AI Policy and Radiologists: Redesigning Work

The missed prediction that illuminates AI workforce policy

In 2016, Geoffrey Hinton, Nobel laureate and widely regarded as the godfather of AI, predicted that radiologists would be replaced by machines within five years.

Ten years later, the evidence tells the opposite story. According to the report published by Ars Technica^[1] in August 2026, radiology professionals are growing steadily, with an expected increase of 26% or more over the next thirty years.

The gap between that prediction and these facts defines the challenge that every talent strategy must address. Hinton did capture something real: physicians now have a silicon colleague that matches or exceeds their performance on defined tasks. This convergence turns radiology into a leading indicator for the adoption of expert decision-making systems.

What is really happening to the people in the room

Radiology represents the hotspot of AI in medicine.

In early 2026, approximately three quarters of the 1,400 AI-enabled medical devices approved by the Food and Drug Administration concerned this field. Some tools make physicians more efficient: they draft reports or flag images requiring urgent attention.

Other tools push beyond human performance. They identify anomalies invisible to the naked eye and interpret images with accuracy equal to, and sometimes greater than, that of trained radiologists. An analysis of 43 clinical studies concluded that AI-assisted colonoscopies detect more polyps than conventional ones.

Accuracy has real stakes for real people: average human error rates on diagnostic imaging range between 3% and 5%, amounting to approximately 40 million errors worldwide each year.

This concentration of approvals signals where technological maturity already meets daily practice. Radiology therefore functions as a leading-edge laboratory for every sector that processes large volumes of data and repeatable decisions.

AI augments; design decides

The central lesson goes beyond the statistical comparison between human and machine.

Curtis Langlotz, director of Stanford's Center for Artificial Intelligence in Medicine and Imaging, observes that even an AI more reliable than average will make mistakes that a human professional would avoid. The radiologist therefore takes on an updated role: evaluating every algorithmic decision and identifying the rare cases in which the machine is wrong.

This task requires profound mental reprogramming. The majority of automated decisions will be correct, and it is precisely this reliability that makes searching for the exception so challenging.

Value emerges from the combination of AI's technical precision and people's flexible expertise. Designing this collaboration becomes the real prize.

The leadership gap that policy must close

Tool adoption differs from organizational readiness, and this distinction governs outcomes.

An organization equipped with excellent AI tools and unprepared leadership will achieve worse results than one with mediocre tools and an AI-literate leadership. The share of leaders who declare themselves ready to govern these deployments remains slim, around 7%, according to surveys previously cited in this column.

That number belongs at board level.

The bottleneck concerns those who commission, govern, and evaluate the introduction of AI, rather than those who use it every day. Effective policy starts at the top, and radiology demonstrates the benefit when clinical governance accompanies the tool.

Organizations that close this gap invest in training decision-makers before investing in tools. They equip leaders to interrogate models, read error rates, and set thresholds for human intervention.

Why workflow matters more than the tool

Inserting AI into an unchanged process generates passive supervision.

When people monitor AI output without active interaction, the problem lies in the design of the workflow, not in the technology. The radiology case shows the alternative path: the human role is redesigned around critical evaluation of exceptions and final clinical accountability.

High-performing organizations rewrite the task sequence before deploying the tool.

Technology enters after the process has been redefined, not before. This sequence determines the difference between an investment that frees up expert time and one that parks people in a sterile surveillance role.

The passive supervision phenomenon absorbs hours of expert work in low-value activities. Redesigning the task returns those hours to complex diagnosis, patient interaction, and review of flagged exceptions.

Converting internal talent beats recruiting

Converting people already within the organization systematically outperforms external hiring for AI capability.

Those who already work inside the organization possess context, relationships, and process knowledge. These elements accelerate adoption and improve performance compared to specialists hired in isolation.

Radiology confirms the principle. The experienced professional who learns to collaborate with the algorithm becomes more valuable than the tool in isolation, because they add clinical judgment where the machine offers patterns.

L&D policy should reward this transition with formal reskilling pathways. The cost of this investment is lower than the cost of external acquisition within an eighteen-month horizon.

Internal adoption data collected from large financial organizations show high participation rates when training reaches people who already know the operational context.

What changes for decision-makers

Each senior function draws a distinct priority from this evidence.

• CEO: bring the conversation on organizational readiness, distinct from tool adoption, to the board.
• CHRO: set L&D and organizational design priorities around role redesign.
• CFO: measure the documented ROI of people development against the cost of external acquisition.
• Talent & Compensation Committee: monitor a human capital metric linked to leader readiness.

Radiology offers a replicable model beyond healthcare. Work changes shape, demand for human expertise increases, and value shifts toward judgment exercised over exceptions.

These orientations share a common root: treating collaboration with AI as a designed organizational condition, rather than an individual choice left to people.

The design question for leaders

The evidence from radiology shifts the center of discussion from replacement to collaboration.

The real policy choice concerns the conditions that leaders design around their people. A redesigned role, a rethought workflow, and an investment in internal capability produce results that hold over time.

Final accountability remains human, and this choice protects both patients and trust in the system. The best policy makes this allocation of responsibility explicit.

The question every CHRO and CEO should ask is direct: what organizational conditions make human–AI collaboration the default mode, rather than the exception? The answer defines the trajectory of the next eighteen months.