AI and Healthcare Jobs: Why Radiologists Are Thriving Despite Automation Predictions
核心洞察
Despite predictions that AI would replace radiologists, the number of radiologists has risen 17% since 2016, with salaries increasing from $350,000 to $570,000.
The FDA has cleared over 1,000 AI radiology tools (搜索), yet human radiologists remain in high demand due to the "strong bundle" nature of their work combining image interpretation with patient interaction.
The Jevons paradox helps explain this trend: as AI makes scans cheaper, clinicians order more of them, increasing demand for radiologists rather than reducing it.
In 2016, AI pioneer Geoffrey Hinton made a bold prediction: "people should stop training radiologists now" because "it's just completely obvious that within five years, deep learning is going to do better than radiologists." Nearly a decade later, he was only half right. Today, the FDA has approved more than 1,000 AI radiology tools (搜索), some capable of analyzing medical images with greater accuracy than human specialists. Yet radiologists—human ones—are in more demand than ever.
Since 2016, the number of radiologists has risen by 17 percent, the field's vacancy rates are near all-time highs, and the average salary has increased from about $350,000 to $570,000, making radiology the third-highest-paid medical specialty in the United States. This counterintuitive outcome offers a powerful lens through which to understand how AI will reshape healthcare employment.
The Strong Bundle: Why Some Jobs Resist Automation
According to Luis Garicano, an economist and co-author of the forthcoming book Messy Jobs, most white-collar jobs combine two very different kinds of work. "Clean" tasks involve predictable problems, objective standards of success, and abundant written data—the kind of work AI systems handle best. "Messy" tasks, by contrast, involve unpredictable situations, subjective measures of success, tacit knowledge, and navigating complex human relationships.
Radiology exemplifies what Garicano calls a "strong bundle" job, where clean and messy tasks are so tightly linked that delegating some to AI would be counterproductive. Properly interpreting a scan is difficult without intimate knowledge of a patient's medical history, symptoms, and general health—information typically gleaned only by interacting with the patient or their referring physician. The radiologist must also oversee imaging exams, explain results, and make recommendations to clinicians.
This framework extends across healthcare. The Forbes analysis identifies roles most insulated from AI replacement as those requiring "sustained, high-stakes human connection in unpredictable environments." These include registered nurses, surgeons, emergency physicians, paramedics, mental health counselors, midwives, and home health aides—professions where physical presence, procedural skill, and therapeutic relationships remain beyond AI's reach.
The Jevons Paradox in Medicine
A second factor explains radiology's surprising growth: the Jevons paradox, named after the 19th-century British economist William Stanley Jevons, who observed that the steam engine's more efficient use of coal paradoxically increased coal demand. In healthcare, as AI makes radiological scans cheaper, clinicians respond by ordering substantially more of them, increasing demand for radiologists.
This phenomenon is not unique to radiology. Job openings for recruiters rose by 30 percent from 2023 to 2025; for software engineers, they have doubled. "It's not hard to imagine this happening with financial services, with legal services, with health care," Torsten Slok, chief economist at Apollo, told The Atlantic. "As AI makes these services cheaper, people are going to want a lot more of them. And that means employment in those sectors will grow."
Expertise: Enhancement Versus Commodification
The third critical factor is whether AI enhances or commodifies a professional's expertise. MIT economists David Autor and Neil Thompson have documented this pattern across more than 300 occupations over four decades. For accounting clerks, computers replaced their least expert skills—recording transactions and manual calculations—freeing them for more complex analytical work. For inventory clerks, however, computers replaced their most expert skill set—encyclopedic knowledge of warehouse inventory—leaving them with more basic, lower-paid tasks.
"The story is almost never as simple as: We're in a race with machines and machines will win," Autor said. "What matters for a given profession is whether technology enhances a worker's expertise or commodifies that expertise."
In radiology, AI tools have automated portions of image interpretation, but the remaining tasks require high levels of formal training and specialized knowledge. The radiologist's expertise is enhanced, not replaced.
Jobs at Risk and Jobs Emerging
The Forbes analysis identifies healthcare roles most vulnerable to AI displacement: those involving pattern recognition, data processing, or rule-based decision-making. These include human scribes, medical coders, appointment schedulers, front desk receptionists, insurance verification specialists, and certain pharmacy technician functions. AI scribes can now listen to patient encounters and produce structured notes in real time. "Autonomous coding" is already operational for high-volume, lower-complexity billing.
Simultaneously, entirely new roles are emerging at the intersection of clinical knowledge and technology. Clinical AI implementation specialists—professionals who translate between technology teams and clinical stakeholders—command salaries of $70,000 to $100,000 annually. Healthcare AI ethics and governance analysts, who evaluate systems for bias, fairness, and regulatory compliance, earn approximately $141,000. Health AI data scientists with clinical domain expertise average $122,738 per year.
The Path Forward
Adoption of AI in clinical practice is accelerating rapidly. A 2024 American Medical Association survey found that nearly two-thirds of physicians now use at least one AI tool in their practice, up from fewer than 30 percent just three years earlier. To date, the FDA has cleared 1,524 AI-enabled medical devices.
Yet AI-driven layoffs have not materialized in healthcare to the same degree as in other sectors. Health systems face chronic labor shortages in many areas. The professionals who will navigate this transition most successfully, experts suggest, are those who combine clinical excellence with AI literacy—a working understanding of how AI tools function, where they fail, and how to critically evaluate their outputs.
As Autor noted, some of the most dramatic consequences of the AI revolution are guaranteed to be surprising. The radiologist story demonstrates that whether AI replaces a given profession is far from straightforward—and that the most intuitive predictions can prove dramatically wrong.