor.
If I told you the accuracy of that was on par with what a doctor would tell you in five to 10 days, say 95% or even better accuracy, it would be no contest. Every patient would choose the app. Even if you had accuracy that was approaching what a doctor would be able to, maybe even 85% or 90% accuracy, there may be an instinct to go there.