The study titled “Labor market impacts of AI: A new measure and early evidence”, published in March by Anthropic researchers Maxim Massenkoff and Peter McCrory, presents a new approach developed to measure the impact of artificial intelligence on the labor force and its first results. The study analyzes how much occupations are affected not by what AI can theoretically do, but by which tasks it is actually used for in real working life.
In the research, a measure called Observed Exposure is used to measure the impact of artificial intelligence on jobs. This measure aims to determine which occupations are truly starting to be affected by combining the tasks that AI can theoretically perform with real usage data.
How is AI impact measured
In the study, three different data sources are used together to determine how much occupations are affected by artificial intelligence: the O*NET database (task descriptions of occupations), real AI usage data, and previous studies measuring whether tasks can theoretically be performed with AI.
When calculating the exposure of an occupation, the following factors are taken into account:
- Whether the task can be done with artificial intelligence
- Whether this task appears in real usage
- Whether the usage is work-related
- Whether the task is automated or done with support
- How much of the occupation this task represents
This method is used to see the difference between theoretical capability and real usage.
There is a large gap between theoretical capability and real usage
According to the research, artificial intelligence is theoretically capable of performing many tasks, but actual usage remains well below this level. For example, in computer and mathematics occupations, it is estimated that AI could theoretically affect about 94% of tasks, while real usage is currently around 33%. The report states that there may be several reasons why some tasks that are theoretically possible do not appear in real usage. These include model limitations, legal constraints, specific software requirements, steps requiring human verification, and the slow diffusion of new technology into working life.
Occupations most affected by artificial intelligence
According to the research, in some occupations a significant portion of tasks can be performed or supported by artificial intelligence. Some of the occupations with the highest exposure are:
- Computer programmers
- Customer service representatives
- Data entry workers
- Financial analysts
- Marketing and research specialists
- Sales representatives
- Software testers
- Information security specialists
In these occupations, a large part of the work consists of tasks such as text production, data processing, analysis, and reporting, which leads to higher use of artificial intelligence.
Exposure is very low in some occupations
The research also shows that AI usage is very low in some occupations. This is related to the nature of the tasks.
Real usage data shows very little AI usage in the following occupations:
- Cooks
- Motorcycle mechanics
- Lifeguards
- Dishwashers
- Agricultural workers
- Maintenance and repair workers
In these occupations, most tasks require physical work, which limits the use of artificial intelligence.
Characteristics of workers more exposed to AI
The research shows that workers who are more exposed to artificial intelligence share some common characteristics.
These workers generally:
- Have higher levels of education, Have a higher share of bachelor’s and graduate degrees
- Earn higher wages
- Have a higher proportion of female workers
These results show that artificial intelligence is used more in knowledge-based and office jobs.
No clear increase in unemployment
The research also examines how unemployment rates change in occupations that are more exposed to artificial intelligence.According to the results, when occupations with high exposure are compared with those with low exposure, no clear difference in unemployment rates is observed. Despite the spread of artificial intelligence, no major change in the labor market is seen during the analysis period. The study interprets this as an indication that the effects of artificial intelligence may still be at an early stage.
Entry into jobs may be lower for young workers
The research also analyzes employment outcomes for young workers. In the analysis of the 22–25 age group, the entry rate into occupations with high AI exposure is found to be about 14% lower. The data suggests that the number of young workers starting new jobs in these occupations may be declining.This may indicate that some entry-level tasks are beginning to be automated. Companies may hire fewer new workers or prefer technology instead of inexperienced employees.
The research shows that artificial intelligence is affecting working life, but the effect is still limited. Although many tasks appear theoretically possible to perform with AI, real usage remains lower. The use of artificial intelligence is more concentrated in data, text, and analysis-based jobs, while occupations requiring physical tasks show more limited exposure.Even though no major increase in unemployment is observed, there are early signals that hiring may slow down in some occupations.
Resource:
Massenkoff, M., McCrory, P. (2026), Labor market impacts of AI: A new measure and early evidence, Anthropic Research : Link
