Artificial intelligence is getting remarkably good at processing medicine.
It can summarize a hospital stay, generate a differential diagnosis, review symptoms and help physicians work through enormous amounts of clinical information.
But there is a part of medicine that happens before any of that information reaches the algorithm.
A doctor looks at the patient.
And notices something is wrong.
That was the lesson a physician and former University of Miami medical student describes in a new TIME essay. As a student in Miami, she watched a supervising doctor meet with a patient complaining of abdominal pain and constipation.
The obvious diagnosis appeared to be irritable bowel syndrome.
The doctor didn't stop there.
He noticed the patient wasn't making eye contact.
He asked why.
The patient eventually revealed that her sister had just been diagnosed with stage 4 colon cancer. Her concern wasn't simply abdominal pain. She was frightened that she might have cancer too.
The encounter changed the student's understanding of what medicine actually involves.
"You learn how to look," the supervising physician told her. "When you care, you get a feel for it."
That lesson has become more relevant as AI moves into clinical environments.
AI Is Very Good at the Information It Gets
Modern medical AI is increasingly capable of working with information that has already been collected.
Give it symptoms, test results, medical histories or clinical notes and it can process them rapidly and identify patterns that might otherwise take a physician considerably longer to find.
But that creates a potentially important distinction.
AI can reason from the input. It doesn't necessarily decide what the input should have been.
A physician may notice a hesitation before an answer, a change in demeanor, something unusual during a physical examination or a detail that doesn't quite fit the patient's initial explanation.
Those observations can determine which questions get asked next.
And those questions determine what information becomes available.
That is a very different role from simply processing a completed medical record.
The Risk Isn't That AI Gets Too Good
The more interesting risk may be that healthcare gets better at turning doctors into the kind of workers AI is already good at replacing.
The TIME essay makes an uncomfortable observation: healthcare systems have increasingly pushed physicians toward documentation, protocols, checklists and electronic records.
Those are precisely the environments in which machines excel.
If doctors spend less time observing patients and developing clinical judgment, there may eventually be less distinction between the human physician and the AI system assisting them.
The answer isn't to keep AI out of medicine.
It is almost certainly the opposite.
Use AI to remove the administrative burden. Use it to summarize records, surface possibilities and process information. Let machines do more of the work they are exceptionally good at.
Then give physicians more time to do the things machines struggle to replicate.
Look.
Listen.
Notice.
Question.
Build trust.
Miami Has a Particular Connection to This Debate
There is an interesting local dimension to this argument.
The story's central lesson was learned by its author while she was a medical student at the University of Miami, long before today's explosion of AI scribes, diagnostic assistants and medical chatbots.
The technology has changed dramatically since then.
The underlying lesson hasn't.
As Miami's healthcare and technology communities increasingly explore AI, the question shouldn't simply be how much medicine can AI automate?
It should also be what kind of medicine do we want AI to help create?
If AI gives doctors more time with patients, it could make medicine more human.
If it simply gives healthcare systems a faster way to process more patients, document more encounters and follow more protocols, it could have the opposite effect.
The difference may not ultimately be determined by the model.
It may be determined by how physicians use it.
AI will probably get much better at recognizing disease.
The harder question is whether the humans using it will still have enough time, and enough training, to recognize the patient.
Reporting Source: This article builds upon reporting from TIME and adds analysis of what the development means for Miami and South Florida.