We’ve all seen the hype.
AI will revolutionize well being care. It’s going to lower documentation time. Enhance diagnoses. Save lives. Perhaps even change medical doctors.
However right here’s what I do know after 11 years as a hospitalist: Hype with out proof is harmful. And AI—particularly in medication—isn’t simply software program. It’s remedy.
If we’re going to let AI affect life-or-death choices, it wants to fulfill the identical commonplace as any scientific intervention. Meaning rigorous trials, clear design, and cultural alignment. Something much less is malpractice.
We’ve been right here earlier than. Bear in mind Theranos? A blinding promise, no peer-reviewed proof, and the medical world’s worst-kept secret. It didn’t simply waste cash—it risked lives. If we deal with AI the identical means—rolling out instruments with out proof, accountability, or ethics—we’re asking for an additional catastrophe.
Medical AI have to be validated like every drug or machine. Randomized managed trials aren’t non-compulsory—they’re important. Dr. David Byrne calls this the “secret sauce” for protected AI implementation, and he’s proper. We’d by no means let a brand new chemotherapy hit the market primarily based on a superb pitch deck and a few retrospective information. So why are we doing that with algorithms?
And but, it’s occurring. Instruments are being deployed with out explainability. With out understanding the info they have been skilled on. With out understanding how they’ll behave in several populations. That’s not innovation—it’s irresponsibility.
Physicians aren’t the enemy of progress. However we’re skeptics for a cause. Skepticism protects sufferers. It’s why we double-check vitals, query assumptions, and push again on protocols that don’t really feel proper. If we’re sluggish to undertake AI, it’s not as a result of we’re resistant. It’s as a result of we keep in mind what occurs when techniques overpromise and underdeliver.
That skepticism will solely develop if we proceed to deal with physicians as implementation obstacles as a substitute of companions. If AI is to reach well being care, it have to be constructed round clinician belief. That begins with schooling. Our colleagues received’t belief a instrument they don’t perceive—nor ought to they.
We want AI literacy woven into coaching packages, hospital onboarding, and government discussions. We want frameworks that guarantee moral and clinically sound growth—like SPIRIT-AI and CONSORT-AI—baked into deployment plans. And we want each chief to grasp that an AI rollout isn’t just an IT mission. It’s a scientific intervention that deserves the identical scrutiny, the identical rigor, and the identical humility.
Simply because one thing’s new doesn’t imply it’s good. In Silicon Valley, pace is a advantage. In medication, security is. The tech world assessments concepts on customers. We take a look at interventions on sufferers. One misstep in a consumer interface might frustrate a buyer. One misstep in medication can price a life.
And right here’s the actual irony.
Physicians need AI to work. We’re bored with clunky EMRs. We would like our notes dictated sooner, our sufferers flagged earlier, our discharges smoother. However what we concern is unhealthy change—change with out proof, implementation with out governance, and know-how that provides burden as a substitute of eradicating it.
We can’t afford to spend thousands and thousands on shiny AI dashboards whereas our EHRs nonetheless frustrate fundamental care. Or roll out “good” triage instruments whereas ignoring the bias of their coaching information. Earlier than we launch AI-powered ambulances, let’s be certain we are able to belief the software program that predicts readmissions.
Physicians don’t concern innovation. We concern irresponsibility.
That’s why it’s time to flip the script. AI shouldn’t be an adjunct—it’s changing into a part of the care plan. And if we settle for that, then it have to be evaluated, regulated, and revered the best way we consider all the pieces else we give to our sufferers.
We have to begin treating AI like chemotherapy.
Not as a result of it’s poisonous—however as a result of it’s highly effective. As a result of it requires precision, vigilance, and consent. As a result of it have to be protected earlier than it’s scaled. And since if we get it unsuitable, the results are too nice.
AI isn’t the way forward for well being care. It’s the current. However it should solely succeed if we construct it on the inspiration that medication was at all times meant to face on: belief, reality, and proof.
Rafael Rolon Rivera is an inside medication doctor.
