In 2016, the “Godfather of AI,” Geoffrey Hinton, stood onstage at a machine learning conference in Toronto and declared AI would soon kill the radiology profession.
At the time, he said people should even stop training new radiologists because it was “completely obvious” that within five years (or 10 at most) AI would do a better job than humans at the same tasks.
“If you work as a radiologist, you’re like the coyote that’s already over the edge of the cliff but hasn’t yet looked down,” Hinton said.
For years, tech experts like Hinton predicted displacement by AI partly because some radiologists’ tasks are seemingly formulaic and repetitive, such as reading scans and writing reports.
And yet, despite the doomsday predictions, radiology may serve as an example that the warnings of AI-fueled job replacement may be oversold. Even Hinton stepped back from his drastic call last year, clarifying that he was speaking purely about image analysis, the New York Times reported. In the future, he said human radiologists will work with AI to be even more efficient and effective.
Over the last 10 years, the number of active radiologists in the U.S. has grown by about 10%, said Christoph Herpfer, an economist and business administration professor at the University of Virginia’s Darden School of Business who studies healthcare finance and physician labor markets.
“We actually have a huge shortage of radiologists. So the exact opposite of this prediction has happened,” he told Fortune.
To be sure, demand for healthcare jobs and health services overall has increased steadily as Americans have gotten older and more people have obtained health insurance under the Affordable Care Act.
But much has been said about AI replacing human jobs. Some tech companies, including Snap and Block, have pointed to AI as a reason for laying off thousands of employees, while Anthropic CEO Dario Amodei has warned AI could eliminate half of all entry-level white-collar jobs within five years.
Still, there were about 4,333 active job listings for radiologists as of March, with a 130-day average to fill each position, according to a report by radboard.io, a job board for radiologists. This shortage has pushed the average salary for radiologists to $571,000 as of 2025, up 9% year over year, according to a study from Medscape.
Whereas years ago AI experts declared the death of radiology as a career, some tech leaders have recently changed their tone.
Nvidia CEO Jensen Huang said on the Dwarkesh Podcast last month that radiology doomers conflate the task of reading scans with the entire job. Netflix cofounder Reed Hastings said on the Possible podcast late last month that radiologists are an example of how even if jobs are affected by AI, they may not be replaced entirely.
“We’re drawn to these scenarios of AI wiping out things,” he said. “And again, it hasn’t happened in radiology. Maybe it will eventually, but not in the last five years.”
Structural issues may make it nearly impossible for AI to fully replace radiologists anytime soon. Medicare and Medicaid will only reimburse for a radiology study if a licensed physician performs the final read. It’s also unclear how AI can be held responsible for a missed diagnosis.
Like Huang, Herpfer pointed out that another reason radiology persists is because reading images is just one of their daily tasks. Radiologists also consult with other physicians, monitor patients, and some, like interventional radiologists, also perform hands-on procedures. Introduce AI automation into scan-reading and report writing, and doctors simply dedicate more time to their other tasks, Herpfer argued.
“Complex jobs like being a doctor consist of many sub-tasks. Even if you can automate one or two of those, you just expand the time you spend on the other tasks,” he said. “Until AI is fully able to do the entirety of all the tasks, the job itself won’t go away.”
Increased demand for scans, fueled partly by AI tools already approved by the Food and Drug Administration that have made imaging cheaper and faster to produce, has also kept radiologists busy. Between 2018 and early 2025, radiology case loads skyrocketed 25%, according to the Journal of the American College of Radiology.
“The number of studies that a radiologist has to do every year keeps going up, the reimbursement keeps going down, and they just get really, really burnt out,” Dr. Jeff Chang, a former practicing ER radiologist who cofounded Rad AI, told Fortune.
Chang knows first hand how burnt out radiologists can get. He spent a decade reading between 150 and 200 imaging studies during each of his night shifts before starting his company in 2018. His company’s AI tools are designed not to replace radiologists but to save them close to an hour per shift by automatically generating the conclusions section of radiology reports.
Despite being the cofounder of a company that introduces AI into radiologists’ workflow, he doesn’t believe the technology can fully replace the profession.
“That entire concept didn’t really make sense to begin with,” Chang said.
On the ground, practicing radiologists say AI cannot simulate the humanity involved in their jobs. Dr. Tonie Reincke, a Texas-based interventional radiologist, said AI can’t offer compassion, empathy, or the nonverbal cues that humans give while speaking to patients.
“A computer can’t hold a patient’s hand when they’re crying,” she told Fortune. “A computer can’t hand them a tissue.”
Reincke worries the hype about AI replacing radiologists entirely may discourage medical students. Radiology is one of the most competitive programs and requires one of the longest residencies in American medicine, often lasting between five or six years depending on the specialization, she said. AI fears, even if they’re not well founded, may worsen the shortage of radiologists, she added.
Herpfer said the broader lesson from radiology extends well beyond medicine. The same dynamic played out with accountants, who were expected to be wiped out by spreadsheet software in the 1990s. Instead, Excel eliminated their routine number-crunching tasks and freed up accountants to move into more complex advisory work.
“As long as AI doesn’t make this quantum leap of becoming sort of AGI [Artificial General Intelligence], as long as this extreme scenario doesn’t happen, most jobs in the medium run are probably going to be reasonably safe,” Herpfer said. “That’s the lesson I think we can learn from the radiologists.”
