Redefine the execution of work with AI agents
AI agents reshape the way in which the work is carried out by offering tools that perform complex tasks and led by objectives. Unlike static algorithms, these agents combine planning in several stages with software tools to manage whole workflows in various sectors, including education, law, finance and logistics. Their integration is no longer theoretical – workers are already applying them to support a variety of professional tasks. The result is an environment of labor in transition, where the limits of human collaboration and machine are redefined daily.
Fill the difference between the capacity of the AI and the preferences of the workers
A persistent problem in this transformation is the disconnection between what AI agents can do and what workers want them to do. Even if AI systems are technically able to resume a task, workers may not support this change due to concerns about work satisfaction, the complexity of tasks or the importance of human judgment. Meanwhile, the tasks that workers are impatient to unload can lack mature AI solutions. This inadequacy presents an important obstacle to the responsible and effective deployment of AI on the labor market.
Beyond software engineers: a holistic evaluation of the workforce
Until recently, AI adoption assessments often focused on a handful of roles, such as software engineering or customer service, limiting the understanding of the impact of AI on professional diversity. Most of these approaches have also given priority to business productivity on workers' experience. They relied on an analysis of current use models, which does not provide prospective view. Consequently, the development of AI tools has failed in full base based on the preferences and real needs of people who do work.
The database of the work bank focused on Stanford surveys: capturing real workers' votes
The Research Team of the University of Stanford introduced an audit framework based on the survey which assesses the tasks that workers would prefer to see automated or increased and compare this to expert assessments of the capacity of AI. Using data from the Database O * Net Database of the US Ministry of Labor, the researchers created the work bank, a set of data based on the responses of 1,500 domain workers and assessments of 52 IA experts. The team used supported mini-interviews to collect nuanced preferences. He introduced the scale of the human agency (HAS), a five -level metric which captures the desired extent of human involvement in the completion of tasks.
Human agency scale (HAS): measure the right level of IA involvement
At the center of this framework is the scale of the human agency, which goes from H1 (complete control of the AI) to H5 (full human control). This approach recognizes that not all tasks benefit from complete automation, and each AI tool should not aim for it. For example, the tasks noted H1 or H2 – such as transcribing data or generating routine reports – are well suited to the independent IA execution. Meanwhile, tasks such as training programs or participation in security -related discussions have often been assessed at H4 or H5, reflecting the high demand for human surveillance. The researchers gathered two inputs: workers evaluated their desire for automation and the favorite level has a level for each task, while the experts evaluated the current AI capacity for this task.
Workbank badges: where workers kiss or resist AI
The results of the work bank database revealed clear models. About 46.1% of tasks received a great desire for automation of workers, in particular those considered to be of low value or repetitive. Conversely, significant resistance has been found in tasks involving creativity or interpersonal dynamics, regardless of AI's technical capacity to perform them. By superimposing workers' preferences and expert capacities, tasks have been divided into four zones: the “Green Light” automation area (high capacity and high desire), “red light” area (high capacity but weak desire), area of R&D opportunity (low capacity but desire) and low priority zone (low and low capacity desire). 41% of tasks aligned with companies funded by Y Combinator have fallen into low priority or red light zones, indicating a potential disparagration between start -up investments and workers' needs.
Towards the responsible deployment of AI on the labor market
This research offers a clear image of how AI integration can be addressed in a more responsible manner. The Stanford team discovered not only where automation is technically possible, but also where workers are receptive. Their framework at the level of tasks extends beyond the technical preparation to include human values, which makes it a precious tool for the development of AI, labor policy and labor strategies.
Tl; DR:
This article presents Workbank, a large -scale data set combining workers' preferences and IA expert assessments on 844 tasks and 104 professions, to assess where AI agents should automate or increase work. Using a new human agency scale (HAS), the study reveals a complex automation landscape, highlighting a misalignment between technical capacity and the desire of workers. The results show that workers welcome automation for repetitive tasks but who resist it in roles requiring creativity or interpersonal skills. The manager offers usable information for responsible deployment of AI aligned with human values.
Discover the Paper. All the merit of this research goes to researchers in this project. Also, don't hesitate to follow us Twitter And don't forget to join our Subseubdredit 100k + ml and subscribe to Our newsletter.
Nikhil is an intern consultant at Marktechpost. It pursues a double degree integrated into materials at the Indian Kharagpur Institute of Technology. Nikhil is an IA / ML enthusiast who is still looking for applications in fields like biomaterials and biomedical sciences. With a strong experience in material science, he explores new progress and creates opportunities to contribute.
