Data
This site provides access to data from the WAL's published research and the code used to generate it. Each data download includes a readme file with futher information, including the reference paper that users should cite as data source.
These datasets can be found
on HuggingFace
.
-
Job-Task Shares
This dataset provides task-level measures of labor input within occupations, constructed from O*NET task frequency data.
For each occupation-task pair, we estimate:
- Task flow (mu): expected number of times a task is performed annually per worker within an occupation.
- Task share (pi): proportion of total labor input allocated to the task.
Under a homogeneous task-duration assumption, task shares can be interpreted as time allocations across tasks within occupations.
Unlike standard O*NET-derived measures, this dataset provides statistically specified estimators of task-level labor input, including uncertainty (variance) for incumbent-based estimates.
-
US AI Labor Exposure Map
Underlying data for the US AI Map: state- and MSA-level labor market facts joined with O*NET 30.2 occupation and task dimensions, task labor input shares, and AI exposure metrics.
The release includes geographic fact tables and dimension files derived from BLS Occupational Employment and Wage Statistics (OEWS 2024), supporting both theoretical and current-exposure views at state and metro levels.
-
Task-Level AI Exposure Scores (Gemini 3.1)
Task-level AI exposure labels for O*NET task statements, classified with a 2026 agentic AI exposure rubric and Gemini 3.1 Pro.
Each task is assigned one of four exposure categories:
- E0: no meaningful direct exposure
- E1: direct exposure to a standalone agentic AI system
- E2: exposure requiring additional software or enterprise integration
- E3: exposure where image capabilities are the critical enabling factor
The dataset includes the original task text, predicted label, and a short model-generated explanation for each task.
-
AI, Jobs and Labor News Corpus
Curated collection of news articles and academic paper abstracts on artificial intelligence, automation, and labor markets.
The corpus contains roughly 49,000 documents: full-text articles from major news outlets and abstracts from scholarly papers indexed in Semantic Scholar, each with publication dates and metadata.
Intended for retrieval, RAG, classification, topic modeling, and other text-based research on AI and workforce change.