AI & human flourishing
The EMH Question: When Is an AI Doctor Better Than No Doctor?
Kenroy George · 2026-09-14 · 6 min read
TL;DR: The rich world debates whether an AI doctor is better than a human doctor. For most of humanity, that is the wrong question. The real choice is between AI-assisted care with honest guardrails and no timely care at all, and pretending otherwise is its own kind of ethical failure.
Please state the nature of the medical emergency
In Star Trek Voyager, a ship gets stranded on the far side of the galaxy, its doctor dies in the first episode, and an Emergency Medical Hologram switches on. The EMH was designed as a stopgap, a program that activates only when no human physician is available. "Please state the nature of the medical emergency," it says, and then it gets to work, because there is no one else who can.
That premise was written as science fiction. It is also, with the serial numbers filed off, an ordinary Tuesday across much of the Caribbean, rural Africa, and rural America. A mother on a small island whose nearest specialist requires a flight and a hotel stay. A farmer whose district clinic has one overworked nurse and a doctor who visits twice a month. A family in a rural county whose hospital closed years ago. For all of them, the question was never AI doctor versus human doctor. The human doctor was never in the room.
The debate we are having versus the reality most people live
When AI in medicine gets debated in wealthy cities, the comparison is almost always AI versus the best available physician. Can a model outperform a board-certified specialist with twenty years of experience, a full diagnostic lab, and a same-day MRI? That is a fair research question, and a demanding bar, and it is the right bar for places where that specialist actually exists and can actually be seen.
But that comparison quietly assumes access. It assumes the patient can get an appointment this month, afford the visit, and reach the building. Strip those assumptions away and the ethical picture inverts. If the realistic alternative is a months-long wait, a day of travel, or simply going without, then measuring AI against the world's best doctor is like refusing to hand out life jackets because they are not as good as a coast guard cutter.
Call it the difference between the standard of care and the standard of access. The standard of care asks what the best possible treatment looks like. The standard of access asks what treatment a specific person can actually reach, this week, with the money and transport and time they actually have. Ethics that only ever reasons from the first standard ends up protecting people who already have doctors while doing nothing for people who do not.
None of this is an argument that AI care is equivalent to a physician. It is not, and saying so would be a lie. It is an argument about the correct comparison class. A tool does not have to beat the best doctor on earth to be worth deploying. It has to be meaningfully better and safer than the nothing it replaces, and it has to be honest about the difference.
What honest guardrails look like
The EMH framing is useful precisely because the hologram knew what it was: an emergency measure, not a replacement for a medical staff. Any real system built for physician-scarce settings should hold itself to the same honesty. In practice that means a few non-negotiables.
AI assists and triages, it does not replace judgment. The highest-value work in a low-access setting is often unglamorous: helping someone understand whether a symptom is urgent, preparing a clear history before a scarce consultation, catching the case that should not wait. That is assistance and triage, and it is where AI can do real good without overreaching.
Every path escalates to a human. An AI that dead-ends at its own answer is a trap. The system has to know its limits and route people toward real clinicians, whether that is a teleconsultation, a referral, or a plain instruction to get to a facility now.
Accountable clinicians stay in the loop for decisions. Diagnosis and treatment decisions belong with licensed professionals who can be named and held responsible. Software can inform those decisions. It cannot absorb accountability for them, and no terms-of-service paragraph changes that.
Honesty about uncertainty. When the system does not know, it must say it does not know, plainly, in language the patient understands. Confident wrongness is more dangerous in a village with no second opinion than it is in a city full of them.
No pretending. An AI assistant is not a licensed physician and must never be dressed up as one. The moment a product blurs that line to seem more capable, it has traded patient safety for marketing, and it deserves every bit of regulatory anger it attracts.
These guardrails are not obstacles to the mission. They are the mission. The goal is not to ship a fake doctor to people who lack a real one. The goal is to extend the reach of real doctors, and to make the time between a worry and a qualified human as short and as safe as possible.
What we are building at Cari
This is the thinking behind Cari Medical. It is a health platform connecting patients with verified doctors across the Caribbean, with booking, records, and AI-assisted care working together rather than in competition. The AI helps people describe what is happening, organizes their history, and helps them get to the right verified clinician sooner. The doctors remain the doctors.
We build it this way because we come from the places where the access gap is not a policy abstraction. When your starting point is a region where seeing a specialist can mean an airplane, you stop asking whether AI is better than a doctor and start asking how technology can put more people in front of one, with better information in hand, sooner than they otherwise would be.
The EMH eventually argued for its own personhood, which made for great television. We are after something much more modest and much more urgent: making sure that when someone in a physician-scarce place states the nature of their medical emergency, something trustworthy answers, and a real human physician is reachable at the end of the path.
Common questions
Is AI-assisted care safe without a doctor physically present?
It is safe only when the system is built to know its limits. That means AI that triages rather than prescribes on its own, escalation paths that connect patients to licensed clinicians, and accountable professionals making the actual medical decisions. A system missing any of those pieces is not ready for patients, no matter how impressive its demos are.
Who is accountable if the AI gets it wrong?
People, not software. In a properly designed system, licensed clinicians remain responsible for diagnosis and treatment, and the platform is responsible for being honest about uncertainty, escalating appropriately, and never presenting an assistant as a physician. Accountability that cannot be traced to a named human is not accountability.
Does this argument excuse lower-quality care for poorer regions?
No. The argument is about the comparison class, not the ceiling. AI-assisted care must clear a real safety bar and must be a bridge toward more human clinical capacity, not a substitute for building it. Accepting a life jacket while the fleet is built is prudence. Calling the life jacket a fleet would be the actual injustice.
If you are a patient or a clinician in the region and want to see how verified doctors, booking, records, and AI-assisted care fit together, you can explore Cari Medical at cari.care.