[Gareth321]

> I wish / hope the medical community will address stories like this before people lose trust in them entirely.

Too late for me. I have a similar story. ChatGPT helped me diagnose an issue which I had been suffering with my whole life. I'm a new person now. GPs don't have the time to spend hours investigating symptoms for patients. ChatGPT can provide accurate diagnoses in seconds. These tools should be in wide use today by GPs. Since they refuse, patients will take matters into their own hands.

FYI, there are now studies showing ChatGPT outperforms doctors in diagnosis. (https://www.uvahealth.com/news/does-ai-improve-doctors-diagn...) I can believe it.

[linsomniac]

GPs don't have time to do the investigation, but they also have biases.

My own story is one of bias. I spent much of the last 3 years with sinus infections (the part I wasn't on antibiotics). I went to a couple ENTs and one observed allergic reaction in my sinuses, did a small allergy panel, but that came back negative. He ultimately wanted to put me on a CPAP and nebulizer treatments. I fed all the data I got into ChatGPT deep research and it came back with an NIH study that said 25% of people in a study had localized allergic reactions that would show up one place, but not show up elsewhere on the body in an allergy test. I asked my ENT about it and he said "That's not how allergies work."

I decided to just try second generation allergy tablets to see if they helped, since that was an easy experiment. It's been over 6 months since I've had a sinus infection, where before this I couldn't go 6 weeks after antibiotics without a reoccurrence.

[BobaFloutist]

There are over a million licensed physicians in the US. If we assume that each one interacts with five patients per weekday, then in the six months since you had this experience, that would conservatively be six-hundred-million patient interactions in that time.

Now, obviously none of this math would actually hold up to any scrutiny, and there's a bevy of reasons that the quality of those interactions would not be random. But just as a sense of scale, and bearing in mind that a lot of people will easily remember a single egregious interaction for the rest of their life, and (very reasonably!) be eager to share their experience with others, it would require a frankly statistically impossibly low error rate to not be able to fill threads like these with anecdotes of the most heinous, unpleasant, ignorant, and incompetent anecdotes anyone could ever imagine.

And this is just looking at the sheer scale of medical care, completely ignoring the long hours and stressful situations many doctors work in, patients' imperfect memories and one-sided recollections (that doctors can never correct), and the fundamental truth that medicine is always, always a mixture of probabilistic and intuitive judgement calls that can easily, routinely be wrong, because it's almost never possible to know for sure what's happening in s given body, let alone what will happen.

That E.N.T. wasn't up to date on the latest research on allergies. They also weren't an allergy specialist. They also were the one with the knowledge, skills, and insight to consider and test for allergies in the first place.

Imagine if we held literally any other field to the standard we hold doctors. It's, on the one hand, fair, because they do something so important and dangerous and get compensated comparitively well. But on the other hand, they're humans with incomplete, flawed information, channeling an absurdly broad and deep well of still insufficient education that they're responsible for keeping up-to-date while looking at a unique system in unique circumstances and trying to figure out what, if anything, is going wrong. It's frankly impressive that they do as well as they do.

[ctoth]

If you fully accept everything BobaFloutist says, what do you do differently?

Nothing. You just... feel more sympathetic to doctors and less confident that your own experience meant anything.

Notice what's absent: any engagement with whether the AI-assisted approach actually worked, whether there's a systemic issue with ENTs not being current on allergy research, whether patients should try OTC interventions as cheap experiments, whether the 25% localized-reaction finding is real and undertaught.

The actual medical question and its resolution get zero attention.

Also though...

You are sort of just telling people "sometimes stuff is going to not work out, oh also there's this thing that can help, and you probably shouldn't use it?"

What is the action you would like people to take after reading your comment? Not use ChatGPT to attempt to solve things they have had issues solving with their human doctors?

[afavour]

> The study, from UVA Health’s Andrew S. Parsons, MD, MPH and colleagues, enlisted 50 physicians in family medicine, internal medicine and emergency medicine to put Chat GPT Plus to the test. Half were randomly assigned to use Chat GPT Plus to diagnose complex cases, while the other half relied on conventional methods such as medical reference sites

This is not ChatGPT outperforming doctors. It is doctors using ChatGPT.

[ryandrake]

For every one "ChatGPT accurately diagnosed my weird disease" anecdote, how many cases of "ChatGPT hallucinated obvious bullshit we ignored" are there? 100? 10,000? We'll never know, because nobody goes online to write about the failure cases.

[ctoth]

> nobody goes online to write about the failure cases.

Why wouldn't they? This would seem to be engagement bait for a certain type of Anti-AI person? Why would you expect this to be the case? "My dad died because he used that dumb machine" -- surely these will be everywhere right?

Let's make our beliefs pay rent in anticipated experiences!

[ryandrake]

Failure cases aren't just "patient died." They also include all the times where ChatGPT's "advice" aligned with their doctor's advice, and when ChatGPT's advice was just totally wrong and the patient correctly ignored it. Nobody knows how numerous these cases are.

[ctoth]

So your failure cases are now "it agreed with the doctor" and "the patient correctly identified bad advice."

Where's the failure?

[ryandrake]

These are failures to provide useful advice over and above what could be gotten from a professional. In the sense that ChatGPT is providing net-neutral (maybe slightly positive since it confirms the doctor's diagnosis) or net-negative benefits (in the case that it's just wasting the user's time with garbage).

[estearum]

This is a doctor feeding the LLM a case scenario, which means the hard part of identifying relevant signal from the extremely noisy and highly subjective human patient is already done.

[Workaccount2]

The problem doctors have is that 99/100 times ABC is caused by xyz, so they prescribe 123 and the problem goes away.

Overtime, as a human, the doctors just turn into ABC -> 123 machines.