As a radiologist, I’ve heard the same question too many times, usually from anxious medical students struggling to decide if they should pursue radiology:
“Is AI going to replace radiologists?”
I understand the concern. I had similar fears during training. But after years of practicing medicine and watching AI promises play out in the real world, I can confidently say this: AI is not replacing radiologists. It’s not even close. And those who claim otherwise either don’t understand our workflow or have something to sell.
Let’s be clear: Radiologists don’t simply “read images.” We are physicians. We study patient charts, review lab values, correlate clinical history, navigate incompatible software systems to pull outside imaging, and synthesize all of that into a report that guides care. We talk to referring doctors, clarify diagnoses, and advise on next steps. It’s a complex, human, and deeply contextual process that cannot be reduced to a code.
Yet some AI vendors speak as if automating this is right around the corner. What they ignore is that no two hospitals are the same. EMRs, imaging platforms, and workflows vary widely. Standardizing all of this to make automation viable would require an extraordinary overhaul of the entire health care system at enormous cost. Who’s paying for that? Not hospitals struggling with reimbursement cuts. Not the vendors selling plug-and-play algorithms.
Even if automation becomes technically feasible in narrow tasks, it would still require massive IT support, a unified infrastructure, and a level of adoption that is, frankly, years if not decades away. The reality is that AI has had minimal real-world impact in health care. We’ve seen plenty of hype, but very little delivery.
What concerns me more than AI itself is how fear of it is deterring bright students from pursuing radiology. That’s a loss not just for our specialty, but for patients. Radiology is intellectually rich, clinically central, and increasingly critical in modern medicine. AI isn’t here to replace us, it’s a tool that, while it may become helpful in the distant future, is nowhere near that point today.
I’ve personally been involved in multiple radiology AI research projects, including some high-profile ones, and I can say with confidence: The data is almost always selectively presented to support a pre-determined narrative. Often, AI is compared to radiologists who are performing the task in their downtime just to check a box for authorship, usually unpaid and not truly invested in meticulously analyzing the images. As someone who takes image interpretation seriously, these comparisons are meaningless. They don’t reflect real clinical practice and certainly don’t represent the standard of care patients deserve.
This leads me to question the real capabilities of some AI vendors. If they were truly focused on supporting radiologists and by extension the patients rather than trying to replace them, progress might be much further along by now. The notion that health care is a monolithic, centralized system ripe for automation shows a fundamental misunderstanding of how medicine works.
Each hospital and clinical environment has its own practices, workflows, and conventions. Surgeons, radiologists, and other specialists collaborate to determine report language and clinical standards. AI vendors who assume they can simply automate these nuanced human processes reveal how disconnected they are from the real-world complexities of health care delivery.
Above all, medicine is the ultimate cornerstone of human connection built on trust, empathy, communication, and judgment. Attempting to automate that is not only deeply misguided but far beyond what any set of algorithms can meaningfully achieve. The essence of patient care cannot be reduced to code.
Fardad Behzadi is a radiologist.
