Skip to content
back to the archive page
#AI

[2/3] Future of Work with AI Agents: Workers’ Preferences and Automation Potential (Category AI)

#AI #Economics #ML #Work

[2/3] Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce (Category #AI)

Continuing my discussion of the study, here are its findings.

Support for automating routine work Contrary to common fears, many workers want AI to take over routine, low-value tasks. Respondents rated automation positively—above neutral on a 5-point scale—for 46.1% of the tasks examined. Before answering, they were asked to consider both the risk of job loss and whether they enjoyed the task. Even so, automation was desirable for almost half the tasks. Supporters gave these reasons:

  • Freeing time for more important work: around 69%.
  • Getting rid of routine work: around 47%.
  • Improving results with AI: around 46%.
  • Avoiding stress: around 25%.

Concerns and resistance The survey also confirmed serious concerns about AI. The most common were:

  • Distrust of AI output quality: around 45%, including doubts about accuracy and reliability.
  • Fear of losing a job: 23%.
  • Lack of a human touch: 16%.

These concerns were particularly strong in creative and humanities-related work. In arts, design and media, for example, only 17% of tasks received positive automation ratings.

Four zones matching preferences with capabilities By comparing workers’ preferences with expert assessments, the authors divided tasks into 4 categories:

  1. Green Light: people want AI to perform these tasks and the technology can already do so. They are the first candidates for adoption and promise the greatest productivity gains.
  2. Red Light: AI can perform the tasks, but people do not want them automated. Adoption may encounter resistance or cause negative social consequences, so caution is needed.
  3. R&D: workers would like automation, but current models cannot handle the tasks. These are promising research directions with clear user demand.
  4. Low Priority: both the desire for automation and technical feasibility are low. These tasks can be set aside for the near term.

The analysis found substantial mismatches: around 41% of tasks fell into Red Light or Low Priority. A significant share of current adoption efforts therefore targets work people do not want automated or work the technology cannot yet handle. Y Combinator startups, for example, often target the red or low-priority zones, while many areas workers want—Green Light and R&D—remain underfunded. This highlights a gap between developers’ and investors’ interests and workers’ needs. It points toward redirecting effort to Green Light and R&D: areas of high demand, with further technological development needed where capabilities are missing.

A preference for collaboration over complete replacement Most workers prefer partnership with AI to full automation. On the study’s Human Agency Scale, or HAS, the most common ideal is H3, equal partnership: on average, 45.2% of respondents would like to work alongside an AI agent as an equal partner. Another approximately 35.6% prefer H4, an AI assistant under human control, with people making key decisions. Together, around 80% favor retaining a substantial human role.

Only a minority, around 19%, prefer either near-total AI autonomy at H1–H2 or, at the opposite extreme, minimal AI use at H5. People clearly do not want to hand work over completely: they want to retain control and critical decisions.

The conclusion is in the next post.

#AI #Economics #ML #Work