The Physician Who Became a Patient and What It Taught Him About AI in Healthcare
When the Doctor Becomes the Patient
Barry Chaiken, MD, MPH, has spent decades in healthcare IT. He was the Board Chair of HIMSS. He has raised money for the Dana-Farber Cancer Institute through 200-mile charity bike rides. And when he was diagnosed with prostate cancer, he almost got treated at a community hospital without a second opinion.
“When you are sick, you do not think rationally as you would as a clinician,” Chaiken told us on the latest episode of Practical AI in Healthcare. “You think emotionally and you don’t think about things in a clear way.”
It took a friend asking three words to change his trajectory: “What are you doing?”
That experience, followed by a second cancer diagnosis (Waldenstrom’s lymphoma), gave Chaiken a perspective that most AI in healthcare commentators lack: he knows what it feels like to be a frightened patient navigating a system he helped build.
The 95% vs. 37% Problem: AI Diagnostic Accuracy
A Nature Medicine study found that when physicians used an AI diagnostic tool, they achieved 95% accuracy. When patients used the same tool conversationally, that number dropped to 37%. The recommendations ranged from “visit the ER immediately” to “just go to sleep and take an aspirin.”
The tool was the same. The difference was entirely in what the user brought to the interaction. Chaiken compares it to the DaVinci surgical robot: a transformative instrument in trained hands, useless (or worse) without the knowledge to operate it.
A Framework That Works
Chaiken’s practical advice for patients is disarmingly simple: don’t use AI to diagnose yourself. Use it to prepare questions for your doctor.
“I don’t want you to give me a diagnosis and I don’t want you to give me a treatment,” he instructs the AI. “I want you to help me understand what the disease I have and what the results mean per test, and what questions I need to ask my clinician.”
This reframe turns AI from a substitute for clinical judgment into a bridge between the patient and their care team. The patient arrives educated and engaged. AI becomes a tool for patient engagement. The clinician gets relevant questions instead of anxiety-driven misinformation.
For clinicians, Chaiken recommends OpenEvidence, an NPI-gated tool that links every recommendation to its source literature. His use case extends beyond his own specialty: he used it to research antibiotics for his wife’s dental complications and to generate patient information sheets for friends facing cancer diagnoses.
The Automation Bias in Healthcare Trap
Chaiken warns about the flip side of high-accuracy AI. When a system is right 99% of the time, clinicians stop checking. His traffic light analogy captures this: if red and green lights swapped positions at an intersection, most drivers would run the red because their brain responds to position, not color.
Healthcare has seen this before. Early EHR implementations had pharmacy departments build medication pick lists organized by pharmacist logic rather than clinical workflow. Doctors scrolled past the right medication and selected the wrong one. “That’s a workflow problem,” Chaiken says. “It’s not a doctor problem, it’s an implementation problem, and I can see the same thing happening in AI.”
From Aviation to Healthcare
Chaiken’s governance proposal draws on the Aviation Safety Reporting System (ASRS), the anonymous, non-punitive reporting system that transformed air travel from dangerous to near-perfect. He argues healthcare AI needs the same: a centralized mechanism for sharing what works, what fails, and how to build workflows that account for human error.
“The current system is trying to distribute the management of AI to all these organizations who are working to discover the same exact fact, and there’s no easy way to share that information with each other.”
His complementary point about interoperability is equally blunt: “If we wanted interoperability, we’d have it.” The ATM analogy makes it concrete: banks turned ATMs from cost centers to profit centers by building interoperable networks and charging transaction fees. Healthcare hasn’t found its equivalent incentive structure.
The Sacred Trust
Perhaps the most resonant moment of the episode comes when Chaiken describes the implicit trust patients place in healthcare organizations: “You walk into a hospital, you walk into a doctor’s office, and you trust every single person that you interact with in that place, from the doctor to the janitor, is focused on you and cares about you. That trust is a sacred trust.”
AI tools that send patients spiraling through misinformation don’t just fail technically. They erode something that took decades to build.
Listen to the full episode – Barry P. Chaiken, MD, MPH: Physician-as-Patient Perspective on AI in Healthcare
You Might Also Enjoy
- S1E24: Bob Wachter on AI in Medicine with Bob Wachter — A leading physician-author’s perspective on where AI fits in clinical practice, with hard questions about implementation timelines.
- S1E28: CancerBot and Clinical Trial Matching with Adam Blum — How AI is already matching cancer patients to clinical trials, directly relevant to Chaiken’s vision for EHR-based trial recruitment.
- S1E30: Patient Advocacy and the Dartmouth Model with Amy Price — The patient perspective on navigating health systems, paired with Chaiken’s physician-patient lens for a complete picture.