If ChatGPT and generative AI live up to even a tenth of the hype surrounding them, wide-scale job losses might seem inevitable. But new economic data shows that the last big leap in AI did not coincide with a reduction of jobs in affected industries—despite widespread fears of rapid replacement at the time.

In a new research paper, economists looked at the job market across a number of European countries between 2011 and 2019. That’s the period during which the AI technique deep learning emerged as a powerful way to automate tasks like transcribing speech, analyzing images, and making algorithmic recommendations for social feeds and ecommerce sites. Back then, deep learning was widely expected to have a broad and swift impact on employment.

To check up on what really happened, researchers at the European Central Bank, Spain’s central bank, and the universities of Oxford and Pittsburgh used two establishedmethods for measuring how vulnerable professions are to AI-powered automation. Both involved examining the tasks workers do and how they compare with the capabilities of algorithms. The researchers cross-referenced that information with survey data on EU workers that shows the number of people leaving or joining different occupations in industries ranging from agriculture to financial services.

The headline result was that industries where AI could be the most useful did not see a reduction of jobs. In fact, for more highly skilled jobs vulnerable to AI, such as white-collar office work that involves working with data, there was around a 5 percent increase in the number of employed workers. The researchers say this supports the idea that new technology can increase demand for more skilled workers at the same time that it replaces those who do routine work. Less skilled workers didn’t seem to be significantly affected by software or AI.

Although fears about new technologies taking jobs are common—and entirely understandable—economic research offers a complex, mixed picture. In general, many economists believe that automation can increase demand for jobs overall, as shown by some recent studies. If you are not one of those more skilled workers, however, the arrival of new technology might be a problem.

None of this research makes it possible to predict what effect ChatGPT or other generative AI technologies will have. They may be too new to cause noticeable changes. And it’s possible that generative AI has an entirely different effect on jobs to what came before. “While in the period of our analysis the association is positive, these results may not be extrapolated into the future,” the authors of the new study write.

Generative AI is undeniably already causing some disruption. You don’t have to look far to find stories of individuals and businesses that are already being changed by the kind of AI behind ChatGPT.

Some copywriters, for instance, are having to find new careers after customers replaced them with generative AI tools. A number of publications are also experimenting with AI-generated content. The German tabloid Bild recently cited AI as one reason for planned job cuts.

But let’s not forget that tools like ChatGPT are still unreliable coworkers, as they make up facts, reinforce biases, and can otherwise misbehave. And it’s worth remembering that the emergence of deep learning in the 2010s prompted some AI experts to predict the elimination of certain jobs, including radiologists—a prognostication that has hardly come true.