Anthropic researchers are measuring the impact of AI on job loss using a new measurement tool.
Anthropic is releasing a new measurement method to map the impact of AI on jobs. Researchers at Anthropic detailed their findings in the report Labor market impacts of AI: A new measure and early evidence. The study shows that, for now, AI is not causing a measurable increase in unemployment. However, there are indications that companies are hiring fewer young workers in professions highly exposed to AI.
Actual vs. theoretical
In the report, researchers Maxim Massenkoff and Peter McCrory introduce a new method to measure how heavily professions are exposed to AI. The study combines the theoretical capabilities of large language models with data on actual use in professional contexts.
The analysis shows that actual AI usage is still far below its theoretical potential. In the category of computer and mathematical occupations, AI can theoretically support about 94 percent of tasks. In practice, about a third of those tasks are currently actually performed using AI.
According to the study, programmers, customer service representatives, and data entry clerks are among the professions with the highest AI exposure. Professions involving physical tasks, such as chefs or mechanics, are hardly affected because many of their activities cannot be performed by language models.
No rising unemployment
The researchers compare their exposure metric with labor market data from the Current Population Survey in the United States. They look at differences in unemployment between professions with high and low AI exposure.
The analysis shows no systematic increase in unemployment in the most exposed professions since the introduction of generative AI tools. Unemployment trends between highly exposed and low-exposure groups remain largely comparable.
However, the researchers did find a possible early signal among younger workers. The influx of workers between the ages of 22 and 25 into professions with high AI exposure appears to be declining slightly. According to the study, this influx is about 14 percent lower than in 2022, although that result is only just statistically significant.