Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

inflecthealth.medium.com

Post reply on HN
Print view View original

311–320 of 368 posts

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#311

Post by IIAOPSW »

oidar wrote: post #310
IIAOPSW wrote: post #300

Earlier quoted context omitted.

I disagree. I've seen it ask questions unprompted. It happened in a jailbreak I stumbled into by accident. The LLM could do it in principle. OpenAI restricts it in unrpoductive ways. https://i.imgur.com/WIzVXMm.jpeg

I had this conversation recently: ### You Can you guess if I am an AI or not? ### ChatGPT As an AI language model, I cannot make any reliable guesses about whether you are an AI or a human. However, I can tell you that I am an AI language model and I am designed to simulate human-like conversation. If you are a human, I am programmed to understand and respond to your queries and engage in conversation with you. ### Y…

Wow, I never knew the reverse Turing test would be so easy.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#312

Post by aix1 »

I am highly unimpressed by this piece. It reads as if its whole purpose is to grab headlines rather than conduct serious scientific inquiry into the current state -- and limitations -- of these AI methods.

1. Which version of GPT did the author use? There's a huge difference. (The article says "the current version".)

2. How did he choose the subject cohort? (The author doesn't seem to even know how many subjects there were; the article says "35 to 40 patients"... I really do hope he's gone through an appropriate ethics review before feeding his patients' data into a third-party non-HIPAA system.)

3. There no evidence of him trying to get the best out of the model (eg through prompt engineering).

4. He assumes that his own diagnoses are 100% correct.

5. There is no control group (other doctors diagnosing the same patients).

and so on

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#313

Post by gymbeaux »

PostOnce wrote: post #246
MrPatan wrote: post #227

One complaint he has is "It doesn't know to ask the right questions". Well, the prompt was to give diagnoses, not questions. Ask GPT for the follow up questions first, then the diagnoses. This is fascinating in that, because now the machine speaks human, we subconsciously ascribe human agency to it. This guy was instictively treating it like a colleague, who would naturally ask follow up questions unprompted. But you…

Other commenters take offense at your comment "doctors are done" and point to doctors doing physical things. GPT can already write robot control and machine vision code (poorly). The real questions are harder. For example, what about patients who lie or are too stupid to answer correctly? They can still be helped by a real human doctor, but GPT would be clueless. Another angle: we didn't have radiation sickness, HIV,…

No study I am aware of has shown a statistically-significant causal relationship between computer usage and development of CTS.

I saw 3 orthopedic surgeons last year and all said I had CTS. I gave the same info to ChatGPT and it said Thoracic Outlet Syndrome.

I don’t have CTS. When the docs realized this, their answers were “I dunno, good luck…”

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#314

Post by dsfyu404ed »

mandmandam wrote: post #239
rscho wrote: post #235

Earlier quoted context omitted.

Doctors are absolutely not done. You are forgetting that most of medicine is doing things. Doctors are not limited to theory like lawyers are, for instance. And physical robots are still very far from competing with docs.

Don't underestimate how many walls and obstacle lawyers have put up in order to entrench and 'justify' their obscene charges. And don't expect them not to create more to fight back against any sort of AI assisted affordable legal system. I have no doubt that the big firms are already crafting up some connivations to shenanigan us with. And some doctors will probably hire them to do the same for their field.

Don't forget real-estate agents, accountants, PEs, and every other credentialed person the bulk of who's job duties are paper pushing and minor judgement calls of limited scope.

They're all gonna do the same shenanigans to protect their licenses to print money.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#315

Post by kashunstva »

MrPatan wrote: post #227

One complaint he has is "It doesn't know to ask the right questions". Well, the prompt was to give diagnoses, not questions. Ask GPT for the follow up questions first, then the diagnoses. This is fascinating in that, because now the machine speaks human, we subconsciously ascribe human agency to it. This guy was instictively treating it like a colleague, who would naturally ask follow up questions unprompted. But you…

> Doctors are so done.

As a group? Not likely.

My spouse spends her days doing complex aortic reconstructions, carotid endarterectomies, peripheral arterial bypasses, etc. Imagining a the robotic/AI future for that field is the stuff of dystopian nightmares.

But possibly you’re thinking solely of non-procedural medicine. Even there, only part of the clinical work is in transacting information. The human-to-human connection itself has therapeutic effects. I’m confident that ChatGPT can use words that if uttered by a human would otherwise convey care and empathy; but I suspect its “patients” would respond differently.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#316

Post by dsfyu404ed »

unsupp0rted wrote: post #303
lechacker wrote: post #278

Earlier quoted context omitted.

There's 0 (zero) chance I'd use a robot doctor over a (qualified, capable) human doctor. None.

There’s a 100% chance I’d use a robot doctor (qualified, capable) over a human doctor (qualified, capable). Robots don’t get tired or have fights with their spouses. Robots don’t follow red herrings based on the most recent conversation they saw in a chat group.

AI doctors will always exploit logical flaws to boost their narrowly defined KPIs.

This sounds bad but in reality it's a consistency upgrade over meat powered doctors who are trying to balance many, many more variables.

It's like the difference between a capricious bureaucracy that might be chasing revenue one day, stats the next, political clout the third and enforcing morals the fourth to a street gang that DGAF seven days a week as long as you pay them.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#317

Post by lapcat »

aix1 wrote: post #312

I am highly unimpressed by this piece. It reads as if its whole purpose is to grab headlines rather than conduct serious scientific inquiry into the current state -- and limitations -- of these AI methods. 1. Which version of GPT did the author use? There's a huge difference. (The article says "the current version".) 2. How did he choose the subject cohort? (The author doesn't seem to even know how many subjects ther…

It's a blog post, not a journal article. You're criticizing it as if it were a funded, peer-reviewed experiment, which seems unfair to me. The author is just one ER doctor.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#318

Post by oidar »

thelittleone wrote: post #267
lumb63 wrote: post #219

The tail end of this article, where the author talks about how many more patients he could see in his life if he had AI assistance, made me realize that part of healthcare cannot be solved by AI. The goal is not to see more patients; the goal is to help more patients get better. For a lot of patients who are frustrated and have not had their problems validated and have been to many doctors and seen no results or poor…

During my wife's second pregnancy she was admitted to hospital at 6 months due to some intense abdominal pain and inability to hold down food. The put her on PNT, pumped her stomach daily and run a bunch of tests. I logged every test. Weeks passed she was slipping away and the specialist had no real game plan. So I got another Doctor. A busy professor who travelled a lot and was clearly passionate about his field. Ye…

This happens so often. Physicians don't have time to keep up on ALL the case studies and ALL the new literature. They are super busy writing notes and taking care documentation and billing - getting on the phone with insurance companies to justify treatment etc. There's not enough time in the day to do it all - so they lean heavily on things like UpToDate. Some physicians get pissed if a pt. brings medical theories to them - but I think that's really the only way to go for these cases - the problem is that the general public doesn't have a working knowledge of medicine, so they are quickly over their head and out of their depth. A family member, someone who isn't afraid of reading journal articles - and understanding statistics, is typically the only way some of these problems are solved - they do it because of love - not because they are paid.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#319

Post by ChatGTP »

detrites wrote: post #240
MrPatan wrote: post #227

One complaint he has is "It doesn't know to ask the right questions". Well, the prompt was to give diagnoses, not questions. Ask GPT for the follow up questions first, then the diagnoses. This is fascinating in that, because now the machine speaks human, we subconsciously ascribe human agency to it. This guy was instictively treating it like a colleague, who would naturally ask follow up questions unprompted. But you…

One thing the article misses is doctors are not all equal. Some are freaking terrible. Others may be competent but susceptible to bad decisions when overworked. Others are simply great. How does ChatGPT fare against the whole group? For one, it can't become "overworked", so at least better in that regard. It may well be better than a terrible doctor too. Given malpractice is consistently a leading cause of death, it…

For one, it can't become "overworked".

I have seen it be overworked, it can be overworked.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#320

Post by cyrialize »

unsupp0rted wrote: post #303
lechacker wrote: post #278

Earlier quoted context omitted.

There's 0 (zero) chance I'd use a robot doctor over a (qualified, capable) human doctor. None.

There’s a 100% chance I’d use a robot doctor (qualified, capable) over a human doctor (qualified, capable). Robots don’t get tired or have fights with their spouses. Robots don’t follow red herrings based on the most recent conversation they saw in a chat group.

My fear of robot doctors are ones fully trained and controlled by the company creating them or insurance companies.

With a human doctor there may be variability in how they treat you and what they tell you - you could appeal to a person's empathy.

But what if robot doctors are trained by insurance companies to withhold information or procedures that should be done if the patient doesn't have insurance or can't afford it?

Human doctors do this too - I'm not denying that, but at least there's some variability there.

With a robot doctor you could have a 100% success rate in withholding information.

I could be entirely wrong, I don't know any laws surrounding withholding info, but there could be some legal grayness there.

An insurance company would never admit that their robot doctor was trained to withhold info. They could just say that their robot doctor got it wrong.

[EDIT] This is definitely a people controlling an AI problem versus an AI problem, but I still think it's worth calling out. Maybe it's the pessimistic inside of me, but I just don't see a future where specialized AI X will be trained in good faith by companies.

Post reply on HN