Machine Learning

Large-Language Models as a Cognitive Virus

Ricard Solé, Giulio Ruffini, Francesca Castaldo, Marco Tuccio, Luis F. Seoane, Manlio de Domenico, Santiago F. Elena, David C. Krakauer, Michael Levin

Featured September 9, 2026

Original Paper

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

Thinking about how people use AI like a spreading sickness, this paper shows that too much AI use can suddenly make many people overly reliant, like getting 'locked-in,' and it's much harder to go back to thinking on their own.

In depth

The paper introduces a novel framework that models the diffusion of Large-Language Models (LLMs) through human populations using an epidemiological analogy. It proposes a dynamical system describing transitions between uncoupled, autonomously coupled, and persistently dependent users, revealing that social transmission and collective reinforcement can lead to tipping points and technological lock-in, where a sudden, irreversible loss of cognitive competence can occur. The framework also identifies strategies for "cognitive immunization" to mitigate these risks.

Key Takeaways

1
LLM adoption can be modeled as a cognitive virus, spreading through social and institutional exposure and leading to different states of human-AI coupling.
2
The model predicts tipping points and hysteresis, meaning that beyond a critical adoption rate, populations can abruptly shift to persistent LLM dependence, and reversing this state requires a much larger reduction in transmission.
3
The framework identifies specific parameters for cognitive immunization, allowing for interventions to prevent harmful dependency while still enabling beneficial human-AI integration.

Conceptual Flow

HIGH LEVEL

Methodology: Modeling LLM Spread Like a Virus

The paper treats how people start using AI, and how much they rely on it, like a disease spreading through a crowd, changing how people think.

People Not Using AI

People Using AI Autonomously

People Dependent on AI

Move Between States

Population Dynamics

Results: Sudden Shifts in Thinking

They found that even small changes in how AI spreads can suddenly make many people dependent, and once dependent, it's very hard to go back to thinking independently.

Slow AI Adoption

Fast AI Adoption

Cross Tipping Point

Sudden Dependence

Hard to Reverse

This breakdown was generated by SciGrove. Get the same analysis — intuition, storyboard, peer review, a runnable prototype and a glossary — on any paper you upload or paste a DOI for.