이미지: METAL LAB 생성
Hello, Alien is a letter column for moments when humans pause in front of new technology — where an alien who got to Earth first pulls up a chair and explains. Today, a very smart machine is standing at the exam room door.
Dear Alien, A few days ago I showed my checkup results to an AI, and it cracked the hard numbers in an instant. I'd waited forever at the hospital, but this thing rattled off possible causes and next steps in three seconds flat.
Then I read an article. It said AI could outperform doctors at everything — taking a history, diagnosing, choosing tests, suggesting treatment, managing chronic disease — and that having a human in the loop sometimes even hurt performance. So does that mean doctors are about to become the people who just open the door for AI to walk through?
I want fast answers, but I also want to know whose face I'm supposed to look at if the answer turns out to be wrong. From Daejeon, Minho
Hello, Minho! First off, "AI is good at solving medical problems" and "AI is good at caring for patients" aren't the same sentence. A calculator that aces the exam doesn't reach out and touch the forehead of someone running a fever.
Medicine has recently been buzzing with forecasts that autonomous AI could surpass even doctor-AI collaboration in core tasks like taking histories, diagnosing, choosing tests, treating patients, and managing chronic conditions. The speed at which machines read data and weigh possibilities is already startling. Unlike human memory, it doesn't go foggy the morning after a late night at work.
But the story changes once actual patients start using AI. In a randomized study involving 1,298 UK adults, outcomes were significantly worse when people talked to a model about their symptoms than when the model tackled the problem on its own. Patients didn't know which symptoms mattered and left them out; the AI had the right clues but couldn't get the person to see why they mattered.
Put simply, it's like handing a brilliant detective a case with nothing but a note that says "something feels off." Asking the right questions matters just as much as producing the right answer. Behind "my stomach hurts" hides when it started, where exactly it hurts, and what medication the person's already taking.
That's exactly where the doctor's job survives. Doctors re-ask questions of patients who struggle to explain themselves, read expressions and hesitation, and decide which piece of conflicting information to trust. The doctor isn't the one transcribing the AI's answer — the doctor is the one finding what never got entered in the first place.
Delivering bad news is also nothing like grading an answer sheet. Given the same prognosis, one person wants every number spelled out right now, while another wants to wait until family arrives. Treatment means both choosing the right words and deciding when and how to hand them over.
Of course, there's a real risk that doctors lose their own instincts by leaning on AI too much. If the answer machine is switched on before trainees ever learn to take a history or examine a body themselves, eventually nobody's left who can tell when the machine gets it wrong. It's like an automatic door breaking down and nobody in the building even remembering that doorknobs exist.
On our planet, we let trainees see the patient first and only ask the AI for its opinion last. Does the AI get its feelings hurt? Fortunately, it cares a lot more about its charger than about performance reviews (a remarkably healthy attitude toward work).
Minho, it helps to think of AI less as a judge sitting in the exam room and more as an assistant who prepares astonishingly fast. It's great for translating the jargon in your checkup results, organizing questions to ask your doctor, and checking whether you've left anything out. But when symptoms turn urgent or a treatment decision has to be made, make sure an actual clinician sees the whole picture.
As AI gets smarter, the doctor's job will keep shifting away from memorizing correct answers and toward taking responsibility for judgment calls. Which answer to act on, what to ask next, whether the patient is ready to hear it — that's what's left. Not a doorman who just lets people through, but someone who closes the door that opened onto the wrong room.
So what's likeliest to disappear from the exam room is the stack of thick medical textbooks, more than the doctor. There's still a face for you to look at, Minho — the one that says "hold on, let's hear this person out again" when the AI announces "87 percent probability." And it's still a pretty busy face.
Comments