TaskNet: Building Occupational Task Inventories from Worker Interviews at Scale

Understanding how AI is reshaping work requires an accurate account of the tasks workers perform.

Workers are often the first to see how their jobs are changing, as new responsibilities and tools often appear in everyday work before these changes reach formal descriptions. Interviews with workers can help capture these emerging changes in job tasks. Yet existing methods support standardized ratings across workers, while open-ended conversations with workers remain costly to collect and synthesize at scale. We introduce the Stanford Task Network (TaskNet), an occupational taxonomy grounded directly in workers’ accounts of their everyday work.

Download TaskNet (CSV)
Tasks by occupation
How TaskNet is built

Every task starts with workers describing their own jobs. Every task statement is then written to O*NET’s standards and validated by job analysts and workers.

Discover

132 Worker Interviews

A worker’s first account of a job is rarely complete, so we treat discovery as active elicitation: our AI interviewer helps workers recall their work, proposes missing tasks, and each worker confirms or revises the list.

Aggregate

2,726 Task Reports

Workers describe the same work in different words, so we group reports by goal and action rather than by vocabulary. Reports that differ only in tools or context become one task.

Validate

318 Raters, 7,923 Ratings

O*NET Center analysts check every statement against O*NET’s standards, and workers rate whether it reflects their job. Tasks rated relevant by at least a quarter of workers are kept.

How to interpret ratings

318 workers rated the tasks, each rating up to 30. Relevance comes from everyone who rated a task, and the other ratings only from workers who said it belongs to their occupation. Tasks rated relevant by fewer than 25% were dropped.

Is It Part of the Job?

Relevance

Share of workers who say the task belongs to their occupation, not just their own role.

Importance

How important the task is to the occupation, averaged on O*NET’s scale from 1 (not important) to 5 (extremely important).

How the Work Happens

Frequency

How often a worker typically does the task, from once a year or less to hourly. Shown as the nearest step on that scale.

Duration

How long one round of the task takes, from under 15 minutes to four or more workdays.

Expertise

Training

How much on-the-job and classroom training it takes to do the task well, from a short demonstration to over 10 years.

Top-worker gap

Share of workers who say a top performer (the best 10% in the same setting) clearly does the task better than an average worker.

Can It Be Specified and Checked?

Articulability

Share who say the standard for doing the task well can be clearly stated in advance. Low values point to tacit know-how.

Verifiability

Share who say a trained coworker could check whether the task was done correctly.

Understanding how AI changes work requires an up-to-date account of what people do in their jobs. ”

TaskNet: Building Occupational Task Inventories from Worker Interviews at Scale

Jennifer Wang, Andy Haupt, Isaac Robinson, Phil Lewis, Erik Brynjolfsson, and Diyi Yang

Read publication

Research team

Jenn is a PhD student in Computer Science at Stanford, co-advised by Diyi Yang and Erik Brynjolfsson. Her research explores how AI is shaping the future of work and how to design AI systems that preserve and enhance human knowledge production. Her current project investigates new approaches to building a next-generation O*NET —a database of occupations and the tasks —to better capture how work is changing.

She holds a BS in Computer Science from Brown University. Her interest in the societal implications of technology has led her beyond computer science into civic tech and AI policy, with experience at the White House Office of Science and Technology Policy and the U.S. Census Bureau. Across these settings, she has been drawn to how research and policy can help ensure that advances in AI serve the people and communities they affect.

Read more

Andreas Haupt is an AI Institute Fellow-in-Residence at Schmidt Sciences, a Digital Fellow at the Stanford Digital Economy Lab, and a Technology and Human Rights Fellow at the Harvard Kennedy School. He studies how algorithmic systems represent and are evaluated against diverse human preferences, using methods from microeconomic theory, structural econometrics, and reinforcement learning. His work has been published in AI venues such as NeurIPS and ICML, in computer science venues such as ACM EC, ACM RecSys, and ACM FAccT, and as a lead article in the American Economic Review.

Read more

Phil is responsible for the implementation, maintenance, and technical quality of the O*NET Program, the nation’s primary source of occupational information. He leads occupational data collection and oversees the Program’s web-based resources and tools.

He develops new projects and methodologies to further support ONET customers. Phil also serves as the ONET Center Liaison to sponsors and stakeholders, supporting O*NET projects within state, federal, industry, and international initiatives.

Erik Brynjolfsson is one of the world’s leading experts on the economics of technology and artificial intelligence. He is the Jerry Yang and Akiko Yamazaki Professor and Senior Fellow at the Stanford Institute for Human-Centered AI (HAI), and Director of the Stanford Digital Economy Lab. He also is the Ralph Landau Senior Fellow at the Stanford Institute for Economic Policy Research (SIEPR), Professor by Courtesy at the Stanford Graduate School of Business and Stanford Department of Economics, and a Research Associate at the National Bureau of Economic Research (NBER).

One of the most-cited authors on the economics of information, Brynjolfsson was among the first researchers to measure productivity contributions of IT and the complementary role of organizational capital and other intangibles.

Read more

Diyi Yang is an assistant professor in the Computer Science Department at Stanford, affiliated with the Stanford NLP Group, Stanford HCI Group, Stanford AI Lab (SAIL), and Stanford Human-Centered Artificial Intelligence (HAI). She is interested in socially aware natural language processing. Her research goal is to better understand human communication in social context and build socially aware language technologies to support human-human and human-computer interaction.

Read more