Reciprocal connections dynamically build consensus between neocortical areas.

The neocortex is organized into specialized areas. Although computations within individual areas have been well studied, it is unclear how these regions function collectively and reconcile potential conflicts to form coherent percepts and decisions. We investigated the joint dynamics of primary (V1) and higher-order lateromedial (LM) visual areas in mice using simultaneous multi-area electrophysiological recordings along with focal optogenetic perturbations to causally manipulate neural activity. We used data-driven nonlinear system identification to construct biologically constrained latent circuit models of both areas. This approach revealed that reciprocal excitatory connections between V1 and LM implement an approximate line attractor in their joint dynamics: this selectively slows the decay of congruent activity patterns while accelerating the decay of inconsistent ones, thereby dynamically achieving inter-area consensus. This mechanism predicts different timescales for consistent versus inconsistent activity patterns across areas, which we verified in our data. These findings, together with our mechanistic theory, identify dynamic consensus building as a general principle of distributed cortical computation.