.png)
In 2016, Geoffrey Hinton—widely known as the “Godfather of AI”—delivered a stark warning at a machine learning conference in Toronto. He declared that artificial intelligence would soon make radiologists obsolete, urging institutions to stop training new ones because deep learning would outperform humans at reading medical scans within five years (or ten at the latest).
“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 famously said.
Ten years later, the opposite has happened. The average salary for a radiologist reached **$571,000 in 2025**, up 9% year-over-year, according to Medscape data. The U.S. has seen roughly a 10% increase in active radiologists over the past decade, yet a significant shortage persists. As of March 2026, there were about 4,333 open radiologist positions, with an average fill time of 130 days.
Why the Prediction Missed the Mark
Hinton himself has walked back parts of his earlier statement. In 2025, he clarified that he was referring specifically to image analysis tasks. He now envisions radiologists working alongside AI to become more efficient and accurate.
Several factors explain radiology’s resilience:
- **Broader job scope**: Reading scans and writing reports is only part of the role. Radiologists consult with physicians, monitor patients, and interventional radiologists perform procedures. Automating one task often allows more time for higher-value work.
- **Regulatory and liability barriers**: Medicare and Medicaid require a licensed physician for final reads. Questions remain about AI accountability in missed diagnoses.
- **Rising demand**: An aging population, expanded insurance coverage, and AI-enabled tools that make imaging cheaper and faster have driven a 25% surge in radiology case loads from 2018 to early 2025.
Tech leaders have echoed this shift in perspective. NVIDIA CEO Jensen Huang noted that critics often confuse reading scans with the full scope of a radiologist’s job. Netflix co-founder Reed Hastings highlighted radiology as a case where AI affects—but does not fully replace—professions.
AI as a Tool, Not a Replacement
Dr. Jeff Chang, a former ER radiologist and co-founder of RadAI, experienced burnout firsthand—reading 150–200 studies per night shift. His company’s AI helps by auto-generating report conclusions, saving nearly an hour per shift. Yet Chang argues the idea of full replacement “didn’t really make sense to begin with.”
Practicing radiologists emphasize the irreplaceable human elements. Dr. Tonie Reincke, a Texas-based interventional radiologist, points out that AI cannot offer compassion, empathy, or simple human gestures like holding a patient’s hand.
Lessons for the Future of Work
Economist Christoph Herpfer argues that radiology offers a broader cautionary tale about AI job predictions. Similar fears in the 1990s about accountants being replaced by spreadsheets proved overstated—software handled routine tasks, freeing professionals for more complex advisory roles.
“Complex jobs like being a doctor consist of many sub-tasks,” Herpfer said. “Even if you can automate one or two of those, you just expand the time you spend on the other tasks. Until AI is fully able to do the entirety of all the tasks, the job itself won’t go away.”
As long as we avoid a sudden leap to artificial general intelligence (AGI), many skilled professions are likely safer than early doomsday forecasts suggested. Radiology’s story shows how AI can augment human work, boost productivity, and meet growing demand rather than simply eliminate jobs.
The hype around total replacement may even risk worsening shortages by discouraging students from entering competitive, lengthy training programs like radiology residencies.
In the end, the coyote is still very much on solid ground—and earning more than ever.