[1/3] Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce (AI column)

I read something fresh and interesting. study About the future of work in view of the emergence of AI agents (June 2025 year). This study was conducted by a group of scientists at Stanford University – in particular, researchers from the Institute for human-oriented AI. (HAI) The Stanford Digital Economy Laboratory. The authors include young researchers. (Yija Shao, Humishka Zoup, Yuchen Jiang, Jiaxing Pei, David Nguyen) under the guidance of recognized experts: Professor of Computer Science Dia Ian (Diyi Yang) economist Eric Brynjolfsson (Erik Brynjolfsson). Eric has a book called Machine, Platform, Crowd. 2017 year I've been talking about told. Also recently. handler A similar article from Erik Brynjolfsson, Bharat Chandar and Ruyu Chen about employment.Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence".

And now that it is clear that the research is from respected people, it is worth telling what the main idea of the authors is. They wanted to understand. What exactly do employees want from the introduction of AI in the workplace, and how modern capabilities of AI agents correspond to these desires? n. It reminded me of a whitepaper.**What Do Developers Want From AI?**From the guys at Google who I am. handler earlier. But in this study, the question goes something like this: what tasks people would like to automate with AI agents or, conversely, would prefer to leave under human control, and whether this coincides with what current technologies are capable of automating at all. Plus, the researchers were interested in how the integration of AI agents could affect the structure of skills and roles in the future.

To answer this question, the authors developed methodologyIt is based on surveys of employees and experts. They interviewed. 1 500 staff 104 various professions throughout the United States. Questions related to specific work tasks (Taken from the U.S. Department of Labor's O*NET Task Base)What each respondent actually does in their job. Employees were asked to indicate whether they want the AI agent to fully automate each task, only to help. (complement) Or they would prefer not to use AI at all. Audio interviews were used for deeper answers (voice-over) and 5-point scale Human Agency Scale (HAS)Reflecting the level of human participation from H1 (fully autonomous AI without human intervention) h5 (Fully manual execution, AI only as a tool). This new HAS score quantifies the preferred degree of human involvement in different tasks, rather than just “automate or not.”

Separately, the authors interviewed 52 AI expert (AI-agent developers) To assess the current technical feasibility of automating the same tasks. Experts estimated how modern AI systems (beginning 2025 d.) capable of performing each 844 the work tasks considered. This dual collection of data – from employees and from experts – formed the basis of the WORKBan database. k (AI Agent Worker Outlook & Readiness Knowledge Bank). WORKBank combines employee preferences and assessments of AI capabilities across a wide range of tasks (844 tasks 104 profession).

Using the data collected, the researchers built a “desire vs. capability landscape,” placing tasks into four zones depending on whether people want to automate them and whether AI can accomplish them. ||(Toast: “Let’s have a drink to make our desires match our possibilities.”)||. Text transcription of audio responses of employees was also analyzed. (fear-and-hope) using thematic modeling, and performed a comparison of tasks with the required skills and wages to identify the shift in the importance of skills in connection with AI.

Continuation of the study results in next post.

#AI #Economics #ML #Work