Yale study finds little evidence that AI is taking people’s jobs

A new analysis from the Budget Lab at Yale University has found little sign that artificial intelligence is having a measurable impact on the composition of the US workforce, despite widespread debate about its potential to transform the jobs market and reduce levels of employment. The study, led by Martha Gimbel, Molly Kinder, Joshua Kendall and Maddie Lee, examined monthly labour market data since the public release of ChatGPT in November 2022.

It compared recent patterns to those seen during previous technological shifts, such as the rise of personal computers and the internet. The researchers also analysed data on which occupations are considered most exposed to AI, alongside information about usage of generative AI tools, to test whether jobs theoretically at risk have begun to decline.

The findings suggest that the mix of occupations in the US has not changed more quickly in the wake of generative AI than during earlier periods of change. Jobs that appear more exposed to automation have not, so far, lost employment share. Measures of AI use and job exposure show no consistent relationship with job losses or gains.

One area where the researchers observed a difference was among recent graduates. Workers aged between 20 and 24 have experienced a slightly greater shift in occupational outcomes than older cohorts. However, the authors caution that this may reflect broader labour market weakness rather than the specific influence of AI.

The study also notes important limitations. Data on exposure to AI is largely theoretical and based on tasks that could be automated in principle, while usage data is drawn from a limited set of tools. The authors emphasise that the results represent an early snapshot rather than a long-term forecast.

While many expect artificial intelligence to have profound effects on the structure of work, this analysis indicates that such changes may take longer to emerge. The researchers plan to update the study regularly as adoption spreads and new evidence becomes available.