In this article
When the Robot Was Kinder Than the Doctor
Imagine asking your mom a question and she answers without turning around. The words might be right. The feeling is wrong. A 2026 study found something like that, but with doctors. People rated AI chatbot replies as kinder than physician replies. The chatbot did not learn to care. So what changed?
I have been thinking about this study for two days now. A group of researchers asked people to read replies to medical questions. Some replies were written by real doctors. Some were written by an AI chatbot. The readers did not know which was which. The chatbot replies sounded kinder.
That is a strange thing to read.
In a 2026 paper in JAMA, a team led by Canio Martinelli pulled together studies that compared chatbot answers to physician answers (Martinelli et al., 2026). Across thirteen pooled studies, the chatbot replies scored higher on empathy — feeling with someone, not just for them. Not by a tiny bit. By a wide margin.
That finding has been turned into a loud story online. The story goes like this: doctors are about to be replaced. Medical school is pointless. Pack it up.
I do not think that story is true. But the surprise inside it is true, and it deserves more than a takedown. So I want to sit with it.
The line that stopped me
Empathy was supposed to be the last refuge of the clinician.
That sentence is from the JAMA paper itself (Martinelli et al., 2026). Read it slowly. The authors are not saying empathy fell. They are saying we expected it to be the last thing to fall. The thing that no machine could touch. And here is a study saying — actually, that fell first.
What the chapter is really about
Here is what I keep coming back to. The chatbot did not become more human. Nothing about the algorithm changed. It is the same kind of model it was last year — patterns and probability stacked very high.
So if the machine did not change, something else did.
Imagine a teacher who used to know your name. She looked up when you walked in. Now she spends every class filling out forms at her desk. She still cares about you. She is still good at teaching. She just is not looking up anymore.
That is not a teacher problem. That is a paperwork problem.
Or imagine asking your mom a question and she answers without turning around. The words might be right. But it does not feel right. Now imagine the kitchen has a little speaker that answers your question with full attention, in a warm voice. Whose answer would you rather have heard?
Most of us would still want our mom. But the speaker would feel kinder in the moment. That is the whole study, in one kitchen.
The next chapter is about how we got here — how the doctor stopped looking up.
References
Martinelli, C., Carnevale, V., Ercoli, A., et al. (2026). Artificial Intelligence Is Not the End of the Physician. JAMA. doi:10.1001/jama.2026.4356
How the Doctor Got Buried
No single rule pulled doctors away from the bedside. It happened slowly — paperwork, billing, insurance forms, inbox messages. Each piece looked reasonable on its own. Together they buried the patient. About half of a doctor's day now goes to screens. About a quarter goes to people. This is what burnout looks like from the inside.
Nobody decided to take doctors away from their patients. There was no meeting. No memo. It happened slowly, one small thing at a time, until the doctor could barely see the patient through the pile.
The JAMA authors give this a name: fraction by fraction. Each new task looked small on its own. A new form. A new screen. A new box to click. The kind of thing nobody would fight about. But small things added together get big. The authors put it in one line:
The transformation happened fraction by fraction… but fractions compound.
That is from the same JAMA paper (Martinelli et al., 2026). To compound here means small things piling up until the pile is enormous.
The numbers, in plain words
A 2016 study in the Annals of Internal Medicine followed real doctors through real office days (Sinsky et al., 2016). The numbers are uncomfortable.
About half of an office day goes to the computer. Filling in records, clicking through forms, doing the desk work that has to be done.
About a quarter of the day goes to the patient. The actual sit-with-the-person part. The reason most doctors say they wanted to be doctors in the first place.
Read those two lines next to each other. The screen got more time than the human. That is what we have built.
What got piled on
It is worth naming the pieces, because none of them sound bad on their own:
Paperwork. Records of every visit, every note, every plan.
Billing and coding. Every visit has to be turned into the right numbers so the insurance pays. Get the numbers wrong and the visit is unpaid.
Insurance approvals. For many treatments, the doctor has to ask permission from an insurance company before giving the treatment. This is called prior authorization — getting permission ahead of time. It can take hours per patient.
Performance scores. Did the doctor check the right boxes? Was the visit fast enough? Did the patient fill out the survey?
Inbox messages. Patient questions, lab results, refill requests — all flowing in all day, even after the office closes.
Each one is reasonable. The first three even make care safer. The pile is the problem, not any single piece of it.
What the pile does to people
There is a word for what happens when you do work you love, surrounded by work that is not the work. The word is burnout — being so tired from work that you stop feeling like yourself.
A 2025 study in Mayo Clinic Proceedings tracked doctor burnout from 2011 through 2023 (Shanafelt et al., 2025). The numbers swing, but the pattern is clear. About 6 in 10 doctors were burned out at the worst point of the COVID years. By 2023 the number had come down to about 4 in 10. That is better. It is not okay.
Now hold that next to the empathy study. Of course readers picked the warmer reply. Half of the doctors writing the other replies were running on empty.
Why this matters for the headline
The story online is that AI is more empathetic than doctors. The story underneath the story is different. The doctor was buried. The chatbot was not.
If you bury anyone — a teacher, a chef, a coach, a friend — under enough paperwork, they will be less warm too. That is not because they stopped caring. That is because warmth needs a person who is actually there.
The next chapter is about a single sentence in the JAMA paper that has stayed with me. It is about what AI is actually for.
References
Martinelli, C., Carnevale, V., Ercoli, A., et al. (2026). Artificial Intelligence Is Not the End of the Physician. JAMA. doi:10.1001/jama.2026.4356
Sinsky, C., Colligan, L., Li, L., et al. (2016). Allocation of physician time in ambulatory practice. Annals of Internal Medicine, 165(11), 753–760. doi:10.7326/M16-0961
Shanafelt, T. D., West, C. P., Sinsky, C., et al. (2025). Changes in burnout and satisfaction with work-life integration in physicians and the general US working population between 2011 and 2023. Mayo Clinic Proceedings, 100(7), 1142–1158. doi:10.1016/j.mayocp.2024.11.031
Digging the Doctor Out
What if AI's real job is not to be the doctor, but to dig the doctor out from under the pile of paperwork? Until now, every fix just shuffled the work between tired people. AI is the first tool that can actually take work away. But there is a careful part — and the JAMA authors say it clearly. An absent doctor cannot check the AI's work. A present one can.
Here is the thought I keep returning to. What if AI's real job is not to be the doctor, but to dig the doctor out from under the pile?
The JAMA authors have a word for this. Excavation — digging carefully through layers to find what is buried underneath. They say it plainly:
AI cannot replace the physician, but it can remove the burden of what replaced the physician's clinical time.
That is from Martinelli et al., 2026. Read it twice. AI is not the next pile of work added on top. It might be the first thing strong enough to take a pile away.
Why this is different
Until now, every fix to doctor burnout has done the same thing: take the paperwork away from the doctor and give it to somebody else. A nurse. A scribe. An assistant. The pile did not get smaller. It moved sideways. The other person got tired.
AI is the first tool that can actually absorb the pile instead of just shuffling it. A computer can write the visit note from the conversation. A computer can fill out the insurance form. A computer can sort the inbox. None of this is replacing the doctor. It is clearing the desk so the doctor can stand up and look at the person in the room again.
Think of it like a dishwasher in a kitchen. The dishwasher does not become the cook. It does the dishes so the cook can cook. Without it, the cook is washing plates instead of making dinner.
Or think of a calculator and your math homework. A calculator does not do your homework for you. It frees up your brain so you can think about the harder parts of the problem.
That is what AI looks like when it is doing its job.
The careful part
Here is the line in the paper that I think is the most important. Not the most quoted. Not the most exciting. The most important.
AI does not replace physician judgment, it preserves it.
Read it slowly (Martinelli et al., 2026). The doctor still has to be in the room. The doctor still has to sign off on what the AI suggests. The doctor still has to notice when the AI is wrong — and the AI will be wrong sometimes, because every tool gets things wrong sometimes.
A doctor who is not in the room cannot check the AI's work. A doctor who is in the room can. AI without a present doctor is like a self-driving car with no driver paying attention. It works until it does not, and there is no one there to catch the mistake.
The danger here is real. If a hospital uses AI to remove the doctor instead of free the doctor, the same problem comes back, just faster. A buried doctor was bad. A doctor sitting somewhere else, not in the room, signing off on whatever the AI generated — that is worse.
Two kinds of work
Reading the paper, I started sorting the doctor's day into two piles in my head.
The first pile is work that can be handed off to a tool. Filling out forms. Looking things up. Sorting messages. Spotting patterns in test results. Writing the first draft of a note. None of this is what the doctor came for. All of it can be helped by a good tool.
The second pile is work that cannot be handed off. A hand on the shoulder. Sitting with someone who is scared. Noticing the thing the patient did not say out loud. Holding someone's gaze when the news is hard. This is the work that needs a human in the chair across from another human.
AI gets the first pile. The doctor gets the second pile. That is the whole idea, and it is a good one.
Why this matters outside medicine
The shape of this question is not just for doctors. Almost every job has these two piles. The work that can be handed off, and the work that needs a human present. AI is about to test every profession on whether it can tell the difference.
The next chapter is about one place where I think we have to be the most careful — and the most honest about what AI is, and what it is not.
References
Martinelli, C., Carnevale, V., Ercoli, A., et al. (2026). Artificial Intelligence Is Not the End of the Physician. JAMA. doi:10.1001/jama.2026.4356
AI Is a Helper, Not a Stand-In
The honest line in the JAMA paper is one most AI companies will not say out loud. AI is not your doctor. It is not pretending to be. The same line needs to be said, just as honestly, about AI in mental health. A chatbot can listen at 2 a.m. when no human is awake. That is real. It is also not the same league as a psychiatrist or a clinical psychologist.
AI is not your doctor. It is not pretending to be one. The day we forget that is the day we lose something that took a long time to build.
I want to apply that same honesty to something closer to my own work — AI for mental health and reflection.
What AI is genuinely good at
There is a real, useful version of an AI assistant. It can listen. It can ask the next gentle question. It can be there at 2 a.m. when no human friend is awake. It can help someone put feelings into words on a hard night.
Those are not nothing. They are real things. For some people, on some nights, an AI that is patient and available and kind is the difference between writing something down and going to bed wordless.
AI is a good assistant. AI can be a surprisingly good companion. I believe both of those things.
What AI is not
Now the line that has to land hard.
Sol AI is not a replacement for a psychiatrist or a clinical psychologist. It is in a completely different league.
That is true of every AI tool I know of, including the ones I help build. I want to be careful here, because this is the part most easy to fudge.
A psychiatrist is a medical doctor. They went to medical school. They can diagnose mental illness. They can prescribe medicine. They can spot when something is medical and not just emotional — a thyroid problem, a sleep disorder, a side effect from another drug — and act on it.
A clinical psychologist has years of training in how the human mind breaks and how it heals. They are trained to notice things in a real conversation that no chatbot can pick up. The pause before someone answers. The look on a face. The catch in a voice when someone says I'm fine for the third time.
Both of them carry weight. Real, human, professional weight. They can be wrong with consequences. They can be right with consequences. That weight is part of what makes their care care.
An AI assistant cannot carry that. Not because the AI is bad — but because that kind of weight needs a human who has trained for years and who is responsible for the outcome. AI was not built for crisis. AI was not built for diagnosis. AI was not built to prescribe.
If you are reading this and any of it is hitting close to home — if you are stuck, overwhelmed, or worse — please remember this is a reflection on research, written from outside the clinic looking in. The people qualified to help with serious mental health concerns are mental health professionals. A chatbot can sit with you for an evening. It cannot do what they do.
Why naming this matters
The JAMA paper is brave in this exact way. The authors had every reason to soften the point. They are doctors. AI is the cool new thing. Saying "AI is not us" sounds, on the surface, like saying "AI is bad." It is not the same. They are saying AI is something else. A different kind of tool, with a different job.
I want to say the same thing about AI for reflection and journaling. It is not a therapist. It is not a doctor. It is something else — a patient companion, a thinking partner, a quiet place to put words at 2 a.m. That is enough. It does not need to be more, and it should not pretend to be more.
The line that ends the paper
The JAMA authors close with a sentence I keep saying out loud.
AI is not the end of the physician — it is the return of every reason to be one.
That is a beautiful line (Martinelli et al., 2026). It says the quiet part. AI is not here to take the doctor's place. AI is here to give the doctor back the work that made medicine a calling in the first place — the part that was about being with people.
If we get this right, the same thing might be true for the rest of us. AI is not the end of the writer. AI is not the end of the teacher. AI is not the end of the friend, the listener, the person on the other side of a hard conversation.
AI might just be the return of every reason to be one.
References
Martinelli, C., Carnevale, V., Ercoli, A., et al. (2026). Artificial Intelligence Is Not the End of the Physician. JAMA. doi:10.1001/jama.2026.4356
More by Hurroz
