In 2016, AI pioneer Geoffrey Hinton suggested that training radiologists should cease, predicting that deep learning would outperform human specialists within five years. Currently, the FDA has approved over 1,000 AI tools for radiology, some demonstrating greater accuracy than radiologists. Despite this, the demand for human radiologists has increased, with a 17 percent rise in their numbers since 2016, high vacancy rates, and an increase in average salaries from approximately $350,000 to $570,000, ranking radiology as the third-highest-paid medical specialty in the U.S.
Concerns about AI rendering numerous careers obsolete have been voiced, including by Anthropic CEO Dario Amodei, who claimed that AI could eliminate half of all entry-level white-collar jobs. However, the situation is complex. The susceptibility of a job to AI replacement can be assessed by three questions.
The first question is whether a job is a weak or strong bundle. Luis Garicano, an economist, explains that white-collar jobs often consist of clean tasks, which are predictable and data-driven, and messy tasks, which are unpredictable and require interpersonal skills. Jobs that can easily separate clean tasks from messy ones are more likely to be affected by AI. For example, a trial lawyer's job is a strong bundle, as the tasks are interconnected, making it counterproductive to delegate them to AI. In contrast, a recruiter’s job, which involves significant clean tasks like résumé screening, can be enhanced by AI without diminishing the overall job.
The second question addresses how much demand would increase if the cost of a product or service decreased due to automation. Historical examples, such as the automobile industry, show that automation can lead to increased demand and employment, a phenomenon known as the Jevons paradox. Job openings for recruiters and software engineers have risen, even as AI is utilized in their fields, suggesting that AI may stimulate demand for these roles.
The third question examines whether AI or the worker possesses more expertise. The impact of technology on professions can vary significantly; for instance, while accounting clerks have seen their roles diminish due to automation, inventory clerks have experienced wage decreases and job growth due to the loss of their expert skills. The interaction between technology and expertise is crucial in determining job security.
In the case of radiology, the combination of clean and messy tasks, along with the need for specialized knowledge, has resulted in sustained demand for human radiologists despite AI advancements. The article concludes by reflecting on the unpredictability of technological changes and their effects on various professions, including journalism, where the balance of clean and messy tasks remains uncertain.