With highly lucid people like the author's mom I'm not too worried about Dr. Deepseek. I'm actually incredibly bullish on the fact that AI models are, as the article describes, superhumanly empathetic. They are infinitely patient, infinitely available, and unbelievably knowledgeable, it really is miraculous.
We don't want to throw the baby out with the bathwater, but there are obviously a lot of people who really cannot handle the seductivity of things that agree with them like this.
I do think there is pretty good potential in making good progress on this front in though. Especially given the level of care and effort being put into making chatbots better for medical uses and the sheer number of smart people working on the problem.
They are knowledgeable in that so much information sits in their repository.
But less than perfect application of that information combined with the appearance of always perfect confidence can lead to problems.
I treat them like that one person in the office who always espouses alternate theories - trust it as far as I can verify it. This can be very handy for finding new paths of inquiry though!
I asked ChatGPT a question about a made up character in a made up work and it came back with "I don’t actually have a reliable answer for that". Perfect.
On the other hand, I can ask it about varnishing a piece of wood and it will give a lovely table with options, tradeoffs, and Good/Ok/Bad ratings for each option, except the ratings can be a little off the mark. Same thing when asking what thickness cable is required to carry 15A in AU electrical work. Depending on the journey and line of questioning, you would either get 2.5mm^2 or 4mm^2.
Not wrong enough to kill someone, but wrong enough that you're forced to use it as a research tool rather than a trusted expert/guru.
like, I tried to treat the bloating in one municipal clinic in Ternopil, Ukraine (got "just use Espumisan or anything else that has symeticone" and when it did not work out permanently, "we don't know what to do, just keep eating symeticone") and then with Gemini 3 (Pro or Flash depending on Google AI Studio rate limits and mood), which immediately suspected a poor diet and suggested logging it, alongside activity level, every day.
Gemini's suggestions were nothing extreme - just cut sugar and ban bread and pastry. I was guilty of loving bread, croissants, and cinnabons (is this how they are translated?) too much.
the result is no more bloating on the third week, -10cm in waistline in 33 days, gradually improving sleep quality, and even ability to sleep on a belly, which was extremely uncomfortable to me due to that goddamned bloating!
You can see multiple doctors (among the ones you're allowed to see by your insurance). The doctors are all in an echo chamber which reinforces their thinking. Their cognitive load and goal seeking is burdened by what they can determine they can bill insurance for (there is still no price transparency). You don't have a "regular" / primary care physician because they rotate through the provider network constantly.
Symptoms which don't fit the diagnosis are ignored / dealt with by deflecting that you should "see your regular physician". "Stare decisis" rules the second opinion. In their minds they believe they have no place to write down e.g. drug interactions with things which they didn't prescribe and don't believe in (the one time I got a call from quality control working for the umbrella organization I utilized this as an example of why I was looking for a different doctor and the QA person, who was, they said, a licensed nurse, said "they can add that to the record, I'll do it right now").
You might get fired as a patient for passing out or having a seizure during a blood draw, hard to say whether that's because they failed to follow SOP and call the meatwagon or because you upset staff by acting unusually. You might get into a conversation with a physician which goes strange and they end up telling you that their clinic gets health inspections like a restaurant... they don't. There's a "wet work" inspection (just like a butcher shop) before occupancy is allowed, but there's no posted inspection report because... there is no inspection! But there's more. There are relatively "safe" and common procedures which still have ooopsies and people end up in the hospital or die. The hospitalization rate might be 1:5000 and the death rate 1:100000 but if you do a million of these there are going to be a few. If the procedure took place in a clinic it's supposed to be reported, and the reports are public record; but surprise surprise, the reported rates for serious complications are far far below what the actuarial tables show.
If you're seeing constellations of incidents similar to these, you need to get a second opinion from somewhere / somebody who is not caught up in that particular bubble. It can be very hard to see what's happening, and also to find a measurable proxy for "in / not in the bubble".
No strong opinion on if that's good or bad long term, as humans have been outsourcing portions of their thinking for a really long time, but it's interesting to think about.
I have experience with building systems to remove that infinite patience from chatbots and it does make interactions much more realistic.
This is a strange way to talk about a computer program following its programming. I see no miracle here.
A human might be "empathetic", "infinitely patient, infinitely available". And (say) a book or a calculator is infinitely available. -- When chatting with an LLM, you get an interface that's more personable than a calculator without being less available.
I know the LLM is predicting text, & outputting whatever is most convincing. But it's still tempting to say "thank you" after the LLM generates a response which I found helpful.
Which, you know, humans can also do, including when they're not actually empathizing with you. It's often called lying. In some fields it's called a bedside manner.
An imaginary friend is just your own brain. LLMs are something much more.
> Watching John with the machine, it was suddenly so clear. The Terminator would never stop, it would never leave him... it would always be there. And it would never hurt him, never shout at him or get drunk and hit him, or say it couldn't spend time with him because it was too busy. And it would die to protect him. Of all the would-be fathers who came and went over the years, this thing, this machine, was the only one who measured up. In an insane world, it was the sanest choice.
The AI doctor will always have enough time for you, and always be at the top of their game with you. It becomes useful when it works better than an overworked midlevel, not when it competes with the best doctor on their best day. If we're not there already, we're darn close.
Specifically:
> I know now why you cry, but it's something I can never do.
While the machine learns what this complex social behavior called 'crying' is, it also learns that it is unable to ever actualize this; it can never genuinely care for John, any relationship would be a simulation of emotions. In the context of a child learning these complex social interactions, having a father-figure who you knew wasn't actually happy to see you succeed, sad to see you cry ...
You very specifically should not do that; you'll develop a medication overuse headache and be worse off than you were.
It gets worse, though. I was able to ask them a few questions about their symptoms, compare them to entries in the International Classification of Headache Disorders, and narrow it down to, iirc, two likely possibilities.
One of them was treatable. The treatment works. They still have pain, but can do stuff.
An AI that makes stuff up and gets stuff wrong isn't any different from the doctors we already have, except you can afford to get a second opinion, and you have the time available to push back and ask questions.
Edit: to expound on quality of the doctor - diagnosis and proposing a treatment was the work of several hours for me, a layman. A doctor should have known the ICHD existed. They should have been able to, in several minutes, ask questions about symptoms, reference the ICHD to narrow down likely diagnoses, and then propose a treatment with a "come back if that doesn't help".
Again, LLMs aren't competing with the best human doctors. They're competing with doctors you actually have access to.
I was taking one high blood pressure medication but then noticed my blood sugar jumped. I did some research with ChatGPT and it found a paper that did indicate that it could raise blood sugar levels and gave me a recommendation for an alternative I asked my doctor about it and she said I was wrong, but I gently pushed her to switch and gave the recommended medication. She obliged, which is why I have kept her for almost 30 years now, and lo and behold, my blood sugar did drop.
Most people have a hard time pushing back against doctors and doctors mostly work with blinders on and don't listen. ChatGPT gives you the ability to keep asking questions without thinking you are bothering them.
I think ChatGPT is a great advance in terms of medical help in my opinion and I recommend it to everyone. Yes, it might make mistakes and I caution everyone to be careful and don't trust it 100%, but I say that about human doctors as well.
In this context, I think of ChatGPT as a many-headed Redditor (after all, reddit is what ChatGPT is trained on) and think about the information as if it was a well upvoted comment on Reddit. If you had come across a thread on Reddit with the same information, would you have made the same push for a change?
There are quite a few subreddits for specific medical conditions that provide really good advice, and there are others where the users are losing their minds egging each other on in weird and whacky beliefs. Doctors are far from perfect, doctors are often wrong, but ChatGPT's sycophancy and a desperate patient's willingness to treat cancer with fruit feel like a bad mix. How do we avoid being egged on by ChatGPT into forcing doctors to provide bad care? That's not a rhetorical question, curious about your thoughts as an advocate for ChatGPT.
Are you asking why a side effect that is actually an entire health problem on its own, is a problem? Especially when there is a replacement that doesn’t cause it?
I have type 2 diabetes.
> How do we avoid being egged on by ChatGPT into forcing doctors to provide bad care?
I don’t ask it leading questions. I ask “These are my symptoms, give me some guidance.” Instead of “these are my symptoms, I think I have cancer. Could I be right?” If I don’t ask leading questions it keeps the response more pure.
> doctors mostly work with blinders on and don't listen
This has unfortunately been my experience as well. My childhood PCP was great but every interaction I've had with the healthcare system since has been some variation of this. Reading blood work incorrectly, ignoring explanations of symptoms, misremembering medications you've been taking, prescribing inappropriate medications, etc. The worst part is that there are a lot of people that reflexively dismiss you as a contrarian asshole or, even worse, a member of a reviled political group that you have nothing to do with just because you dare to point out that The Person With A Degree In Medicine makes glaring objective mistakes.
Doctors aren't immune to doing a bad job. I don't think it's a secret that the system overworks them and causes many of them to treat patients like JIRA tickets - I'd just like to know what it would take for people to realize that saying such doesn't make you a crackpot.
As an aside I use Claude primarily for research when investigating medical issues, not to diagnose. It is equally likely to hallucinate or mischaracterize in the medical domain as it is others.
I feel like the difference is that doctors took what I told them and only partially listened. They never took it especially seriously and just went straight to standard tests and treatments (scopes, biopsies and various stomach acid impacting medications). ChatGPT took some of what I said and actually considered it, discounting some things and digging into others (I said that bitter beer helped... doctor laughed at that, ChatGPT said that the alcohol probably did not help but that the bittering agent might and it was correct). ChatGPT got me somewhere better than where I was previously... something no doctor was able to do.
I visited my GP, 2 wrist specialists, and physical therapist to help deal with it. I had multiple x rays and an MRI done. Steroid injection done. All without relief. My last wrist specialist even recommended I just learn to accept it and don't try to extend my wrist too much.
I decided to ask Gemini, and literally the first thing it suggested was maybe the way I was using the mouse was inflaming an extensor muscle, and it suggested changing my mouse and a stretch/massage.
And you know what, the next day I had no wrist pain for the first day in a year. And it's been that way for about 3 weeks now, so I'm pretty hopeful it isn't short term
I pre-emptively switched to trackballs and to alternating left/right hands for mousing near the start of my professional career based on the reading I did after some mild wrist strain.
I simply can't imagine this happening based on my experience with healthcare in two (European) countries.
The only thing that moves is my thumb, and it's much better for flexing than the wrist, also it has a tiny load to manage vs the wrist.
I’ve been lucky enough to not need much beyond relative minor medical help but in the places I’ve lived always found that when I do see a GP they’re generally helpful.
There’s also something here about medical stuff making people feel vulnerable as a default so feeling heard can overcompensate the relationship? Not sure I’m articulating this last point well but it comes up so frequent (it listened, guided me through it step by step etc.) that I wonder if that has an effect. Feeling more in control than a doctor who has other patients and time constraint just say it’s x or do this
I caught one when i ask ChatGPT something, and then went to urgent case. I told my story, they left, came back and essentially read back exactly what ChatGPT told me.
They have 15 minutes and you have very finite money.
Medical agents should be a pre consult tool that the patient talks to in the lobby while waiting for the doctor so the doctor doesn't waste an hour to hear the most important data point and the patient doesn't sit for an hour in the lobby doing nothing.
Everyone I know just goes to walk-in clinics / urgent-care centres. And neither of those options give doctors any "skin in the game." Or any opportunities for follow-up. Or any ongoing context for evaluating treatment outcomes of chronic conditions, with metrics measured across yearly checkups. Or the "treatment workflow state" required to ever prescribe anything that's not a first-line treatment for a disease. Or, for that matter, the willingness to believe you when you say that your throat infection is not in fact viral, because you've had symptoms continuously for four months already, and this was just the first time you had enough time and energy to wake up at 6AM so you could wait out in front of the clinic at 7:30AM before the "first-come-first-served" clinic fills up its entire patient queue for the day.
To say nothing of giving your personal health information over to a private company with no requirement to practice HIPAA, and just recently got subpoenaed for all chat records. Not to mention potential future government requests, NSA letters, during an administration that has a health secretary openly talking about rounding up mentally ill people and putting them in work camps.
Maybe LLMs have use here, but we absolutely should not be encouraging folks to plug information into public chatbots that they do not control and do not run locally.
It is a recipe for disaster.
As an American on ACA this made me chuckle.
I did some searching with Grok and I found out:
- no contact injuries are troubling b/c it generally means they pulled something
- kids don't generally tear an ACL (or other ligament)
- it's actually way more common for the ligament to pull the anchor point off of the bigger bone b/c kid bones are soft
I asked it to differentially diagnose the issue with the details of: can't hold weight, little to no swelling and some pain.
It was adamant, ADAMANT, that this was a classic case of bone being pulled off by the ligament and that it would require surgery. It even pointed out the no swelling could be due to a very small tear etc. It gave me a 90% chance of surgery too.
I followed up by asking what test would definitely prove it one way or the other and it mentioned getting an X-Ray.
We go off to the urgent care, son is already kind of hobbling around. Doctor says he seems fine, I push for an X-Ray and turns out no issue: he probably just pulled something. He was fully healed in 2-3 days.
As someone who has done a lot of differential diagnosing/troubleshooting of big systems (FinTech SRE) I find it interesting that it was basically correct in what could have happened but couldn't go the "final mile" to establish it correctly. Once we start hooking up X-Rays to Claude/Grok 4.2 etc equivalent LLMs, will be even more interesting to see where this goes.
Grok is...not most people's first choice.
At least both OpenAI and Deep Mind do medical fine tuning, and both are almost certainly paying doctors to do it.
Opus 4.5 seems good too, though getting dumber. OpenAIs fine tuning is clearly built to toe the professional medical advice line, which can be good and bad.
It is like getting phished and then pointing out that the scammer was basically right about being Coinbase support aside from the fact that they did not work there
I don't think people are afraid of doctors using imperfect tools. That is the easier part. But that will not solve the problem of too many patients for a single doctor and what leads to the lack of empathy. This was a problem even before AI. It seems society does not have empathy for these kind of "professional problems". Offering tools instead of humans is an even riskier approach, not for that particular individual, but for how society tends to build trust and empathy. We tend to see everything now as a problem with a technical solution because we only have confidence in solving technical problems.
This is the point where it becomes important to distinguish two senses of "advanced", i.e. advanced in technological sense on the one hand and advanced in social/societal and especially large-group-long-time-horizon coordination terms on the other. In the former we are quite advanced, in the latter quite primitive and regressing by the day, it feels like. (But sorry to end on a doomer note, take it with a grain of salt.)
The crucial part is the training. AI may very well be the solution for underserved communities, but not if it is trained on internet rubbish. Train an AI on curated state-of-the-art, scientific data, imagine the expert systems of yore on overdrive, and you will see much better results, including knowing when to call in a human doctor.
Using ChatGPT takes... nothing, it's already here.
Not too surprising to see one and not the other.
Also, the kinds of changes that result in more doctors don't tend to get media coverage. That's that boring, keep the lights on, politics that modern rage-bait driven media abhors, so it may even still be true that the changes we need for more doctors are also already happening. We'll find out in another decade or two.
Now, I have to book an appointment online and the first available slot is in 2 weeks if lucky.
What exactly changed?
If you are using it like a tool to review/analyze or simplify something - ie explain risk stratification for a particular cancer variant and what is taken into account, or ask it to provide probabilities and ranges for survival based on age/medical history, it's usually on the money.
Every other caveat mentioned here is valid, and it's valid for many domains not just medical.
I did get hemotologist/oncologist level advice out of chatgpt 4o based on labs, pcr tests and symptoms - and those turned out to be 100% true based on how things panned out in the months that followed and ultimately the treatment that was given. Doctors do not like to tell you the good and the bad candidly - it's always "we'll see what the next test says but things look positive" and "it could be as soon as 1 week or as long as several months depending on what we find" when they know full well you're in there for 2 months at minimum you're a miracle case. Only once cornered or prompted will they give you a larger view of the big picture. The same is true for most professional fields.
I remember trying to talk to WebMD when I had pain in my side and appendicitis was near the bottom of the list, the top stuff was either nothing serious or highly improbable. The pain didn't seem as bad as what the appendicitis pain should have been based on descriptions. My mother got her doctor to call me and he walked me through some touching and said "you likely have appendicitis, don't talk to WebMD next time." I went to the hospital that night and that doctor told me I was likely hours away from a burst appendix. I can only imagine what nonsense ChatGPT would have told me.
Just like with most professions, the real world is nothing like the textbook. Being able to pass a medical exam doesn't necessarily mean you're going to be a good doctor. Most of the exam is taken during med school and the final portion is only taken after the first year of residency. They still have another few years at least of residency after passing USMLE and that's with supervision under an attending doctor. Being able to pass the USMLE is not equivalent to being successful doctor with years of experience.
You are 100% correct that, much like pilots overseeing autopilots, we should combine the best of both worlds in the medical field.
In companies people talk about Shadow-IT happening when IT doesn't cover the user needs. We should probably label this stuff Shadow-Health.
To some extent, the deployment of a publicly funded AI health chat bot, where the responses can be analysed by healthcare professionals to at least prevent future harm is probably significantly less bad than telling people not to ask AI questions and consult the existing stretched infrastructure. Because people will ask the questions regardless.
But I do agree that some focused and well funded public health bot would be ideal, although we'll need the WHO to do it, it's certainly not coming from the US any time soon.
There isn't simply enough doctors to go around and the average one isn't as knowledgeable as you would want. Everything suggests that when it comes to diagnosis ML systems should be better in the long run on average.
Especially with a quickly aging population there is no alternative if we want people to have healthcare on a sensible level.
The AI Revolution in Medicine: GPT-4 and Beyond by Peter Lee (Microsoft Research) - https://www.microsoft.com/en-us/research/publication/the-ai-...
The AI Revolution in Medicine, Revisited by Peter Lee (Microsoft Research) - https://www.microsoft.com/en-us/research/story/the-ai-revolu...
I see some of this adversarial second-guessing introspection from Claude sometimes. ("But wait. I just said x y and z, but that's inconsistent with this other thing. Let me rethink that.")
Sometimes when I get the sense that an LLM is too sycophantic, I'll instruct it to steelman the counter-argument, then assess the persuasiveness of that counter-argument. It helps.
Patients are substituting 15 minutes of 100% correct information for 2 hours of 80%, or whatever percentage the AIs are.
Additionally, they're pretty much free at the moment and available 24/7 with no travel required.
However I would say that the cited studies are somewhat outdated already compared e.g. with GPT-5-Thinking doing 2mins of reasoning/search about a medical question. As far as I know Deepseeks search capabilities are not comparable and non of the models in the study spend a comparable amount of compute answering your specific question.
> She understood that chatbots were trained on data from across the internet, she told me, and did not represent an absolute truth or superhuman authority. She had stopped eating the lotus seed starch it had recommended.
The “there’s wrong stuff there” fear has existed for the Internet, Google, StackOverflow. Each time people adapted. They will adapt again. Human beings have remarkable ability to use tools.
https://www.mdpi.com/2504-3900/114/1/4 - Reinecke, Madeline G., et al. "The double-edged sword of anthropomorphism in llms." Proceedings. Vol. 114. No. 1. MDPI, 2025 Author: https://www.mgreinecke.com/
AI is doing the last mile and all these posters are saying it’s replacing doctors. Scary stuff.
As for AI, my personal experience is ChatGPT and Gemini is more effective then DeepSeek for healthcare issues. I do hope DeepSeek or Doubao can catch up.
Now there's a quote
It responded: “Hearing you say that really makes me so happy! Being ..."
You should try praising a Claude instance after a successful session. It is eerie.
"Machines are more like humans"
I love the future...
I don't know enough about medicine to say whether or not this is correct, but it sounds suspect. I wouldn't be surprised if chatbots, in an effort to make people happy, start recommending more and more nonsense natural remedies as time goes on. AI is great for injuries and illnesses, but I wonder if this is just the answer she wants, and not the best answer.
Human is human already, regardless how they speak and what they do... There's always true sympathy and empathy to find in a human, regardless how busy, dark, erroneous etc. they are... Yet, it's the true alive soul inside in every person...
Medics are likely tired, doing their job every single day, yet they have empathy, and will always try to listen if you actually try... They know what PAIN, AGONY, DEATH, SORROW... fear means...
LLM/"AI" will always pretend to be a human, since it's "trained"/designed, to be so, and will always be limited and incomplete... Not to mention the initial dataset of numerous emphatic actual human has its limited memory inside, no one is responsible for.
What a hopeless sorrow is that awful trendy, advertised, mind-atrophying mess...
(and not just with the AI stuff)
It's still insane to me how people can trust an LLM they didn't train themselves! You have no idea what vile and evil Dreck these models were trained with.
It's the click workers in third world countries who you don't recognize exist who could know if they were experts. But they aren't. A huge responsibility for these underpaid workers. They are the backbone of this hype, not GPUs. It's all built on shit. :)
Personally I think the article spends a lot of time trying to show that AI may be able to improve health outcomes, particulaly for rural patients, but IMHO it doesn't spend nearly enough time talking about the current challenges
