Anthropic Refines Job Exposure to AI Through Usage

Anthropic details a method for assessing job exposure to AI that combines theoretical capabilities with usage data from Claude and its API. A preliminary list places programmers and customer service representatives among the most exposed, while about 30% of workers would fall into low-risk occupations. The company anticipates gradual economic disruptions, more akin to those caused by the Internet or trade with China than to a sudden shock.

Gradual Disruptions and a Large Low-Risk Block

Anthropic aims to detect economic changes related to AI in advance, hoping to identify disruptions before their effects become significant. According to the company, these changes could occur gradually, similar to those triggered by the Internet or trade with China, rather than in the form of a sudden shock like that of COVID. In this context, Anthropic identifies a group of low-risk jobs, which would represent 30% of workers. These professions rank at the bottom of the established list, with their tasks rarely appearing in the analyzed usage data. Among them are cooks, motorcycle mechanics, lifeguards, and bartenders.

Measuring Exposure: Combining Theory and Real Usage

Anthropic's method relies on considering the gap between what AI is supposed to be able to automate and what is actually achieved in practice. The company cites several factors that may limit actual adoption: model limitations, legal constraints, specific software requirements, human verification steps, or other obstacles. The approach thus combines theoretical assessment of task automation with analysis of real usage derived from Claude and Anthropic's API. A graph presented by the company illustrates this gap, with the theoretical level represented in blue and the observed exposure to automation in red.

Initial Categories Published and Planned Follow-Up

By applying this methodology, Anthropic has published an initial list of jobs most exposed to automation by AI, which notably includes computer programmers and customer service representatives. The details of this method were communicated in March, and the company plans to regularly reassess its analyses to refine the measurement of AI's impact on employment. This approach comes as the use of artificial intelligence expands across many sectors, and evaluations often rely on the theoretical capacity of AI to replace humans.