Stanford Study of 1,500 Workers: What They Really Want from Artificial Intelligence

Researchers from the Stanford SALT Lab conducted a groundbreaking nationwide study that reveals a fundamental disconnect between what employees actually want from AI agents and the current direction of technology research and investment. The study, titled "The Future of Work with AI Agents," analyzed data from 1,500 workers across 104 different occupations and provides the first comprehensive view of how employees' voices are often lost in the debate over workplace automation.

Extensive WORKBank Database Reveals Workers' True Preferences

In collaboration with economists from the Stanford Digital Economy Lab, the researchers created an innovative survey- and audio-based auditing framework that maps the risks and opportunities of AI agents across the full spectrum of U.S. occupations. This worker-centered approach gathers insights directly from the people who actually perform the specific tasks. The result is the AI Agent Worker Outlook & Readiness Knowledge Bank (WORKBank), the first database to capture both AI agent capabilities and worker preferences.

The WORKBank database currently contains responses from 1,500 workers across 104 occupations and annotations from 52 AI experts, covering 844 occupational tasks. It uses the U.S. Department of Labor's O*NET database as its source of tasks and is designed to be easily expanded with additional tasks while reflecting evolving technological capabilities and worker preferences.

Fear and Desire—What Really Motivates Employees

Through an analysis of interview transcripts, the study identified workers' three most common concerns regarding AI automation. Lack of trust accounts for 45% of concerns, fear of job displacement for 23%, and the absence of a human touch for 16.3%. Interestingly, only 17.1% of tasks in the arts, design, and media sectors received positive ratings, indicating significant resistance in creative fields.

Despite these concerns, workers expressed a positive attitude toward AI automation for 46.1% of tasks (a rating above 3 on a 5-point Likert scale), even after explicitly considering concerns such as job loss and reduced job satisfaction. The most frequently cited motivation for automation was "freeing up time for more valuable work," selected by 69.4% of respondents. Other common reasons included the repetitive nature of tasks (46.6%), stress (25.5%), and opportunities to improve quality (46.6%).

Four Automation Zones Reveal Investment Misalignments

Contrasting the views of workers and AI experts made it possible to classify occupational tasks into four zones. The Automation "Green Light" Zone contains tasks for which both the desire for automation and AI capabilities are high—these tasks are prime candidates for the deployment of AI agents, with the potential for broad productivity and societal benefits.

The Automation "Red Light" Zone includes tasks for which AI capabilities are high but workers' desire for automation is low. Deployment here requires caution because it may face worker resistance or have broader negative societal impacts. The R&D Opportunity Zone contains tasks for which workers' desire for automation is high but AI capabilities remain low—these represent promising directions for artificial intelligence research. The Low Priority Zone then includes tasks for which both desire and capability are low.

The researchers used Y Combinator (YC) companies as a proxy and mapped them to tasks in the WORKBank database. Unfortunately, current YC investments are not focused on the Automation "Green Light" Zone and the R&D Opportunity Zone. A full 41.0% of YC companies are mapped to the Low Priority Zone and the Automation "Red Light" Zone, while many promising tasks in the "Green Light" Zone and Opportunity Zone remain underserved by current investments.

New Human Agency Scale Changes the View of Automation

A notable aspect of the auditing framework is that it goes beyond the typical focus on automation and also examines augmentation—where technology supplements and enhances human capabilities. To provide a common language for quantifying automation versus augmentation, the researchers introduced the Human Agency Scale (HAS), a five-point scale ranging from H1 (no human involvement) to H5 (human involvement essential). This new scale complements the SAE L0-L5 automation levels by quantifying the degree of human involvement needed to complete occupational tasks and ensure their quality, rather than focusing on an "AI-first" perspective.

Human Agency Scale level H3 (Equal Partnership) emerged as the dominant level preferred by workers in 47 of the 104 occupations analyzed. This indicates a strong preference for a balanced, collaborative partnership with AI across many professions.

Workers Prefer More Human Agency Than Technology Requires

Of the 844 tasks, workers prefer higher levels of human agency than experts consider technologically necessary for 47.5% of tasks. Notably, for 16.4% of tasks, the level preferred by workers is two levels higher than the experts' assessment. This finding potentially foreshadows friction as AI capabilities increase and the technology becomes capable of taking over a larger share of tasks than workers desire.

Tasks within the same occupation can vary significantly in their desired levels of human agency. The researchers suggest that the development of AI agents should account for different levels of human agency to enable higher-quality and more responsible adoption.

Skills Transformation—from Information Processing to Interpersonal Abilities

The study also analyzed shifts in human skills using the WORKBank database. Each occupational task was linked to the specific skills on which it relies. For example, the task "approve, reject, or coordinate the approval or rejection of lines of credit or commercial, real estate, or personal loans" (performed by financial managers) is mapped to "decision-making and problem-solving" and "guiding, directing, and motivating subordinates."

For each skill, the researchers estimated the level of human agency based on expert assessments and the average wage according to data from the U.S. Bureau of Labor Statistics as a measure of current economic value. By comparing skill assessments across these two dimensions, they identified three emerging trends that could potentially shape the future of human work.

The first trend is declining demand for information-processing skills. Skills related to analyzing data and updating knowledge, although common in today's high-paying occupations, are less prominent in tasks requiring high levels of human agency.

The second trend is a greater emphasis on interpersonal and organizational skills. Skills involving human interaction, coordination, and resource monitoring are more frequently associated with tasks that have high HAS levels, even though they are not currently prioritized in wage-based assessments.

The third trend is that high-agency skills span diverse areas. The top 10 skills with the highest average required level of human agency cover a broad range, from interpersonal and organizational abilities to decision-making and quality assessment.