Seminars

Occupations or what people do for a living have long been central to how social scientists understand inequality and opportunity. Yet much of today's research relies on simplified categories or income-based models, often overlooking how jobs are actually described, experienced, and socially constructed in the real world. One major challenge is surprisingly practical: people describe their jobs in their own words, producing messy, open-ended data that are difficult to translate into standardized occupational categories. For decades, this bottleneck has limited what researchers can learn about the structure of work. In this talk, I introduce a new, AI-assisted approach that uses large language models (LLMs) to make sense of these free-text job descriptions at scale. By training models on detailed occupation data including tens of thousands of job titles we can more accurately connect what people say they do to how occupations are formally classified. Beyond improving measurement, this approach opens up new ways for social scientists to study the hidden heterogeneity within occupations and to better understand emerging forms of labor market inequality.