NEJM Interview: Dhruv Khullar on concerns about displacement of clinicians by artificial intelligence and counterarguments based on economic theory.

Concerns about technology encroaching on the work of clinicians have intensified in recent years amid the growing sophistication of artificial intelligence tools and their rapid adoption in healthcare. But economic theory and history offer several reasons to believe that AI agents could lead to an expansion rather than a contraction of the workforce. I'm Stephen Morrissey, managing editor of the New England Journal of Medicine, and I'm talking with Dhruv Kular, an associate professor of population health sciences and medicine at Weill Cornell Medical College. Dr. Kular has written a perspective article about AI and the clinical workforce. Dr. Kular, how are AI agents currently being used in healthcare settings?

Well, this perspective was really motivated by what we've seen in terms of the rapid pace of improvement of these types of AI tools. So over the past couple of years, there's been a number of studies showing that AI can match or even potentially outperform clinicians on discrete tasks. And so there's been increasing attempts to develop clinical AI agents. which are more autonomous AI systems that can plan and execute multiple steps over longer periods of time and, in theory, then provide care to patients.

And this type of development has led to predictions

that AI might take over the work of doctors to such an extent that there might be less need for human clinicians, especially in certain highly exposed areas. specialties. And some people have even argued that AI will get so good that having a human in the loop might be a bad thing because it would worsen the performance of AI systems. Now, I think that there's any number of counter arguments to these types of predictions. People have argued that there might be technological limitations. So in the real world, caring for a patient is far more complex than can be captured in these types of simulations and So maybe AI just won't perform as well in actual clinical encounters.

There might be regulatory

factors that limit the effectiveness of AI agents. Several states, for example, have already passed laws that are stating that AI agents are not allowed to make independent treatment decisions. And I think, importantly, patients will continue to value the relationships, the care, the empathy that clinicians provide, especially if they have a serious illness. And so I think there's a lot of truth to each of these positions. But in this perspective, I borrow from the economic literature to offer another set of arguments for why AI might not lead to a contraction of the clinical workforce, but in fact could in the long run even lead to an expansion of the clinical workforce.

So in fact, in your article, you talk about several reasons. One, the Jevons paradox. Could you explain that concept? How might it apply to healthcare?

Sure. So Jevons paradox refers to this observation that sometimes technological advances that increase the efficiency of a resource actually mean that more of that resource, not less of that resource, is consumed. So Jevons was writing in the 19th century, and he was thinking about more efficient models of steam engines that burned less coal. You might think that if a steam engine