In 2013, two researchers at Oxford University wrote a famous paper. They said about 47% of jobs in the United States were at high risk from computers. Many people thought that half of all workers would lose their jobs.

But the paper did not say that. Later studies found that technology often changes the tasks inside a job. It does not usually remove the whole job.

In 2013, Carl Benedikt Frey and Michael Osborne of Oxford University estimated how likely 702 US occupations were to be automated. They concluded that about 47% of US jobs were in high-risk occupations. News reports often said that half of all jobs would disappear, but the paper only measured what might be technically possible. Whether automation actually happens also depends on cost, laws and social acceptance.

In 2016, an OECD report pointed out that people with the same job title often do different tasks. When the researchers analyzed individual tasks instead of whole occupations, the share of high-risk jobs fell to about 9% on average across 21 countries. Today, many experts believe AI changes the content of jobs rather than removing them completely.

In 2013, Frey and Osborne estimated the probability of computerization for 702 US occupations and concluded that about 47 percent of employment fell into a high-risk category. The number is usually quoted as "47 percent of jobs will disappear." That is not what the study measured. It estimated technical susceptibility, and whether a technology is actually adopted also depends on costs, regulation and public acceptance.

Critics soon went after the method. A 2016 OECD paper pointed out that people in the same occupation perform different tasks. When the analysis looked at tasks instead of whole occupations, the high-risk share across 21 countries fell to roughly 9 percent. Australian economists later reported that the original predictions did a poor job of explaining how employment actually changed. Generative AI has since reopened the question.

A 2024 paper in Science estimated that around 80 percent of US workers could see at least 10 percent of their tasks affected, and highly educated, well-paid jobs were among the most exposed. Exposure is not the same as job loss. A 2026 working paper, whose data overlap with the pandemic years, found that employment and wages grew even in occupations where AI was highly applicable.