“The rapid rise of compound AI systems (a.k.a., AI agents) is reshaping the labor market, raising concerns about job displacement, diminished human agency, and overreliance on automation,” said “Future of Work with AI Agents,” a recently published paper by researchers from the Stanford Institute for Human-Centered AI and the Digital Economy Lab. “Yet, we lack a systematic understanding of the evolving landscape.”
As explained in their project website, the Stanford researchers addressed this understanding gap by conducting a nationwide survey that collected data from 1,500 workers across 104 occupation to learn which of 844 tasks workers wanted AI agents to automate or augment, and which, for a variety of reasons, they didn’t want AI agents involved. In addition, they interviewed 52 AI experts to help them understand AI’s current technological capabilities and how those capabilities align with the workers desires. The researchers then compared worker desires alongside AI capabilities to identify opportunities and tasks that warrant reconsideration for automation.
Let me summarize their main findings.
Understanding workers desires and fears
Which occupational tasks do workers desire AI agent automation? For 46.1% of tasks, workers currently performing them express a positive attitude toward AI agent automation. The top three tasks they voted for are
- Tax preparer: scheduling appointments with clients;
- Public safety telecommunications: maintaining files of information relating to emergency calls; and
- Timekeeping clerks: issue and record adjustments to pay related to previous errors.
Why do workers want AI agent automation? The most cited reasons for pro-automation responses are:
- Automating the task would free up time for high-value work (selected in 69.4% of responses);
- The task is repetitive or tedious (selected in 46.6%);
- Automating this task would improve the quality of work (46.6%);
- The task is stressful or mentally draining (25.5%).
Where do workers resist AI agent automation? The three most common concerns that workers have about relying on AI automation in their work are:
- Lack of trust in AI systems’ accuracy, capability, or reliability (45%);
- Fear of job replacement (23%); and
- The absence of human qualities or capabilities in AI (16.3%).
Contrasting worker and AI expert perspectives delineate four task zones
Based on the survey data, the research team came up with an innovative approach for classifying the occupational tasks into four categories based on whether the workers desire to automate that particular task was high or low, and whether experts rated AI’s capability to automate the task as high or low, leading to four distinct zones of occupational tasks:
- Automation “Green Light” Zone: Tasks with both high workers desire to automate and high capability for AI automation in occupations like Tax Preparers and Mechanical Engineers. These are prime candidates for AI agent deployment with the potential for broad productivity and societal gains.
- Automation “Red Light” Zone: Tasks with high capability for automation but low desire for workers to automate in occupations like Logistics Analysts and Municipal Court Clerks. Deployment here warrants caution, as it may face worker resistance or pose broader negative societal implications.
- R&D Opportunity Zone: Tasks with high workers desire for automation but low AI automation capability in occupations like Technical Writers and Video Game Designers. These represent promising directions for AI research and development.
- Low Priority Zone: Tasks with both low workers desire and low AI automation capability, in occupations like Art Directors, and Ticket Agents. These are less urgent for AI agent development.
Opportunities for Human-Agent Collaboration: the Spectrum of Automation and Augmentation
“Traditional technology impact studies often ask: To what degree can this task be automated?,” wrote the Stanford researchers. “Besides this view of automation, we consider the view of augmentation — where technology complements and enhances human capabilities (Autor, 2015), — as this new wave of technology holds significant potential to augment human workers through human-agent collaboration, enhancing both productivity and work quality.”
Their article references a 2015 paper by MIT economist David Autor, “The History and Future of Workplace Automation,” where he explained that “tasks that cannot be substituted by automation are generally complemented by it.” Automation does indeed substitute for labor, but automation also augments labor, raising economic outputs in ways that often lead to higher demand for workers.
There is no established framework for comparing the various levels of automation vs augmentation. To fill this gap, the Stanford researchers introduced the Human Agency Scale (HAS), a five-level scale from H1 (no human involvement) to H5 (human involvement essential).
- H1: AI agent handles the task entirely on its own without human involvement;
- H2: AI agent only needs human involvement at a few key points;
- H3: AI agent and humans collaborate in an equal partnership;
- H4: AI agent often requires human input to successfully complete the task;
- H5: AI agent cannot function without continuous human involvement.
Human Agency Scale H3 (equal partnership) was the dominant worker-desired level for 43.2% of the occupations analyzed, followed by H2 (human involvement only at key points) at 35.6%; and H4 (human involvement the majority of times) at 16.3%. Both H1 (no human involvement) at 1.9%, and H5 (human involvement essential) were significantly less popular choices.
A Shift in Valued Skills
Finally, the Stanford researchers noted that as AI and automation redefine work, the worker skills that are most valued are likely to evolve. To explore this shift, they used the O*NET database from the US Department of Labor to match each occupational task to the skills it relies on. For each skill, they then estimated two key values:
- the current economic value of a skill based on wage data from the O*NET database, and
- the human agency level of the skill based on expert assessment of its susceptibility to replacement by AI.
Comparing the skills rankings based on these two dimensions uncovered three emerging trends that could potentially shape the future of human work:
- Shrinking demand for information-processing skills. Skills related to analyzing data and updating knowledge — while common in today’s high-wage occupations are less prominent in tasks that demand high human agency.
- Greater emphasis on interpersonal and organizational skills. Skills involving human interaction, coordination, and resource monitoring are more frequently associated with tasks requiring high human input, even if they are not currently prioritized in wage-based evaluations.
- High-agency skills span diverse aspects. The top 10 skills with the highest average required human agency (H4 and H5) encompass a broad range, from interpersonal and organizational abilities to decision-making and quality judgment.