Experts predicted AI would replace radiologists, but the profession is soaring — even as AI outperforms humans
Made with Midjourney
Computer scientist and Nobel prize winner Geoffrey Hinton predicted a few years ago that AI would make human radiologists obsolete. But demand for radiologists is at record highs in the US today, with salaries soaring to $520,000. The success of the profession continues despite AI models outperforming human radiologists on several benchmarks. But why?
Benchmarks are tricky. While AI models excel in curated datasets, they often struggle in real hospital conditions. Performance can drop 20 percentage points when tested at new sites, and models fail to combine text and image reliably. Recent research from Microsoft shows many medical AI systems still guess diagnoses without images and produce misleading explanations. Clinical reasoning remains elusive.
Regulation and risk slow adoption. Only a handful of AI tools, like IDx-DR, operate autonomously under strict guardrails. Due to liability concerns, insurance limits, and FDA rules, hospitals must rely on human radiologists to review every scan. AI accelerates workflows but cannot replace clinical judgment, patient communication, or the legal responsibility carried by radiologists.
Radiologists do more than just read scans. They consult clinicians, advise patients, and supervise procedures. When AI speeds up interpretation, radiologists often shift toward these higher-value tasks. Historical patterns show faster scans increase demand rather than shrink jobs — efficiency opens new avenues for human labor.