In 2013, researchers at the University of Oxford published startling figures on the future of work: they estimated that 47% of American jobs were “at risk” of automation “in an unspecified number of years, perhaps – be a decade or two.”

Ten years later, the unemployment rate in the country has reached a historic low. The tsunami of grim headlines of the time – like “The rich and their robots are about to wipe out half the world’s jobs” – seem utterly misguided.

But the study authors say they did not intend to suggest the apocalypse was near. Rather, they were trying to describe what the technology was capable of.

It was the first attempt at what has become a long-term thought experiment, with think tanks, business research groups and economists publishing article after article to determine how far work is “affected by” or “exposed to” technology.

In other words: if the cost of tools was irrelevant and the only goal was to automate as much human work as possible, how much work could technology support?

Back when Oxford researchers Carl Benedikt Frey and Michael A. Osborne were conducting their study, IBM Watson, an artificial intelligence (AI)-powered question answering system, had just shocked the world by winning the Jeopardy game! Test versions of autonomous vehicles were on the roads for the first time. Today, a new wave of studies is following the rise of tools that use generative AI.

In March, Goldman Sachs estimated that the technology behind popular AI tools like DALL-E and ChatGPT could automate the equivalent of 300 million full-time jobs. Researchers from Open AI, the designer of these tools, and the University of Pennsylvania found that 80% of the American workforce could see an effect on at least 10% of their tasks.

“The uncertainty is huge,” said David Autor, a professor of economics at the Massachusetts Institute of Technology (MIT), who has studied technological change and the labor market for more than 20 years. “And people want to provide those answers. »

But what exactly does the claim that, for example, the equivalent of 300 million full-time jobs could be affected by AI mean?

It depends, according to Mr. Autor: “Affected can mean improved, worsened, gone, doubled. »

The fact that technology tends to automate tasks, not entire professions, is a complicating factor. In 2016, for example, AI pioneer Geoffrey Hinton investigated a new “deep learning” technology capable of reading medical images. He concluded that “if you work as a radiologist, you are like the coyote who has already crossed the edge of the cliff, but has not yet looked down.”

He predicted five years, possibly ten years, before algorithms would do “better” than humans. What he probably forgot is that reading images is just one of many tasks (30, according to the US government) of radiologists. They also deal with “talking to medical professionals” and “giving advice”. Today, some professionals in the sector are concerned about an impending shortage of radiologists. And Mr. Hinton has since become a vocal public critic of the very technology he helped create.

Frey and Osborne calculated their 47% figure in part by asking tech experts to assess the feasibility of automating entire professions like “telemarketer” or “accountant.” Three years after their paper was published, a group of researchers from the ZEW Center for European Economic Research, based in Mannheim, Germany, published a similar study that assessed tasks like “explaining products or services” and found that only 9% of professions in 21 countries could be automated.

“People like numbers,” said Melanie Arntz, lead author of the ZEW study.

In some scenarios, the AI ​​essentially created a tool, not a complete job replacement. You are now a navvy who can use an excavator instead of a shovel. Or a nurse practitioner with access to better information to diagnose a patient. You may have to charge more per hour because you will get much more results.

In other cases, technology replaces your work instead of complementing it. Or it turns your job that requires special skills into one that doesn’t. This is unlikely to go well for you.

Even if a job becomes completely automated, the fate of displaced workers will depend on how companies decide to use technology for new types of work, especially those we cannot yet imagine, explained Daron Acemoglu, professor at MIT and author of Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity. These choices will relate in particular to the complete automation of work or the use of technology to increase human expertise.

He said the seemingly chilling numbers predicting how many jobs AI could eliminate, although it’s unclear exactly how, were a “wake-up call”.

He thinks people might “go in a better direction,” but he’s not optimistic. He doesn’t think we’re on a “pro-human” path.

Any estimates of how much work AI could take on depend heavily on humans: researchers making assumptions about what the technology can do. Frey and Osborne invited experts to a workshop to assess the likelihood of automating professions. More recent studies rely on information such as a database of AI capabilities created by the Electronic Frontier Foundation, a nonprofit digital rights group. They also rely on workers using platforms such as CrowdFlower, where people complete small tasks in exchange for money. Workers rate tasks based on factors that predispose them to automation. For example, if the error tolerance is high, it is a better candidate for automation for a technology like ChatGPT.

Exact numbers aren’t the point, argue many researchers involved in this type of analysis.

“I would describe our methodology as almost certainly flawed, but hopefully pointing in the right direction,” said Michael Chui, an AI expert at McKinsey who was one of the authors of a 2017 white paper suggesting that about half of the work and 5% of the professions could be automated.

What the data describes is, in some ways, more mundane than is often realized: big changes are coming, and they are worth paying attention to.