EmpiricalRequires evidence

Near-criticality is the best-motivated operating point for a neural network in models, and in brains the data are consistent with it without establishing it. At a second-order phase transition a system keeps correlations up to system size while both dwelling in metastable states and moving among many of them; below the transition it freezes into ordered domains, above it correlations vanish. A network of excitable elements tuned to a branching ratio near one has the widest dynamic range, sensitive to weak input without saturating on strong input; subcritical networks cannot sustain input and supercritical ones explode. In brains, neuronal avalanche sizes follow a power law matching a critical branching process and fMRI correlation networks are scale-free. But summed Poisson processes or anatomy-driven waves can mimic the avalanche statistics, the fMRI results admit other readings, and the mechanism had no formal account. The signatures are necessary for criticality, not sufficient.

Collective intelligence