AI as a Cognitive Virus: An Epidemic Model of LLM Dependence

A group of nine authors — among them Ricard Solé, David Krakauer and Michael Levin — has published a paper on arXiv with the provocative title “Large Language Models as a Cognitive Virus.” The preprint appeared in the “Physics and Society” section and examines not the technical side of neural networks but the social dynamics of their penetration into human culture.

Infection by Analogy

The authors propose looking at the spread of LLMs the way epidemiologists look at an infection. The use of these models “propagates” through the population like a virus and gradually embeds itself in cognitive and cultural practices. The model distinguishes three states: people with no connection to the tool at all, those connected to it from time to time, and those who have crossed over into a state of stable dependence.

The key mechanism is the interplay of three forces: social transmission (people pick up the habit from one another), “recovery” (a return to working on one's own), and collective reinforcement, in which widespread use makes abstaining ever less likely. It is from the combination of these factors that tipping points and technological lock-in arise — a state that is hard to escape.

The Avalanche Effect

The paper's most alarming conclusion is the possibility of “runaway dynamics.” Once a critical threshold has been crossed, even a small increase in the number of users can trigger an abrupt shift of the entire population toward stable dependence. This, in the authors' view, is accompanied by a sudden loss of cognitive competence: the skills people have delegated to the machine atrophy quickly and in many people at once.

This is not a linear process but precisely a collective phase transition — like water coming to a boil rather than heating up gradually. Hence the aptness of the viral metaphor: an individual may fail to notice the moment when a “gentle aid” turned into a necessity.

A Vaccine Against Dependence

The same mathematics, however, also points to a defense. The authors speak of the possibility of “cognitive immunization”: if you reduce the intensity of the habit's social transmission and make it more reversible — that is, make giving up constant use simple and painless — the system can be held on the healthy side of the threshold.

The work remains a theoretical model, twelve pages long with three graphs, and makes no claim to empirical measurement of real-world society. But it offers a language in which the risk to cognitive autonomy can be discussed — not in terms of the benefit or harm of a particular tool, but in terms of threshold effects beyond which turning back becomes costly.

Source: Hacker News