Technological lock-in
Definition here: sole-2026-llm-cognitive-virus’s term for path dependence in which a population stays in the coupled, cognitively offloaded state after the pressure that produced it has fallen.
The mechanism in the model
The dynamics carry two thresholds. The saddle-node at lambda_SN = 2*sqrt(kappa*rho) introduces a stable coupled
branch. The transcritical at lambda_TC = rho + kappa is where the uncoupled state loses stability. When kappa > rho, the first arrives before the second, so for 2*sqrt(kappa*rho) < lambda < rho + kappa both the uncoupled and
the coupled state are stable, separated by an unstable branch. The width of that region is lambda_TC - lambda_SN = (sqrt(kappa) - sqrt(rho))^2.
Raising lambda leaves the population near U = 1 until lambda_TC, where it jumps. Lowering lambda does not
reverse the jump until lambda_SN, which is lower. That loop is the lock-in, and it makes prevention cheaper than
reversal.
The runaway
The paper also describes a runaway transition. Past a critical point, greater reliance on LLMs promotes further
cognitive offloading, which reinforces the reliance. The effective-potential picture shows the minimum for autonomy
flattening and the minimum for offloading deepening. Once the autonomy well disappears at lambda_TC, the state rolls
toward the low-competence basin. The authors stress that this can happen even when individual adoption is gradual.
Limits of the model
The minimal model fixes the attractor structure but not the speed of the transition in real time, and the paper says
so. The thresholds also rest on a mean-field approximation that absorbs network structure into lambda. The authors
flag that heterogeneous and correlated networks can shift them.
Related
- cognitive-immunization: the intervention framework the lock-in motivates
- cognitive-virus
- cognitive-offloading
- sole-2026-llm-cognitive-virus