The brain’s structural connections span several orders of magnitude in strength. Standard practice in connectomics is to filter out the weak ones — connections so faint they look like noise. Threshold the connectivity matrix, keep the strong links, analyze what remains. The implicit assumption: if a connection is weak, it doesn’t matter.

Wang et al. (arXiv:2505.24125) show, using data from nearly 1,000 participants, that weak structural connections are not noise. They predict general cognitive ability and memory performance. They are essential for reproducing realistic patterns of brain activity in simulation. They balance global integration against local specialization in ways that strong connections alone cannot.

The mechanism is nonlinear amplification. A weak connection between two regions contributes negligibly to any single transmission. But weak connections create alternative pathways — detours through which information can flow when direct routes are saturated or inappropriate. In a nonlinear dynamical system like the brain, these indirect routes can be amplified far beyond what their synaptic weight would suggest. The contribution of a weak link is not proportional to its strength.

A specific subset of weak connections — those linking visual and motor cortex to limbic regions — has disproportionate functional impact. These connections bridge the brain’s action and emotion systems through pathways that are anatomically tenuous but functionally critical. Remove them from a computational model and the simulation fails to reproduce observed patterns of brain dynamics.

The field’s longstanding practice of thresholding weak connections amounts to discarding the signal that predicts individual differences in cognition. The strong connections define the highway system. The weak connections define who drives where — and that turns out to be where the individuality lives.