Reproduce “Working with AI: Measuring the Applicability of Generative AI to Occupations” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kiran Tomlinson, Sonia Jaffe, Will Wang, Scott Counts, Siddharth Suri
The authors analyze 200,000 anonymized U.S. Bing Copilot conversations from 2024, classifying users’ goals and AI actions separately before mapping them to O*NET’s work activities. Copilot use centers on information work such as gathering information, learning, writing, and communicating, connecting AI assistance to activities across occupations. The resulting scores estimate potential applicability, not productivity gains, job loss, or wage effects.
Key insights
- Copilot use centers on information tasks such as learning, writing, explaining, and communicating; in 40% of conversations, the AI’s identified activities do not overlap with the user’s stated goal activities.
- Applicability is not limited to high-wage professional jobs: sales and administrative occupations score highly, and the employment-weighted relationship between applicability and wages is weak (r = 0.13).
- From January to September 2024, use broadened across more activity types; measured capability scope changed little, indicating that expanding usage breadth drove most of the increase in applicability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.