AI in Healthcare Idea

1 points by ultimaterr ↗ HN
So I found gpt3, tried it out, and was generally impressed by what it could do. I wondered if I could make something that can do symptom analysis and ultimately replace a clinician entirely. So it is an AI symptom analyzer that can adjust/suggest new medication a patient needs to take if their current one doesn't suit them or refill via e-prescription. I want to completely replace medical doctors from the entire landscape and make it frictionless. Thoughts?

14 comments

[ 3.2 ms ] story [ 43.5 ms ] thread
Microsoft has a version of this from what I'm aware of, would it still be worthwhile to pursue this venture
Using GPT-3 for that sounds dangerous.
Not gpt-3 specifically, I’m talking about making my own language model.
not dangerous at all; it would allow for frictionless checkups via chatbot interface and allow you to give dietary recommendations and refill their prescription instantly without human intervention. This would be advantageous to the patient and their insurance company since they don't need to fork up the cash for insurance and copay for a visit.
IBM has been doing massive investments into this space. There are still lots of problems and you can't completely replace things such as specialists.

>refill via e-prescription

No, no, no, absolutely not. This is a horrible idea. IMMEDIATELY, when you let robots prescribe medicine without any human interference, two things will happen:

>Immediate posts on drug forums saying things like "I found out if you give inputs X, Y, and Z, you get a giant prescription for Xanax, Ambien, and Adderall"

>Any small bug can be massively magnified into potentially prescribing the wrong medication to literally MILLIONS of patients

There has to be human involvement at some point in the process.

What if it refills that are not for controlled substances like Lipitor, metformin, and beta blockers instantly without human intervention? Most of the healthcare in the US is for treating chronic disease.
I am a doctor, so take my answer how you will. Clinical medicine is much more than just symptom analysis, and is often applying general reasoning to medical problems. Also, Diagnosis is much more about the subtleties of how someone tells you about a symptom than the symptom itself. Communication skills are also important (even though I acknowledge not everyone has a good experience of this), and regulatory and risk issues come in to play. I don’t think I would work with a GPT language model for healthcare as a patient or clinician. For simple things, basic flowcharts or written algorithms work well (and should be used more!). For anything more, I’d skip straight to a human.
Gotcha doc, thanks for you insight. But what about the millions of Americans that just go over and over to the same doctor to get a refill on their Lipitor, metformin, or beta blocker (metroprolol)? wouldn't it be easier to cater to that sector of healthcare with an AI that could automate the checkup process? just a thought.
For some things, there is definitely room for automation, but I think more basic algorithms/programs would solve that without the complexity of GPT like AI which can be somewhat unpredictable. But huge room for simple automation definitely.