Not yet human-reviewed
This source record was written automatically by the research pipeline and has not yet been reviewed by the maintainer or integrated into a topic page. Treat it as a provisional summary and verify claims against the original source.
Large-Language Models as a Cognitive Virus
Citation. Solé, R., Ruffini, G., Castaldo, F., Tuccio, M., Seoane, L. F., de Domenico, M., Elena, S. F., Krakauer, D. C., & Levin, M. (2026). Large-Language Models as a Cognitive Virus. arXiv:2609.03344.
Summary. This preprint develops a population-level dynamical model of LLM-mediated cognitive coupling, distinguishing weakly coupled, autonomously coupled, and persistently dependent users. It argues that social transmission and reinforcement can produce tipping points and hysteresis in cognitive offloading, while preserving unaided practice and verification can reduce dependence without requiring low overall AI use.
Key claims
- The model predicts a bistable region in which gradual increases in the transmission of substitutive LLM use can trigger discontinuous, path-dependent shifts toward persistent cognitive offloading.
- In the model, interventions that preserve routes back to unaided cognition and active verification can reduce dependency or prevent lock-in; recovery after a tipping point requires stronger change than prevention.
Evidence & limitations
- This is a theoretical mean-field model with illustrative parameter values, not an empirical estimate of LLM adoption, dependence, or cognitive competence; its proposed mechanisms require empirical testing with heterogeneous users and institutions.