Could These “Robot Doctors” Put Your Life At Risk?
You get a CT scan. Maybe your doctor suspects a tumor…A blood clot…Or a dangerous blockage in your heart or brain.
The scan is sent to a specialist called a radiologist, a physician trained to analyze medical images and detect life-threatening problems hidden inside the body.
Radiologists spend more than a decade in training learning how to recognize subtle patterns that could mean cancer, stroke, aneurysm, or internal bleeding.
But now…the CEO of America’s largest public hospital system just announced he wants to replace radiologists with artificial intelligence as soon as regulations allow it.
In other words…The next time you get an X-ray or mammogram…
A robot algorithm might be the first, or only, “expert” interpreting your results.
And if that makes you uneasy… well, it should.
Because these AI models have already made some terrifying mistakes… things you were never supposed to know about.
Radiology isn’t just reading pictures. It’s detective work inside the human body.
A faint shadow on a lung image could mean:
- Early lung cancer
- Pneumonia
- A benign scar
- Or nothing at all
Missing the difference can have enormous consequences.
Radiologists are trained to recognize thousands of subtle clues and to combine imaging with clinical context, medical history, and experience.
AI systems don’t actually “understand” any of that. They’re trained on massive datasets of images and taught to recognize patterns.
Sometimes they do it well. But sometimes they don’t.
And when mistakes happen, the consequences aren’t a typo in an email…
They’re missed tumors, delayed strokes, or unnecessary surgeries.
In fact, radiologists have already warned that relying on AI alone to read scans could put patients at risk.
One imaging specialist responded bluntly to the hospital CEO’s remarks, saying attempts to implement AI-only readings would likely result in patient harm.
Mistakes from medical AI aren’t theoretical.
A Reuters investigation found an AI-assisted surgical guidance system used in sinus operations was linked to more than 100 malfunctions and at least 10 patient injuries reported to the FDA.
This included strokes after the software misidentified where surgical instruments were inside the patient’s head.
In another incident, a medical AI system actually diagnosed a patient with damage to the “basilar ganglia.”
The problem? That structure doesn’t exist. Experts later said the AI had essentially invented a body part.
Researchers have also found that some diagnostic AI systems show major bias, including melanoma detection algorithms that produced 28% more missed cancers in darker-skinned patients.
Here’s the uncomfortable truth.
Radiologists are expensive. Hospitals are under pressure to cut costs. And artificial intelligence promises something administrators love: Automation.
If a computer can read scans, hospitals could process more images with fewer specialists.
That may look efficient on a spreadsheet. But when it comes to life-and-death diagnoses, efficiency should never replace expertise.
Artificial intelligence will likely become a bigger part of healthcare. But patients should never become passive passengers in that transition.
Here are five ways to protect yourself:
- Ask if AI is involved in your scan results.
Some hospitals already use algorithms to assist in reading mammograms, CT scans, and X-rays. - Don’t accept “AI approved” as the final answer.
Always confirm that a qualified physician reviewed your results. - Request human oversight.
For serious conditions like cancer, stroke, or heart disease, insist on expert interpretation. - Ask questions about unusual findings.
Doctors should be able to explain results, not just print a computer report. - Treat AI as a tool, not the authority.
Technology can assist doctors. But it should never replace medical judgment.
To protecting the human side of medicine,
Ray Thatcher
Research Director, Health Sciences Institute
Sources:
- Li, Y., Yi, X., Fu, J., Yang, Y., Duan, C., & Wang, J. (2025). Reducing misdiagnosis in AI-driven medical diagnostics: a multidimensional framework for technical, ethical, and policy solutions. Frontiers in medicine, 12, 1594450. https://doi.org/10.3389/fmed.2025.1594450
- Ullah, W. (2026, April 3). Replace yourself with AI, CEO Katz — Let’s test out this new tech where the downside is a bad spreadsheet, not a missed cancer. MedPage Today. https://www.medpagetoday.com/opinion/second-opinions/120627
- Field, H. (2025, August 4). Google’s healthcare AI made up a body part — what happens when doctors don’t notice? The Verge. https://www.theverge.com/health/718049/google-med-gemini-basilar-ganglia-paper-typo-hallucination
