You want to understand how artificial intelligence could you change jobs? Consider the radiology as a clue.
Radiology has recently become a talking point in the AI race. This was mentioned several times last month by technology executives at the World Economic Forum in Davos as well as in a White House white paper on AI and the economy.
Radiology is far from being the only profession impacted by AI, which is gradually integrating into the work of software engineers, teachers and even plumbersamong many others. If widely adopted, Goldman Sachs estimates that advances in AI could displace 6 to 7% of the American workforce, although technology is also expected to create new jobs.
But radiology the field has become a case study in how AI could enhance, not replace, jobs. Radiology type of work is also ideal for AI assistance, said Dr. Po-Hao Chen, a diagnostic radiology physician at the Cleveland Clinic.
Radiology has a lot of data available for research and AI applications, which need large amounts of data for training. AI can analyze reams of data much faster than human workers, and it is already helping to speed up certain processes in radiology — for example, determine which analyzes require immediate attention.
But human doctors still have to do most of the work – like making diagnoses, physically examining patients and writing reports.. And radiology jobs are expected to grow faster than jobs in other fields, as the field continues to adopt technology.
“(AI) not only does not replace these workers, but it actually increases the amount of work they can do and increases the demand for their services.” » said Jack Karsten, a researcher at Georgetown’s Center for Security and Emerging Technologies. “It’s a bright future that the tech industry can foresee as AI does good for the economy.”
AI is very effective at analyzing images and spotting patterns in data, both of which are essential to radiology. And the field has been digitized for years, meaning there is an abundance of data, according to Chen.
“There are still smaller use cases that are analog, but in the United States, for the most part, every X-ray, every CT (scan), every MRI can be available as zeros and ones,” Chen said.
Today, radiologists are using AI to determine which scans to prioritize, improve image quality and make reports easier to summarize, according to Dr. Chen and two other radiology experts who spoke with CNN.
“It’s something that doesn’t replace anyone, it just makes our work more efficient and more meaningful,” said Dr. Shadpour Demehri, who works in interventional radiology at Johns Hopkins Medicine.
René Vidal, professor of engineering and radiology in the Penn Engineering department at the University of Pennsylvania, sees AI as particularly useful for capturing high-quality MRI scans with fewer measurements. This speeds up the process and allows more patients to be seen in the same amount of time.
Other applications are being explored in research, such as using AI to measure the volume of a tumor or automatically fill out reports, although they are probably still some way off, Vidal said.
AI tools must be approved by the U.S. Food and Drug Administration for medical use, which could take about eight years given the development process and clinical testing, Vidal said. But these approvals are certainly happening: of the 1,357 AI-enabled medical devices Currently with FDA approval1,041 are intended for radiology.
At the same time, radiology professions seem to be developing. The Bureau of Labor Statistics projects that radiology employment will increase 5% between 2024 and 2034, which is higher than the 3% average across all occupations. Data from Indeed provided to CNN also indicates that there were more radiology jobs in 2025 than there were five years ago.
The demand for imaging during the medical diagnostic process, as well as the increasing aging of the population, is likely driving the need for more imaging. radiology services, say radiology experts who spoke to CNN.
But this was not always the case. Geoffrey Hinton, Nobel Prize-winning computer scientist, also referred to as godfather of AI, said in 2016 that “people should stop training radiologists now” because deep learning – a A subset of AI that models the way the human brain learns – would do its job better in five to ten years.
Hinton said in an email to New York Times last year, he spoke too broadly in those 2016 comments.
Demehri remembers there was a sense of anxiety in the radiology field about human roles being replaced by AI around 2015 and 2016. Today, the technology is seen as a “second pair of eyes,” he said.
There are still risks around bias and potential over-reliance on AI, according to Chen. Unlike human radiologists, for example, AI can accurately predict a person’s race based on an X-ray, according to a MIT Study 2022raising concerns about bias in diagnoses.
Chen says he also worries about the temptation to make staffing decisions — like replacing a doctor with a nurse or a subspecialty radiologist with a primary care physician – if AI becomes sufficiently advanced. This may work in some cases, but not for the majority of conditions for which radiology is used, such as detecting cancer or life-threatening infections.
“We need to understand that a lot of the performance of (the) algorithm comes from the fact that the automation result is reviewed by an expert,” he said. “And together, that collaboration, if you will, between the machine and the expert is what makes improvement real.”