✨ 😷 AI and the radiologist apocalypse that wasn't

AI was supposed to replace medical imaging experts. Instead, they're working together – revealing crucial lessons about technology's true economic impact

May 15, 2025

My fellow pro-growth/progress/abundance Up Wingers,

Here’s another great example of "predictions are hard especially about the future, " artificial-intelligence edition. In an eye-opening (for some) New York Times story, “Your A.I. Radiologist Will Not Be With You Soon, ” reporter Steve Lohr recounts AI pioneer Geoffrey Hinton's 2016 prediction that radiologists would become obsolete within five years.

Yet the opposite has proven true, according to the story. At the Mayo Clinic, where radiologist numbers have jumped 55 percent since then, AI has become a powerful assistant rather than a job killer. The technology handles routine tasks while doctors retain decision-making authority. Though AI excels at spotting specific abnormalities, it can't replace radiologists' comprehensive role, such as consulting with other physicians, talking with patients, and applying experienced judgment.

From the piece:

Radiologists do far more than study images. They advise other doctors and surgeons, talk to patients, write reports and analyze medical records. After identifying a suspect cluster of tissue in an organ, they interpret what it might mean for an individual patient with a particular medical history, tapping years of experience.

Economic signals across industries

The radiologist apocalypse that wasn't brings together some important economic themes here in the emerging Age of AI:

✨ First, note that staff numbers have risen even as more than 250 algorithms now shoulder the drudgery of measuring kidneys or flagging stray blood clots. Each scan saves specialists some time, yet the ultimate diagnosis remains firmly human-centric. That is precisely the “augmentation, not automation” pattern documented by “Large Language Models, Small Labor Market Effects,” a working paper that I wrote about on Monday.

✨ Second, Mayo’s experience also squares with “Industry-Level Growth, AI Use and the U.S. Postpandemic Recovery,” a new analysis by St. Louis Fed economists Oksana Leukhina and Mickenzie Bass. They find that productivity gains are emerging inside industries that marry abundant brainpower with targeted AI. By plotting industry-level gains against survey data on AI chatbot use, its economists reveal a clear positive relationship. Sector groups like information services and corporate management — those with highly educated workforces and high AI uptake — led the pack. While the analysis stops short of claiming causation, the pattern aligns with the view that AI tools are starting to deliver measurable economic benefits: “It is quite possible the industries that had more to gain from AI tools and had the educated workers who could effectively use them were the ones that experienced higher growth.”

✨ Third, the radiologist story can also be interpreted as supporting the caution expressed in “AI and recent productivity growth,” in which JPMorgan economists Michael Feroli and Abiel Reinhart note that US productivity growth has improved modestly since the pandemic, particularly in sectors like real estate, professional services, and manufacturing. Yet when they compare this trend with three measures of AI diffusion — government surveys, occupational task-exposure data, and usage logs from large language models — they find no clear relationship between industries with higher AI use and those with faster productivity growth.

This could be because the technology is still early in its adoption curve, or because the metrics themselves are noisy. Alternatively, it may suggest that AI's promise is overstated, at least in the short run. JPM: “Until we get more evidence of that, though, our assumption is that growth in nonfarm business productivity will remain below 2% in the near term, and could converge back to rates close to the prior cycle.”

In the case of the Mayo radiologists, shaved minutes from workflows barely register once averaged across a sprawling health-care labor force, let alone a $28 trillion economy. The real macro dividend will come only when thousands of Mayos, in industry after industry, bring their algorithms and changing work patterns to scale.

From investment to innovation

Setting aside different data sources (such as on AI usage) and interpretive approaches, the Productivity J-Curve would seem to offer a tidy macroexplanation for the seemingly contradictory (or at least less-than-fully synced) findings of those three studies. According to the PJ-C, when a general-purpose technology like AI first arrives, firms pour resources into intangible assets — training workers, rewriting code, redesigning processes — that don’t show up in official statistics.

A line diagram illustrating the productivity J-curve, with a y-axis representing measured growth and an x-axis representing years elapsed. The curve begins at the origin on the left, makes a downward hump before returning to the x-axis at the midpoint, and then goes on to make an upward hump on the right side of the graph before leveling out along the horizontal axis. The initial downward hump represents underestimated productivity, with capital and labor going toward intangible technological gains, and the ensuing upward hump represents overestimated productivity, with the intangible gains from prior years showing up in the data. — Why Artificial Intelligence Isn’t Boosting the Economy—Yet

The Danish study, which finds modest productivity gains and no wage growth for white-collar workers, reflects this early, costly phase. JPMorgan’s broader look across industries sees no clear link between AI use and productivity growth — just what one would expect during the trough of the J-curve, when benefits are building but not yet visible. By contrast, the St. Louis Fed finds some sectors — those with high AI uptake and educated workforces— already posting faster productivity growth. Think of them as the savvy front-runners, climbing the upward slope. Rather than conflict, the three studies together sketch a familiar (if messy) trajectory: slow gains now, but the promise of acceleration to come.

One more thing: Google's algorithmic evolution

Beyond task automation and augmentation, the real dream is AI as a novel and reliable generator of ideas and solver of problems. To make that dream a reality, one of the milestones we should probably pass along the way is something like Google's new AlphaEvolve. It’s an AI agent that the company says “evolves algorithms for math and practical applications in computing” via its Gemini models and then systematically evaluates against objective performance metrics — keeping only the top-performing versions as seeds for the next generation of improvements. Through that techno-Darwinian process, AlphaEvolve creates code that humans might never conceive.

Already delivering measurable benefits — reclaiming 0.7 percent of Google's vast computing resources and advancing decades-old mathematical problems — the company claims AlphaEvolve demonstrates how this automated verification system transforms AI from creative but unreliable assistant to practical problem-solver. Google: “We believe AlphaEvolve could be transformative across many more areas such as material science, drug discovery, sustainability and wider technological and business applications.”

Reliable problem-solving, happening across industries economy-wide, is what would result in massive and measurable economic gains. That would be a profitable payoff for all that corporate dough being shoveled at AI infrastructure.

As Congress Does Less, The Courts Are Doing More

Breaking news: The Dispatch has acquired SCOTUSblog, the gold standard in Supreme Court analysis.

Reliable coverage of the Supreme Court has never been more important. With an unbridled executive branch and a Congress eager to surrender its constitutional prerogatives, the federal judiciary is playing an increasingly important role in shaping the country’s direction.

Join 600,000 loyal readers and check out The Dispatch today. No insulting clickbait, no false outrage, no annoying auto-play videos—just reliable journalism that helps you understand the big decisions that will shape our nation’s future.

Faster, Please! readers: Take 25% off a Dispatch membership today

Claim your 25% off membership