Modeling the adoption of large models as an epidemic: the tipping point towards cognitive addiction

Ricard Solé and his co-authors transpose a three-state epidemiological model to the spread of LLMs and describe a threshold beyond which dependence becomes collective and difficult to reverse.

In an article on arXiv on September 3, 2026, Ricard Solé, Giulio Ruffini, Francesca Castaldo and six co-authors propose to read the dissemination of great language models with the tools of epidemiology. The analogy is not rhetorical: the authors construct a model of transitions between three user states — decoupled from the system, coupled, and permanently dependent.

The interest of formalism is to show the dynamics that linear observation of adoption lacks. Social transmission, re-use of outputs and collective reinforcement generate tipping points and technological locking effect. Having passed a critical threshold, small adoption increases are sufficient to rapidly move the entire population to a regime of persistent dependence, accompanied by sudden loss of cognitive competence.

This is a theoretical work and not an empirical measure of these losses: the model indicates where to look for threshold effects, not that they have already been crossed.