Jobbapokalypsen dröjer – AI ersätter inte människor

And any jobs-pocalypse seems a long way off.

Almost every week the world takes another step in the direction of artificial general intelligence. The most powerful AI models can do an astonishing array of tasks from writing detailed reports to creating video on demand. Hallucinations are becoming less of a problem.

Small wonder, then, that so many people worry they will soon be surplus to requirements. Earlier this year global Google searches for “AI unemployment” hit an all-time high. In cities such as London and San Francisco, “How long do you reckon you have left in your job?” is a common topic of conversation. But is ChatGPT actually putting anyone out of work?

Lots of pundits claim that it is. Many point to a recent paper by Carl Benedikt Frey and Pedro Llanos-Paredes, both of the University of Oxford, which suggests a link between automation and declining demand for translators. At the same time, however, official American data suggest that the number of people employed in interpretation, translation and the like is 7% higher than a year ago. Others point to Klarna, a fintech firm, which had boasted about using the tech to automate customer service. But the company is now undertaking an about-turn. “There will always be a human if you want,” Sebastian Siemiatkowski, its boss, has recently said.

Others still scour the macroeconomic data for signs of the forthcoming AI jobs-pocalypse. One popular measure is the ratio of the unemployment rate between recent college graduates and the overall American average. Young graduates are now more likely than the average worker to be jobless (see chart 1). The explanation runs that they typically do entry-level jobs in knowledge-intensive industries—such as paralegal work or making slides in a management consultancy. It is exactly this sort of task that AI can do well. So maybe the technology has eliminated these jobs?

Well, no. The data simply do not line up with any conceivable mechanism. Young graduates’ “relative unemployment” started to rise in 2009, long before generative AI came along. And their actual unemployment rate, at around 6%, remains low.

Returning to a measure we introduced in 2023, we examine American data on employment by occupation, singling out workers that are believed to be vulnerable to AI. These are white-collar employees, including people in back-office support, financial operations, sales and much more besides. There is a similar pattern here: we find no evidence of an AI hit (see chart 2). Quite the opposite, in fact. Over the past year the share of employment in white-collar work has risen very slightly.

Across the board, American unemployment remains low, at 4.2%. Wage growth is still reasonably strong, which is difficult to square with the notion that AI is causing demand for labour to fall. Trends outside America point in a similar direction. Earnings growth in much of the rich world, including Britain, the euro area and Japan, is strong. In 2024 the employment rate of the OECD club of rich countries, describing the share of working-age people who are actually in a job, hit an all-time high.

There are two competing explanations for these trends. The first is that, despite the endless announcements about how firms are ushering AI into their operations, few make much use of the technology for serious work. An official measure suggests that less than 10% of American companies employ it to produce goods and services. The second is that even when companies do adopt the tech, they do not let people go. AI may simply help workers do their jobs faster, rather than making them redundant. Whatever the explanation, for now there is no need to panic.

© 2025 The Economist Newspaper Limited. All rights reserved.