The medical perils of unmitigated AI
I’m writing in response to a May 8 Valley News article, “AI is reshaping how Vt. doctors care for patients.” As an old guy with an age-appropriate panoply of ailments, and the history to go with them, I’ve had lots of opportunities over the last 10 years or so to read medical notes in my growing chronicle of consultations and procedures. This has made for some fascinating reading.
For example, I learned from one report that I had been unable to move my arm above a certain level when, in fact, I’d never been asked to move a muscle during the appointment. Or: Some after-visit notes recommended high doses of a particular medication that the provider had specifically agreed, during the appointment, that I should not take. Or: The reading of some crucial imaging commented on the current state of my appendix. The wonders of modern medicine, given that it had been removed 65 years before! Or: Some surgical notes indicated that a certain procedure been executed “to good effect,” when in fact it hadn’t been done at all.
In each of these instances, an otherwise dedicated but overburdened provider had adopted without change some pre-fabricated verbiage or a report template designed around the general case at hand.
It will indeed constitute progress if AI can help compose and organize medical notes and records, providing essential relief to those working in our over-stressed system. But that will only be tolerable if there is an active, critical human in the loop, to assure that the particulars that result correspond to reality.