Physicists Model LLM Adoption Like an Epidemic — With Tipping Points Into Dependence

A new arXiv preprint from physicist Luis Seoane borrows the mathematics of viral spread to describe how large language models are working their way into everyday cognition. The model tracks people moving between three states — not using LLMs, using them casually, and being persistently dependent on them — and treats adoption as something transmitted socially and reinforced collectively, much like a contagion moving through a population.

The central warning is nonlinearity. Rather than a smooth, gradual uptake, the dynamics produce tipping points and technological lock-in: once adoption crosses a critical threshold, a small nudge can flip an entire population toward entrenched reliance, accompanied by what the paper frames as sharp drops in cognitive competence. The same equations that describe runaway dependence also point to the antidote — ‘cognitive immunization’ — achieved by slowing transmission and keeping the transition reversible so people can back out of dependence.

The framing is provocative and deliberately loaded, and as a society-and-physics modeling exercise it offers no empirical measurement of actual skill loss — the ‘cognitive competence’ term is a parameter, not a finding. Still, it’s a useful lens for a debate usually dominated by productivity claims, reframing questions about AI’s societal footprint around collective phase transitions and the erosion of cognitive autonomy rather than individual choice.