Decreased costs of computing power and virtually infinite cloud storage capacity have created a fertile environment for artificial intelligence (AI) to disrupt industries across the globe. Computers won’t replace radiologists in the next 10, 20 or 30 years, but I do believe increasingly large parts of the job will be automated—and it may be up to radiologists to carve out space for themselves.

Diagnostic uses of AI technology will likely follow computer-assisted detection’s example and boost the efficiency of radiologists. Computer vision algorithms could identify exams with critical findings and push them to the top of the stack, prioritizing cases that need immediate attention. Other productivity-boosting technologies might include deep learning modules that comb patient records for relevant information, seamlessly presenting valuable context to the radiologist.

Perhaps the most promising possibility is the ability for AI technology to discover new associations and independently develop new ways to diagnose patients, as explored in this Science article. However, this also opens up what is commonly referred to as “the black box problem”: If deep learning can identify a lung nodule, it can’t tell us why it identified it. It doesn’t use prescribed rules. It teaches itself how to differentiate tumors from healthy tissue, and it’s nearly impossible to unravel how it arrived at its conclusion.

Looking into that black box is exceedingly challenging, but it’s not entirely necessary. Some people may be uneasy with trusting algorithms we can’t really explain, but there will always be oversight—the medical community would never be comfortable with fully automated diagnoses. Also, the World Health Organization predicts a drastic shortage of physicians by the end of this century, and AI is one way we could improve efficiency among current physicians. However, it’s also possible we could see some duties formerly reserved for radiologists farmed out to other clinical staff. Think nurse practitioners.

Increasing specialization caused a shortage of primary care physicians in the early 1960s, especially in rural areas, and the advent of advanced clinical staff shored up the staffing deficit while controlling cost. Deep leaning algorithms that read as accurately as radiologists could allow for a similar transition, forcing radiologists to adjust their role.

I could foresee radiologists of the future moving out of the orchestra pit and up to the conductor stand, doing less direct reading of images and more overseeing an ecosystem of algorithms and image interpreters—signing off on correct diagnoses and investigating those that don’t seem quite right. This future would require a new set of skills, but radiologists would be more than up to the challenge.

This transition may prove challenging for the imaging industry, and it may require a lot of significant changes, but there are no signs that the radiologist will just fade out of existence. Radiology may evolve radically, but there will always be a need for radiologists.