Study links AI skill erosion to hard-to-reverse adoption
A modelling paper argues that reliance on AI can erode human skills, making widespread adoption bistable and difficult to undo once it happens.
In plain words
Researchers modelled what happens when people can do a task themselves or with AI help. In their model, skills grow with practice but fade with reliance on AI, so a population can settle into very different long-term states. Once adoption spreads, it may be hard to reverse even if AI's advantage fades.
What was announced
The arXiv paper, by researchers studying generative AI adoption, models a population of agents who perform a task either with their own skill or with AI assistance. Agents imitate more successful peers, while skills improve with practice and erode with reliance on AI. Analysing this analytically and through simulations, the authors find that skill erosion makes adoption bistable: populations with the same adoption level and mean skill can end up in opposite long-term states depending on how expertise is distributed between AI users and non-users.
A deterministic description tracking the mean skills of both groups, plus the variance of skills among non-users, predicted individual simulation runs as accurately as the simulations allowed; mean skills alone were not sufficient. The authors also model AI benefits improving over time: adoption can then jump abruptly, and a temporary benefit increase, such as during a period of low prices, can leave a population relying on AI after benefits return to earlier levels. They conclude that skill erosion creates a history-dependent switching cost that can make widespread adoption difficult to reverse.
What it does not show
- The source is an abstract, so methods, assumptions and full results are not available for verification.
- Findings come from analytical and simulation models of agents, not from measurements of real-world populations or workplaces.
Sources
Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 19 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.