The Safe and Meaningful Integration of AI with Medicine is Possible

by James Lyons-Weiler, PhD, Popular Rationalism, ©2025

(Nov. 30, 2025) — In 2004, as a young Assistant Professor in the Department of Pathology in the School of Medicine at the University of Pittsburgh, I made a suggestion that, in retrospect, required neither radical technology nor prophetic genius. I proposed that a hospital system—one of the most advanced in the nation, boasting research budgets, satellite campuses, and a fleet of administrators whose titles barely fit on business cards—could do something so basic, so deeply moral, that it shook the institutional spine to its core: it could learn. Not just in theory. Not just in seminars or postmortems. But in practice, in real time, from its own decisions, its own outcomes, its own near-misses and unnoticed recoveries. The proposal was simple. Equip the system with memory. Allow it to reflect. Build a mirror that helped it learn.

The idea did not die because it lacked clarity. It died because it made too much sense. After I sent it to then-CEO Jeff Romoff, I was warned by my division chief, a creature with too many elbows, to never speak of it again, or I would lose my job. The message was unambiguous: anyone proposing a system trained to perform well and improve patient outcomes is a threat.

For years, that warning remained lodged beneath the surface of everything I saw. Each time a preventable error passed without remark, each time a clinician wondered aloud whether there might have been a better path, the silence became its own indictment. Medicine, for all its credentials and rituals, does not know what it does. It does not track. It does not measure. It guesses, defers, and moves on to keep up. It hides its outcomes in abstraction and deflects blame to avoid liability. It destroys every opportunity to improve through real-world learning.

In every other domain that touches complexity—aviation, engineering, even finance—systems learn. When a plane nearly falls from the sky, the event is dissected across continents. Pilots receive private report cards. When a patient crashes post-op, the paperwork is filed, the chart closed, and the data forgotten. The hospital moves forward as though nothing has happened. There is no reckoning, only the performance of one.

Ask any institution: how many of your readmissions last month were preventable? Which of your surgeons consistently produces the best long-term outcomes? How many medication combinations led to harm that could have been anticipated? The answers will not come in numbers. They will come in stories, evasions, or silence with shaking heads. Somewhere right now, the worst-performing kidney transplant specialist in the United States is operating on a patient. That surgeon does not know their standing. Shouldn’t they know? Shouldn’t they want to know? And we tolerate this.

But we should not.

For a long while, the idea existed only in fragments. Whispered among reformers. Sketched in conference margins. But the world changed. The volume of data grew too vast to ignore. Algorithms began diagnosing tumors more accurately than generalists. Pandemic dashboards mapped outbreaks before agencies did. The myth that healthcare could remain artisanal in a computational age began to fray. And the core insight—that a hospital, if allowed to reflect, could become more than a place of procedure—began to take root again.

Total Health Outcome Awareness—THOA, a phrase I first uttered in 2004 — should be designed to offer something medicine never permitted itself to hold: a conscience. Not a judgmental ledger. Not a bureaucratic surveillance tool. A conscience in the truest sense: a capacity for reflection, refinement, and humility. Paired with it are Private Report Cards (PRCs) for physicians and nurses, not the crude metrics that humiliate or penalize, but the gentler clarity of comparison. These reports will show, without shame and in private, how a physician’s decisions align or do not align with outcomes from colleagues facing similar cases based on health outcomes. Days to wellness. Survivorship. Saved lives and limbs. They do not punish. They illuminate.

———————–

Read the rest here.