Neither labor apocalypse, nor imminent disruption. Almost three years after the launch of Chatgpt, artificial intelligence has not caused any discernible structural change in the labor market. This is demonstrated by the most thorough and systematic study published to date by the Budget Lab of the University of Yale, in collaboration with the Brookings Institution.

The report, which has been updated monthly since November 2022, analyzes million data from the US labor market based on indicators such as unemployment, occupational rotation or real and theoretical exposure to AI tools. Its conclusion is that the labor market has not experienced any “shock” attributable to generative artificial intelligence. “The most alarmist holders do not correspond to real data. What we see is stability, not disruption”indicates the document.

Data denies the narrative

The research team has measured the evolution of the occupational mixture, a metric that reflects how many people change their work or sector, or enter/leave the market, and has compared the AI ​​period with three historical moments: the adoption of the PC in the 80 2022. “Historically, technological effects on employment occur throughout decades, not months. AI is no exception”they conclude.

Thus, in the 33 months since the launch of Chatgpt (November 2022), occupational disguise, that is, the change in the job mix, has reached a maximum of 4.75%. To put it in context, in the era of computerization (1984), this index reached up to 3.47%. With the popularization of the Internet (1996), it reached 3.76%. Even during the control period (2016-2019) without major technological disruptions, the index touched 4.02%.

This suggests that the change in current work profiles is not more accelerated or more extreme than lived with previous technologies.
Even in sectors traditionally more susceptible such as information, finance or professional services, observed changes began before Chatgpt’s public emergence, and have not been significantly accelerated since then.

Another axis of the study has been to confront two dimensions of the potential impact of the AI: the theoretical exposure (measured by OpenAi according to what percentage of tasks of an occupation can be carried out with AI) and the real use (according to data from Anthropic and Claude). The results show that:

• The proportion of workers in jobs highly exposed to AI remains stable since 2022, around 18%
• The presence of workers in really automated or augmented occupations has not grown up appreciably
• Even among the unemployed, a bias is not detected towards professions exposed to the

The media overexposure of sectors such as programming or content creation, which do make intensive use of generative, distort, therefore, a much more diverse reality: most sectors, from administration to health or manufacturing, are not actively using these large -scale technologies.

According to OpenAI data analyzed by Budget Lab, 18% of American workers are in occupations with high exposure to AI, that is, where more than 50% of the tasks could be performed by tools such as Chatgpt. However, since November 2022, this proportion has not varied significantly. In addition, 45.5% of jobs are in medium exposure categories and the remaining 29%, in occupations with low exposure.
In other words, there is no visible labor displacement towards less exposed occupations, as one would expect if AI was massively replacing human work.

Research has also examined whether AI is disproportionately affecting recent graduates or young workers, but has not found conclusive evidence. And it is that a common hypothesis suggests that the first affected by automation would be younger or less experience. However, Budget Lab data shows that the difference between the occupations that young graduates (20-24 years) and their older counterparts (25-34 years) have not increased relevantly since the launch of ChatGPT in November 2022. The dissimilarity index remains stable around 30-33% for several years.

Although there is a slight upward trend since mid -2023, this is consistent with the behavior of the previous labor market and could also be explained by a general slowdown of hiring. Thus, there is no pattern indicating that young people are being expelled from their work niches by artificial intelligence. “The transformations that we attribute to AI are often prolongations of previous dynamics. The risk is to overcome before having clear evidence”warns the report.

A warning about fear

The authors of the study recognize that their findings are not predictive: the fact that IA has not yet transformed the labor market does not mean that it cannot do it in the long term. However, they emphasize that generalized anxiety around automation is based more on perceptions than on data.

Thus, a key element of the analysis is the distinction between two different concepts: the theoretical exposure, such as the one calculated by OpenAI, which estimates what percentage of tasks could, in theory, be affected by models such as chatgpt. And the real use, such as the one collected by Anthropic in his adoption metrics of the Chatbot Claude in professional environments.

When comparing both databases, the report shows that there is no strong correlation between the sectors most exposed to AI and those where it is really being used on a day -to -day basis. For example, areas such as programming or communication have high levels of use and exposure, while sectors with high theoretical exposure such as administration or accounting show little real use. And some professions such as production, cleaning or transport appear as exposed but barely use in practice.
This lag suggests that many “at risk” professions continue to operate with little or no integration of generative, which delays any possible tangible labor effect.

Waiting for more complete metrics and standardized use data by large technological ones, the Budget Lab report marks an important guideline: before designing policies or making radical corporate decisions, it is convenient to look beyond the noise and observe the market with perspective.