In theory, AI could take away millions of jobs, but for now it is only replacing a few

The artificial intelligence (AI) is not yet causing a massive wave of unemployment, despite warnings about its potential impact on the labor market, according to a study published by the company Anthropic that analyzes the real use of these systems in professional environments.

The research, carried out by economists Maxim Massenkoff and Peter McCrory, introduces the concept of “observed exposure”a metric that measures which tasks are actually being performed by AI models, rather than those that could be automated in theory. To do this, the authors analyzed data on the use of the chatbot Claude in work contexts.

MIRA: Creating, editing or altering images with AI: when it stops being a record and becomes a language

The results show a significant gap between the theoretical capability of AI and its actual use. In areas like computer science and mathematics, for example, language models could execute up to 94% of the tasks associated with these professionsbut in the practice They only do around 33 %according to the study.

The researchers point out that this difference is explained by multiple barriers, including legal restrictions, the need for human verification and limited integration with companies’ IT systems, which slows down the automation of many tasks.

According to the analysis, the jobs with the greatest current exposure to AI are computer programmers (74.5%), customer service personnel (70.1%) and data entry workers (67.1%). On the other hand, around 30% of workers – such as cooks, mechanics, lifeguards or waiters – have zero exposure, since their functions require physical presence.

The study also notes that the most exposed workers tend to have higher educational levels, 47% higher average incomes and are 16% more likely to be women, suggesting that automation could mainly affect skilled professionals such as financial analysts, lawyers or developers.

Despite this, the authors indicate that since the appearance of ChatGPT at the end of 2022, there is no statistical evidence of a systematic increase in unemployment in the sectors most exposed to AI. The effect, according to the report, is “indistinguishable from zero,” although these jobs are projected to grow at a slower rate in the next decade.

The analysis warns, however, that the impact of artificial intelligence could manifest gradually, similar to other structural changes such as the expansion of the Internet or globalization, so its effects on employment could take years to become fully visible.

It should be noted that the research measured the use of Claude, not that of artificial intelligence in the economy as a whole, which constitutes a limitation if its conclusions are to be extrapolated to the real scope of the labor market.