The CEO of America's largest public hospital system publicly stated he's prepared to replace radiologists with AI systems. The comments have sparked backlash from medical professionals concerned about job displacement and patient safety—and they mark a turning point in the AI displacement conversation.
This isn't "someday AI might replace jobs." This is a CEO at a major hospital system saying he's ready to fire doctors now. Mitchell Katz, who leads NYC Health + Hospitals, told a healthcare conference that AI radiology systems have reached a level of accuracy where he'd be comfortable replacing human radiologists for routine scans.
The question isn't whether AI can read scans—it can. Multiple studies have shown AI systems matching or exceeding human radiologists on specific tasks like detecting lung nodules or classifying breast tissue. The technology works, at least for well-defined diagnostic tasks with clear imaging protocols.
But whether a hospital executive should be making this call without oversight is another matter entirely. Radiology isn't just pattern matching on images. It's clinical judgment, contextual understanding, communication with patients and referring physicians, and knowing when an unexpected finding requires human attention. AI systems excel at the pattern matching part; they struggle with everything else.
What's particularly concerning is the framing. Katz positioned this as a cost-saving measure—AI is cheaper than radiologists. But radiologists don't just cost money; they provide value through expertise that goes beyond reading scans. When hospital administrators view doctors as line items to be optimized, patient care becomes secondary to budget management.
The medical community response has been fierce. Radiologists point out that AI systems are trained on curated datasets and struggle with edge cases, poor-quality images, or unusual presentations. A human radiologist seeing something unexpected can adjust their approach, request additional imaging, or flag concerns for follow-up. An AI system just outputs a probability score.
Having built technology products, I know the temptation to automate everything. But there's a difference between using AI to augment human expertise and replacing humans entirely. The best radiology departments are already using AI—as a tool to help radiologists work more efficiently, flagging concerning findings for human review. That's augmentation. Firing the radiologists and trusting AI alone is replacement, and it carries risks that administrators may not fully appreciate.
