Don’t become a radiologist. That was the advice to medical students from AI pioneer and Nobel Prize winner Geoffrey Hinton in 2016. Computers would take over their work within a few years. Because how can a specialist who will view at most about a hundred thousand X-rays over an entire career compete with a machine that has been trained with hundreds of millions of images? But ten years later, there appears to be a shortage of radiologists worldwide, their salaries are rising and the number of training places is increasing. What happened?
The answer is as simple as it is counterintuitive. Digitalization and AI have made the production and analysis of medical images faster and cheaper, exploding the use of X-rays, CT scans, MRIs and ultrasounds. Thanks to computer assistance, radiologists process more images in less time and it is precisely this efficiency gain that has sharply increased demand. Radiologist turned out not to be a dead-end job, but a shortage profession.
This pattern – that efficiency does not lead to less but rather to more use – has a name: Jevons’ paradox. In 1865, the English economist William Stanley Jevons described how James Watt’s improved steam engine did not lead to less coal consumption, but rather to a huge increase in it. The new machine was so much more efficient that countless new industries began to mechanize and therefore consume more coal. Jevons formulated a crucial economic insight: when a raw material is used more efficiently, demand can increase so much that total consumption increases instead of decreasing.
Jevons’ paradox crops up again and again in technological innovation and can undermine predictions about the labor market. In the nineteenth century, machine looms replaced the manual work of weavers. Mass unemployment was expected. But textiles became so cheap that demand skyrocketed. Everyone could suddenly afford multiple items of clothing and modern ready-to-wear was born. Employment in the textile industry did not decrease but increased, although the work in the factories – twelve hours a day, six days a week, in deafening noise – was dangerous and poorly paid.
We saw the same thing when PCs came onto the market. Now that everyone could start tinkering with electronic spreadsheets themselves, the end seemed to be in sight for accountants. The opposite happened. Because complex questions could now be answered in seconds instead of days, the demand for financial analyzes and therefore for accountants and advisors increased. Instead of long calculations, complicated considerations were now required. Here too, automation lowered the threshold and demand increased.
Every time technology turns something scarce and precious into something abundant and cheap, Jevons’ pattern repeats itself. Taking a photo was an expensive and complex task until the digital camera came along. Now more photos are taken in just a few minutes than in the entire nineteenth century.
However, the current AI revolution seems to be of a different order. In coal, textiles and spreadsheets, a specific technology became available that made one type of labor more efficient. The arrival of smart machines makes intelligence cheap and widely accessible – a meta-commodity that you use in everything. With intelligence there is no natural upper limit to demand, because there is hardly a domain in which more brainpower is not useful. Think of education, with a personal teacher for every student. Or healthcare, with access to the best diagnostics for patients all over the world. This makes Jevons’ paradox with intelligence so much more powerful than with coal or cotton. And predictions about the labor market are so much more difficult.
Among radiologists, we still think it is important that a doctor assesses the entire patient based on the data and analysis that machines provide. Nobody wants to hear a life-threatening diagnosis from an algorithm. Chess grandmaster Garry Kasparov called this combination of man and machine the “centaur”, after the mythical creature that is half man, half horse. After his humiliating loss to the computer program Deep Blue in 1997, Kasparov organized tournaments in which man-plus-machine competed against both individual computers and grandmasters. His surprising discovery: it was not the strongest computer or the best chess player who won, but the team that worked together the smartest. Many professionals, from radiologists and consultants to generals and scientists, are now in such a centaur phase, moving from routine work to roles that require judgment and smart teamwork with machines.
The arrival of smart machines makes intelligence cheap and widely accessible – a meta-commodity that you use in everything
However, the Jevons effect does not guarantee that everyone will keep their job, only that aggregate demand will grow. History provides plenty of examples where a technological… tree in employment was followed by an equally dramatic one bust caused by a subsequent innovation. The invention of the spinning machine in the 1770s led to a boom in the number of hand weavers, because there was suddenly so much more yarn available. The subsequent automation of the loom decimated that same profession. The telephone created a huge job market for telephone operators. Automatic circuits washed those lanes away again.
The centaur phase is therefore real, but the question is how long it lasts. In chess that was exactly twenty years. In 2017, AlphaZero appeared on the scene, the self-learning computer program that pulverized every combination of man and machine after just four hours of training. It turned out that the centaur could do very well without a human head.
The uncomfortable question is not whether AI will take over many tasks from us, but whether there will always be something left that only humans can do. For now, the answer seems to be yes: judgment, the bigger picture, the difficult conversation, taking responsibility. In this way, we are all a bit like a radiologist, convinced of our irreplaceable qualities, while the machine quietly learns to see what we are missing.
