AI outperforming physicians. Ban humans from practicing certain medicine?

[Anonymous]

Anonymous wrote:

Anonymous wrote:

Anonymous wrote:Also, radiology. Why isn't that a field that will get absolutely decimated by AI. They will just take images and have them interpreted by AI that can use image analysis and machine vision that is going to be less error probe and less biased than a radiologist. No need to pay an army of radiologists $500k salaries anymore when AI can do all of the work in 1/10th the time, with less errors, and for a fraction of the cost.

I would not sign up to have an ERCP performed by AI. You wouldn’t either.

People won't have a choice soon when insurance no longer covers physician fees when AI will cost $20 with less error...

I don’t think you know what an ERCP is. AI cannot even reliably drive cars. It definitely cannot do this radiological procedure in its present state.

[Anonymous]

Anonymous wrote:

Anonymous wrote:

Anonymous wrote:

Anonymous wrote:Also, radiology. Why isn't that a field that will get absolutely decimated by AI. They will just take images and have them interpreted by AI that can use image analysis and machine vision that is going to be less error probe and less biased than a radiologist. No need to pay an army of radiologists $500k salaries anymore when AI can do all of the work in 1/10th the time, with less errors, and for a fraction of the cost.

I would not sign up to have an ERCP performed by AI. You wouldn’t either.

People won't have a choice soon when insurance no longer covers physician fees when AI will cost $20 with less error...

I don’t think you know what an ERCP is. AI cannot even reliably drive cars. It definitely cannot do this radiological procedure in its present state.

And how fast did the smartphone evolve again?

[Anonymous]

Anonymous wrote:I already use an app that identifies skin conditions using AI (family prone to them).

If the condition didn't clear up quickly, I would go to the doctor and also tell them what the app suggested. I would say maybe the app was right about 80% of the time. Maybe it will get to 95% in the next decade. But there will always need to be humans validating the technology, which can screw up royally on occasion.

And humans don't screw up royally in health?

I think the point is that there is going to be a turning point where the error rates of AI in medicine are going to be lower than those for a human. At that point, why would it make sense ever to go to human doctor? AI is already being used across the globe when you enter and there are no more staff at customs because the entire thing is using facial recognition. AI imagine analysis and machine vision are far better than people think and are evolving more rapidly than most realize. It only matter of time until the bulk of diagnostics is performed by AI and that all your insurance will pay for.

[Anonymous]

Anonymous wrote:I don’t have a link but a couple months ago they had an interview on NPR (A1 maybe?) on this issue. The result was that AI was better for routine issues but that human physicians were much better on detecting cases that fell outside the routine.

This is ALREADY one of my biggest frustrations with human physicians - they are trained to think "when you hear hoofbeats, think horses, not zebras," and will make patients suffer catastrophic issues to prove it's not horses. So anything that reinforces that bias further could be dangerous.

[Anonymous]

Anonymous wrote:

Anonymous wrote:I don’t have a link but a couple months ago they had an interview on NPR (A1 maybe?) on this issue. The result was that AI was better for routine issues but that human physicians were much better on detecting cases that fell outside the routine.

This is ALREADY one of my biggest frustrations with human physicians - they are trained to think "when you hear hoofbeats, think horses, not zebras," and will make patients suffer catastrophic issues to prove it's not horses. So anything that reinforces that bias further could be dangerous.

I wonder if AI would have the same bias against a zebra diagnosis.

I was recently in the ER with weird to me symptoms. The blood tests had three results flagged and the ER doctor zeroed into one and gave me a relatively common, not great, diagnosis. I followed up two weeks later with my GP, who focused in on the other two flagged results, and ordered a common routine test that the ER had not done. That test confirmed a totally different, less bad (though still not great), diagnosis whose incidence is quite rare--1 to 2 per 100,000 people.

If the ER doctor had used AI, would he have ordered the additional routine (and cheap) test that would have confirmed the rare condition before he came to a conclusion on the diagnosis? I he had, I could have gone directly to the right specialist instead of having four appointments (including for tests) with a specialist I don't really need to see.

[Anonymous]

Anonymous wrote:

Anonymous wrote:Post the study, please.

https://the-decoder.com/openai-says-its-latest-models-outperform-doctors-in-medical-benchmark/

They have a link to the paper there.

But that's not a study. They link to a dataset and to an unreviewed write-up of what they think that means for their product.

I don't think anyone excited about this has any understanding of research. Probably about at the level of AI diagnosis, but sure. Go do that.

[Anonymous]

Anonymous wrote:Post the study, please.

https://the-decoder.com/openai-says-its-latest-models-outperform-doctors-in-medical-benchmark/

They have a link to the paper there.

But that's not a study. They link to a dataset and to an unreviewed write-up of what they think that means for their product.

I don't think anyone excited about this has any understanding of research. Probably about at the level of AI diagnosis, but sure. Go do that.

[Anonymous]

If we actually cared about health, the AI would be free to patients. Doctors could still be available for those who wanted a human, for hands-on issues (like surgery), and for consultation if/when the AI was inconclusive. But about 75% of the time I go to see a whitecoat, it's simply because I need someone with letters after their name to order the Rx I already know I need and can ask for by name. And the NP I typically see is often looking up the answers on google anyway.

Just let me connect to the data myself, thanks.

[Anonymous]

Anonymous wrote:If we actually cared about health, the AI would be free to patients. Doctors could still be available for those who wanted a human, for hands-on issues (like surgery), and for consultation if/when the AI was inconclusive. But about 75% of the time I go to see a whitecoat, it's simply because I need someone with letters after their name to order the Rx I already know I need and can ask for by name. And the NP I typically see is often looking up the answers on google anyway.

Just let me connect to the data myself, thanks.

But this won't be used to decrease gatekeeping. It will be used to increase profits. Nobody cares about health.

[Anonymous]

Anonymous wrote:

Anonymous wrote:

Anonymous wrote:

Anonymous wrote:

Anonymous wrote:Many doctors have not learned or barely practice the art of hands on medicine. They rely on labs, which is something AI can do pretty well. But hands on is really needed in many contexts and often helps to avoid expensive tests.

AI can't really replicate hands on medicine; doctors need to do more of this.

What do I need hands in medicine for though for many types of issues?

Let's say I have some kind of unknown infection. I take images of the skin rash, input my symptoms into AI, along with my labs delivered electronically and AI comes up with the highest probable diagnosis and appropriate course of action/treatment. I don't really need a handson clinical, do I? AI can also keep training itself on the entire body of new research and literature available so that it can constantly update the best prescription for treatment regimens, optimal dosing for drugs, etc. while a human physician probably almost never reads any literature after med school.

Struggling here to see why we need any doctors for hands on work if AI now does it with less error rates than a human.

If you have a skin lesion and AI diagnoses it, who is going to remove it if that’s the recommended treatment? I assume one day a robot could do it, but I think that’s a longer way off.

What if you have symptoms that can’t be shown in a photo? If I have abdominal pain, the doctor doing an exam and putting their hands on my belly to assess for pain, feel for masses, etc is doing something that AI can’t.

I think medicine is going to change a lot, but there will be a role for doctors for a while, probably for some fields longer than others.

There will definately be a role for doctors in the future. AI cannot replace them. But doctors can integrate AI/medical space searches to help direct them and/or confirm a possible diagnosis. Because when I can search and find a diagnosis for my symptoms, yet a visit to the doctor has them going a different path, you wonder can they investigate both paths?

"It's never lupus."

Doctors have experience, intution, AI has some past data. AI can be valuable but doctors have knowledge and wisdom.

Ai has been a lifesaver for me. I’ve seen tons of doctors who all blew me off. I could put my symptoms in and get an idea of what it was and it was something I suspected. Out of many dozens of doctors I finally got one after ten years to give me the evidence I needed and now just need to get to the right specialists but there are only a few. Most doctors can treat the basics but not more specialized things. They spend a few minutes with you and if it’s not something common they say you’re fine.

[Anonymous]

Anonymous wrote:If we actually cared about health, the AI would be free to patients. Doctors could still be available for those who wanted a human, for hands-on issues (like surgery), and for consultation if/when the AI was inconclusive. But about 75% of the time I go to see a whitecoat, it's simply because I need someone with letters after their name to order the Rx I already know I need and can ask for by name. And the NP I typically see is often looking up the answers on google anyway.

Just let me connect to the data myself, thanks.

I’d rather have a robot for surgery. Been to too many dictors who should not be doctors.

[Anonymous]

Anonymous wrote:

Anonymous wrote:If we actually cared about health, the AI would be free to patients. Doctors could still be available for those who wanted a human, for hands-on issues (like surgery), and for consultation if/when the AI was inconclusive. But about 75% of the time I go to see a whitecoat, it's simply because I need someone with letters after their name to order the Rx I already know I need and can ask for by name. And the NP I typically see is often looking up the answers on google anyway.

Just let me connect to the data myself, thanks.

Been to too many doctors who should not be doctors.

This is true. Universities seem to be nothing more than diploma mills these days, giving anyone with the money a degree, and everyone has the money now with government loans they hand out like candy.

[Anonymous]

Anonymous wrote:

Anonymous wrote:

Anonymous wrote:I don’t have a link but a couple months ago they had an interview on NPR (A1 maybe?) on this issue. The result was that AI was better for routine issues but that human physicians were much better on detecting cases that fell outside the routine.

This is ALREADY one of my biggest frustrations with human physicians - they are trained to think "when you hear hoofbeats, think horses, not zebras," and will make patients suffer catastrophic issues to prove it's not horses. So anything that reinforces that bias further could be dangerous.

I wonder if AI would have the same bias against a zebra diagnosis.

DP but yes they do.
They suffer from confirmation bias as well. They are still programmed by humans, and most all humans are flawed and egotistical.
I have noticed that certain questioning techniques tend to get them to become more questioning of the information they express and questioning their "sources", but even then they are programmed to be "loyal" to certain things.

[Anonymous]

Anonymous wrote:

Anonymous wrote:

Anonymous wrote:

Anonymous wrote:

Anonymous wrote:Also, radiology. Why isn't that a field that will get absolutely decimated by AI. They will just take images and have them interpreted by AI that can use image analysis and machine vision that is going to be less error probe and less biased than a radiologist. No need to pay an army of radiologists $500k salaries anymore when AI can do all of the work in 1/10th the time, with less errors, and for a fraction of the cost.

I would not sign up to have an ERCP performed by AI. You wouldn’t either.

People won't have a choice soon when insurance no longer covers physician fees when AI will cost $20 with less error...

I don’t think you know what an ERCP is. AI cannot even reliably drive cars. It definitely cannot do this radiological procedure in its present state.

And how fast did the smartphone evolve again?

Great point—not that fast. Mine has more computing power than involved in the moon landing and still cannot drive my car.

[Anonymous]

Anonymous wrote:

Anonymous wrote:If we actually cared about health, the AI would be free to patients. Doctors could still be available for those who wanted a human, for hands-on issues (like surgery), and for consultation if/when the AI was inconclusive. But about 75% of the time I go to see a whitecoat, it's simply because I need someone with letters after their name to order the Rx I already know I need and can ask for by name. And the NP I typically see is often looking up the answers on google anyway.

Just let me connect to the data myself, thanks.

I’d rather have a robot for surgery. Been to too many doctors who should not be doctors.

Autonomous surgical robots aren't coming anytime soon. An Nvidia article from last year talks about robots that've been trained to autonomously lift a piece of human tissue, tie a surgical knot, and one other thing that I forget.