Keeping the Clinician in the Loop
Ambient AI can give clinicians back the most precious resource in medicine, time. It works only if the human stays the final authority on the note.
We are living through a rare chance to give medicine back its most precious resource, time. For every hour a clinician spends with a patient, research has found that nearly two more disappear into the electronic health record, into the clicks and the checkboxes and the after-hours "pajama time" that has become a hallmark of modern practice and one of the great engines of burnout. Ambient AI documentation offers to reverse that. Let the machine listen and draft, and give the human back the minutes that were taken.
Optimism is not abdication
I share that optimism without reservation. But optimism is not the same as walking away, and as these systems grow more capable, one principle has to be held firmly in place. The clinician stays in the loop.
Why the human stays
The reason is fundamental. A large language model does not understand a patient. It predicts the next word. Most of the time the draft it produces is remarkably good. Every so often it confabulates, producing a medication that was never ordered, a dose that is quietly wrong, or a symptom no one mentioned. And the output arrives fluent, confident, and beautifully formatted, which is exactly what makes those errors dangerous. Fluency is not accuracy. A polished note can lull even a seasoned physician into signing what they never truly read.
Automation bias
That is automation bias, and it strikes hardest at the worst possible moment, late in a long clinic day, when the draft looks complete and the inbox is full. The answer is not to slow clinicians down. It is to design for real review. Place the AI's suggested codes beside the note so both are checked at once, make every edit visible so that a colleague, or an auditor, can see what changed, and never let the mere absence of an edit stand in for verification. Silence is not sign-off.
Deep medicine
This is the heart of what Eric Topol calls deep medicine. The point of lifting the keyboard off the clinician was never to slide an algorithm between doctor and patient. It was to remove the one that was already there. Used well, an AI scribe handles the transcription and the first-pass assembly so the human can do what no model can, bringing judgment, context, accountability, and above all presence to the record and to the person across the room.
The technology will keep improving. The drafts will keep getting better. But the signature at the bottom of a clinical note is a human act, and it has to remain one. Give clinicians the gift of time. And trust them, still, to be the ones who decide what is true.