Three Ways to Assess AI's Threat to Jobs

The Atlantic published "Three Ways to Think About AI and Jobs" on June 11, 2026, revisiting Geoffrey Hinton's 2016 remark that "people should stop training radiologists now" because deep learning would soon outperform them at reading scans. A decade later, The Atlantic reports radiologist pay has risen from about $350,000 to roughly $570,000, making radiology the third-highest-paid U.S. medical specialty, with vacancy rates it describes as near all-time highs. A separate Fortune report from May 4, 2026 puts 2025 average radiologist compensation at $571,000 (citing Medscape data, up 9% year over year) and cites University of Virginia economist Christoph Herpfer's estimate that the active radiologist workforce grew about 10% over the past decade, somewhat lower than The Atlantic's reported 17% since 2016; both accounts agree on the direction, growth and higher pay, which is the opposite of Hinton's prediction. The Atlantic pairs this with Dario Amodei's May 2025 warning, made to Axios, that AI could eliminate half of all entry-level white-collar jobs within five years.

Fortune's reporting fills in why: Medicare and Medicaid reimbursement generally requires a licensed physician's final read, radiologists spend much of their time on tasks beyond image interpretation such as consulting with other physicians, monitoring patients, and performing procedures, and imaging case volumes rose an estimated 25% between 2018 and early 2025 per the Journal of the American College of Radiology. AI-assisted efficiency gains appear to have been absorbed by rising demand rather than shrinking headcount. Nvidia's Jensen Huang and Netflix cofounder Reed Hastings have both cited radiology publicly as an example of AI changing a job's task mix without eliminating the role.

The Atlantic's diagnostic questions are the transferable part beyond this specific case: instead of asking only whether AI can perform a task, ask how reimbursement or regulatory rules assign liability for the final decision, how the job's sub-tasks are bundled (automating one sub-task can free time for others rather than cut headcount), and whether automation is likely to expand overall demand for the service. Applying that lens to Amodei's white-collar-jobs prediction suggests the same structural questions, not just model capability, will determine the outcome.

Track whether radiologist vacancy and compensation data continues to diverge from AI-capability headlines, how reimbursement policy for AI-assisted reads evolves, and whether Amodei's five-year entry-level-jobs prediction, which points to roughly 2030, plays out closer to the radiology pattern or the more severe scenario he described.