Ask HN: Should primary care doctors be replaced with AI?
4 comments
I believe it absolutely should be, and it can even be applied to rare disease diagnosis.
My child was just saved by AI. He suffered from persistent seizures, and after visiting three hospitals, none were able to provide an accurate diagnosis. Only when I uploaded all of his medical records to an AI system did it immediately suggest a high suspicion of MOGAD-FLAMES — a condition with an epidemiology of roughly one in ten million.
Subsequent testing confirmed the diagnosis, and with the right treatment, my child recovered rapidly.
For rare diseases, it is impossible to expect every physician to master all the details. But AI excels at this. I believe this may even be the first domain where both doctors and AI can jointly agree that deployment is ready to begin.
My child was just saved by AI. He suffered from persistent seizures, and after visiting three hospitals, none were able to provide an accurate diagnosis. Only when I uploaded all of his medical records to an AI system did it immediately suggest a high suspicion of MOGAD-FLAMES — a condition with an epidemiology of roughly one in ten million.
Subsequent testing confirmed the diagnosis, and with the right treatment, my child recovered rapidly.
For rare diseases, it is impossible to expect every physician to master all the details. But AI excels at this. I believe this may even be the first domain where both doctors and AI can jointly agree that deployment is ready to begin.
By LLM, you mean some flavor of ChatGPT? Have you checked its results using non-LLM sources? It sounds like
a mild case of AI mania. Your doctor presumably knows what a dangerous LDL level is, maybe they weren’t concerned because it isn’t dangerous?
Yes, precisely these tools and yes, I did cross-reference with studies. In-fact the physician immediately misattributed the cause, and I had to guide her into self-correcting herself, since she failed to ask conditional followup questions that I forced the LLM to.
In-fact, in another similar case, an almost identical thing happened to my partner. She was experiencing medical issues and asked her friends, who were doctors. They confidently gave a flurry of knee jerk responses of the cause without carefully considering all the variables I forced the LLM to do. With the guidance of the LLM, we took a local diagnostic at a nearby pharmacy, CORRECTED the pharmacist's recommendation who attributed it to 'the heat', and the problem was solved within a few hours after determining it was related to a magnesium deficiency.
I'm not saying it's perfect out of of the box, but I remember getting excited when OCR medical imaging 5 years or so was 85% as effective as a doctor, now its surpassed human performance and for much the same reasons that LLMs are superior.
The primary mistake of the physician, and most, is their knowledge window is limited, as another comment cites. Also, they tend to, based on my observation be more reactionary. If you're not visibly sick then you're not sick and several studies demonstrate the long-term compounding effects of LDL at these levels without side effects for years.
Again, the argument isn't to replace primary care physicians with self-diagnostic ChatGPT usage, but rather than in the near term we will observe a threshold similar to medical image recognition surpassing primary care physicians specifically and at some point we will reach a threshold where physician interaction is in-fact meddlesome.
In-fact, in another similar case, an almost identical thing happened to my partner. She was experiencing medical issues and asked her friends, who were doctors. They confidently gave a flurry of knee jerk responses of the cause without carefully considering all the variables I forced the LLM to do. With the guidance of the LLM, we took a local diagnostic at a nearby pharmacy, CORRECTED the pharmacist's recommendation who attributed it to 'the heat', and the problem was solved within a few hours after determining it was related to a magnesium deficiency.
I'm not saying it's perfect out of of the box, but I remember getting excited when OCR medical imaging 5 years or so was 85% as effective as a doctor, now its surpassed human performance and for much the same reasons that LLMs are superior.
The primary mistake of the physician, and most, is their knowledge window is limited, as another comment cites. Also, they tend to, based on my observation be more reactionary. If you're not visibly sick then you're not sick and several studies demonstrate the long-term compounding effects of LDL at these levels without side effects for years.
Again, the argument isn't to replace primary care physicians with self-diagnostic ChatGPT usage, but rather than in the near term we will observe a threshold similar to medical image recognition surpassing primary care physicians specifically and at some point we will reach a threshold where physician interaction is in-fact meddlesome.
I totally get how annoying all this running around is! Having to call back and forth just to get a lab order or a follow-up request, and then the doctor just jumps to conclusions and labels it a "diet issue"—but then the LLM actually asks for details first before giving advice? Anyone would start wondering, “Can AI actually do the doctor’s job?”
But honestly, I don’t think it’s about “replacing” them—it’s about “filling in the gaps.” Think about it: what’s LLM good at? Digging through data and picking up on little details, like spotting right away that your LDL doesn’t match your age, even linking it to genetics. Way more accurate than doctors relying on their gut and making assumptions. But what’s it missing? The ability to be there in person to get a feel for things. Like when you see a doctor—they can check if you look pale, if your ankles are swollen, or feel your pulse to tell if your heart rate’s steady. That’s the invisible stuff AI can’t catch. And then there are the tiny symptoms you might not mention—like occasional dizziness or tossing and turning at night—doctors can pull that out just by chatting, but LLM can’t exactly stare through a screen and notice that, right?
And you mentioned doctors “only referring people”—but primary care docs are supposed to be the “first filter,” y’know? It’s not just writing orders or sending you to specialists. They should help connect the dots with all your random symptoms. If your LDL’s high, a good one would ask, “Does anyone in your family have this?” or “Do you ever get chest tightness?” Then they’d pair that with your blood work to decide—do we prescribe statins now, or run more tests to rule out other stuff? The doctor you saw just didn’t do that. It’s not that “the primary care role is useless”—she just didn’t do her job as the “gatekeeper.”
And about the statins: LLM was right you need them, but the details? How much to take? Do you need to check your liver (since statins can mess with that)? Are you on other meds that might clash with it? Only a doctor can call those shots. I helped my mom with her high blood pressure once—AI suggested a pill that interacted with her diabetes meds. Thank god the doctor caught that! AI can read medical guidelines, but it can’t wrap its head around all your personal stuff the way a human can.
And when people say doctors “mess up a lot”—it’s not all their fault. So many primary docs see 30+ patients a day, five minutes each. They don’t have time to ask all the questions, so they default to the “most common cause” (like blaming diet for high LDL). But AI doesn’t care about “saving time”—it’ll list every possibility and ask for clarifications first. That’s why it’s perfect as a helper, not a replacement. Imagine: before the doctor sees you, AI organizes your medical history, flags the weird parts (like your LDL vs. age), then the doctor just focuses on checking those. Saves time and avoids misses.
At the end of the day, your frustrating experience isn’t about “do we need primary care docs?” It’s about “how do we stop them from being swamped and careless?” AI can fix the data stuff and the detail questions, but it can’t replace that human call. Taking meds isn’t nothing—you need someone to watch if you get side effects, or calm you down when you’re panicking (“It’s okay, let’s run one more test first”). Right now, only a real doctor can do that.
In order to get this done, I needed to have a call with my primary care physician essentially as a hurdle to get their authorization.
Then, when the results came back, I uploaded them to the LLM and found I had dangerously high LDL for my age. The LLM recommended a statin since it was likely hereditary, which my additional research suggests is the correct suggestion.
To get a second opinion I scheduled another call and got a dismissive response that essentially concluded there was nothing to be done.
First, she misdiagnosed it directly assuming it must be my diet and gave me the generic 'eat healthy' routine, until I told her I already did, the LLM first asked some clarifying questions before jumping to conclusions, so I guided her later to realize it was hereditary to correct her.
Now I need to call her again to re-rest in another month, then again a forth time to get a statin.
The strange ethical question is should primary care doctors, which typically only route the patient to diagnostics and specialty care physicians be replaced by an LLM entirely?
The paradox is we seem to be of the belief they are needed and superior, but is it not the case they are often wrong?