Why artificial intelligence isn't ready to be the clinician yet

There's a phrase that gets repeated at conferences and pitches: "AI will replace the doctor." It usually comes with an impressive number and a rising chart. It's a good phrase for selling — and a bad way to think about healthcare.

Not because artificial intelligence is bad. On the contrary: at many tasks, it's already better than any of us. The question is different, more surgical: is it ready to take, on its own, the place of whoever decides what happens to a patient? Looking at what AI is today, the answer is no — and it's worth understanding exactly why.

1. It doesn't perceive the patient. It processes data about them.

A health professional, beside a person, is reading far more than what they measure. They read the expression, the breathing, the hesitation, the "it's nothing" that means the opposite. They catch the signal that fits in no form.

AI doesn't see the person — it sees the data we give it about them. And in the populations that matter most to us, that's a serious problem: an older person with dementia, a patient in distress, someone who can't describe what they feel. It's precisely where self-report fails that the human gaze becomes irreplaceable. The machine wasn't in the room.

2. It is, at times, confidently wrong

This may be the most dangerous trait of today's AI: it errs with the same fluency with which it gets things right. It produces an articulate, plausible, self-assured answer — and sometimes simply a false one. Without hesitating, without flagging the doubt.

In a search engine, a wrong answer is an inconvenience. Beside a patient, it's another matter. A professional knows how to say "I'm not sure, let's check." AI tends not to know that it doesn't know — and that false confidence, in healthcare, is a real risk.

3. It cannot be responsible

When a clinician decides, someone is responsible for that decision: a trained person, subject to rules, ethics, accountability. If something goes wrong, there's someone to answer and someone to learn.

An algorithm doesn't carry that responsibility — it can't. "The model decided" is not an acceptable answer in front of a patient or a family. And a clinical decision with no one to own it isn't autonomy: it's an accountability vacuum dressed up as efficiency.

4. It fails on the rare case — and medicine lives on rare cases

Artificial intelligence is brilliant at the average. It learns from the most common patterns and answers them very well. The problem is that healthcare is full of exceptions, and the exception is often the vulnerable patient: the one who reacts atypically, who has several conditions at once, who doesn't fit the mould.

It's exactly there, in the long tail of unusual cases, that AI is weakest — and where an experienced professional makes the difference. Betting the safety of those who need it most precisely on the technology's weak point is a hard trade to justify.

5. It learned from data that doesn't represent everyone

A model is only as good as the data it learned from. And that data tends to under-represent precisely some of the populations we serve: the oldest, the multi-morbid, certain groups. The result is blind spots — and a system's blind spots tend to land on those already at a disadvantage.

6. It knows patterns, not meaning

Finally, there's something AI doesn't do: understand what's at stake. It doesn't weigh the dignity of a frightened person, a family's wishes, the difference between treating to cure and accompanying to comfort. It recognizes correlations; it doesn't assign meaning. And much of what a professional does, especially in caring for fragile people, is exactly that — making sense, deciding with values, caring.

"Not yet" is an open door

We say "not yet" on purpose. This isn't technophobia or a rejection of AI — it's a judgment about its current state. As reliability improves, evidence accumulates, regulation matures and accountability becomes clear, the role of artificial intelligence will grow, and rightly so.

Until then, the responsible way to use AI in healthcare is simple to state: in the service of the professional, not in their place. That's the line. And it's by respecting it that we keep building RVer around an old and still irreplaceable idea — that behind every session, there is one person caring for another.

← Back to the blog