Anthropic study finds AI adoption remains well below its technical potential in the workforce

Anthropic has published a research report introducing a new framework for measuring how artificial intelligence is affecting the workforce.

While earlier studies focused on what large language models could potentially do, the report reveals that real-world AI adoption rates remain substantially below the technology’s total technical potential.

The authors note the prediction that occupations with high AI exposure, such as computer programming and customer service, will see slower employment growth over the coming decade.

Despite those projections, the data compiled in the report shows no systematic increase in aggregate unemployment linked to AI exposure since late 2022.

The occupation currently most exposed to AI is computer programming, with an “observed exposure” rate of 74.5%, as the writing, updating and maintenance of software programs undergoes extensive automation.

Customer service representatives follow closely, with an exposure rate of 70.1%. Their core tasks, such as engaging with customers and handling complaints, are becoming increasingly automated.

The remaining top ten occupations with the highest AI exposure are:

Data entry operators (67.1%): at high risk due to significant automation of core tasks such as reading source documents and entering data into systems.

Medical records specialists (66.7%): at risk due to the automation of compiling, summarising and coding patient data.

Market research analysts and marketing specialists (64.8%): at risk in tasks involving report preparation, data visualisation and the conversion of complex findings into written text.

Sales representatives (wholesale and manufacturing, excluding technical and scientific products) (62.8%): at high risk in tasks involving contacting customers to promote products and take orders.

Financial and investment analysts (57.2%): tasks involving the analysis of financial information to forecast business, industry or economic conditions in order to guide investments.

Software quality assurance analysts and testers (51.9%): at risk due to tasks involving modifying software to correct errors or improve performance.

Information security analysts (48.6%): exposed to AI automation in tasks related to conducting risk assessments and testing data processing security.

Computer user support specialists (46.8%): focused on answering user questions concerning computer software or hardware.

By contrast, approximately 30% of workers currently fall outside the scope of AI, either because their tasks are highly physical or because they do not appear with sufficient frequency in AI usage data to meet the minimum threshold.

Occupations with zero AI exposure include cooks, electricians, nurses, motorcycle mechanics, lifeguards, bartenders, dishwashers and dressing room attendants.

Workers with high AI exposure tend to be slightly older.

Data comparing workers with zero AI exposure to those in the top quartile of exposure shows that the most heavily exposed group has an average age of 42.9, compared with an average age of 41.0 for the unexposed group.

Women are significantly more likely to be in occupations with high AI exposure. Women account for 54.4% of the most highly exposed group, but only 38.8% of the unexposed group.

AI exposure correlates strongly with higher levels of education. The most AI-exposed workers are far more likely to have advanced educational attainment.

By contrast, those with less formal education are overwhelmingly concentrated in unexposed roles.

Individuals with a master’s degree comprise 17.4% of the high-exposure group, but only 4.5% of the unexposed group.

Individuals with a bachelor’s degree account for 37.1% of the most exposed workers, but only 13.3% of unexposed workers.

Workers whose highest level of education is a high school diploma or less are far more prevalent in the unexposed group (52.1% in total) than in the top quartile of exposure (20% in total).

Workers in the top quartile of AI exposure earn an average of $32.69 per hour.

That figure is 47% higher than the average hourly wage of $22.23 for workers with zero AI exposure.