The Setup
In 1865, British economist William Stanley Jevons noticed something strange. Steam engines were becoming more efficient, yet coal consumption across England was rising. His conclusion, now called the Jevons Paradox, was simple but unsettling: when you make a process more efficient, you often increase total demand for that process.
Today, artificial intelligence is the steam engine of the information age. Efficiency gains are staggering, and predictions of mass obsolescence follow every new breakthrough. Among the boldest was the claim by Geoffrey Hinton, a Nobel-level pioneer in deep learning, who once said we should stop training radiologists because AI would soon do their work better.
That claim aged poorly. The number of radiologists has risen. Imaging volumes are exploding. Pay remains high. The paradox is alive and well.
The Original Prediction
Around 2016, Hinton said, “People should stop training radiologists now.”His reasoning was straightforward. Image interpretation is a pattern-recognition task, and deep learning excels at pattern recognition. Once machines could outperform humans at reading X-rays and MRIs, human radiologists would be redundant.
The logic sounded convincing, until it met reality.
The Reality Check
Nearly a decade later, radiology is thriving:
Training programs are expanding. U.S. residency slots hit record highs in 2025.
Average incomes are climbing, with radiologists earning about $520,000 annually, second only to orthopedics.
Demand keeps rising, and hospitals are performing more imaging per patient than ever before.
The assumptions behind the prediction simply did not hold.
Why the Prediction Failed
The job is more than image readingRadiologists interpret images, but they also guide procedures, design protocols, teach, consult, and manage clinical risk. Only a fraction of their time is spent on direct interpretation. Replacing that one slice does not remove the rest of the role.
Lab accuracy is not real-world performanceAI can excel in controlled datasets but still struggle with messy hospital data that contain artifacts, variable quality, or incomplete scans. Real-world adoption requires reliability, regulatory approval, and trust, none of which move quickly.
Efficiency breeds demandDigital imaging already proved this point. When scans became faster and cheaper, hospitals ordered more of them. From 2000 to 2008, imaging volume per 1,000 patients rose by roughly 60 percent. Efficiency lowered the barrier, and usage exploded.
Humans remain the trust anchorIn medicine, responsibility cannot be automated. Patients and regulators still expect a human to validate results, explain outcomes, and assume accountability.
The profession adaptedModern radiologists now study data science, workflow design, and AI validation. They are not being replaced. They are evolving.
The Jevons Paradox Revisited
Jevons’ observation remains clear:
When technology reduces the cost of using a resource, total use often rises instead of falling.
In radiology, the “resource” is expert image interpretation. As AI makes scanning faster and cheaper, healthcare systems perform more scans, creating more work that still requires human oversight.
Efficiency expands the system rather than shrinking it.
Broader Lessons for Security and Technology
Anyone working in corporate security, logistics, or supply chain risk should take note. The same dynamic applies.
When AI improves the speed of threat analysis or cargo screening, it often increases the total number of analyses performed. The system becomes broader, not smaller. New data flows, new anomalies, and new layers of interpretation appear.
Efficiency rarely reduces complexity. It multiplies it.
A Caution Against Premature Abandonment
When leaders assume that automation replaces human expertise, they risk misallocating talent. If universities had followed Hinton’s advice, there would now be a severe shortage of radiologists. Similar mistakes occur in security, operations, and intelligence when efficiency is mistaken for finality.
The correct strategy is not to stop training humans. It is to train humans differently—those who can guide machines, audit outputs, and bridge the gap between algorithms and accountability.
The Pattern Behind the Paradox
Efficiency lowers cost.
Lower cost drives demand.
Demand expands system scale.
System scale reintroduces complexity.
Complexity restores the need for human oversight.
Automation bends the curve temporarily, but it still loops back to people.
Final Signal
The Jevons Paradox is the ghost in every efficiency dream. Each time we think we are automating ourselves out of the equation, we find new reasons to stay in it.
AI did not make radiologists obsolete. It made them indispensable in new ways. The same will be true in every field that deals with complexity, judgment, and risk.
Efficiency is not the end of expertise. It is the next test of it.
Call to Action
If this perspective resonates, share this post and subscribe to Ice Station Briefing. Each week, we explore how technology reshapes risk, perception, and the human element behind both.