Research Warns AI Agents Are Pushing Humans Out of the Decision-Making Loop, Eroding Cognitive Skills
A new academic paper argues that increasing AI Agent autonomy is weakening humans’ ability to oversee and audit these systems.
On August 24, 2026, a position paper titled "AI Agents Push Humans Out of the Loop" was published on the academic repository arXiv. Its authors are Margaret Mitchell and Avijit Ghosh of Hugging Face, together with Samir Passi of Data & Society.
The paper focuses on safety risks as AI Agents gain greater autonomy. While human oversight is commonly proposed as a solution, the authors argue that current development practices not only hinder effective oversight but may also gradually erode human cognitive abilities through prolonged AI use.
To address this, the authors recommend making the goals and cognitive needs of human overseers a top priority in AI development. They propose design measures such as mandatory pauses for reflection or review, alongside organizational protocols including training and job-rotation policies, to prevent human skills from deteriorating over time.
Overreliance on AI could weaken people’s own analytical thinking and decision-making abilities—a risk Thai organizations and users should consider when adopting autonomous AI systems.