Item

Abstract

We use internal models of the external world to guide behavior, but little is known about how these cognitive maps are created. The orbitofrontal cortex (OFC) is typically thought to access these maps to support model-based decision-making, but it has recently been proposed that its critical contribution may be instead to integrate information into existing and new models. We tested between these alternatives using an outcome-specific devaluation task and a high-potency chemogenetic approach. We found that selectively inactivating OFC principal neurons when rats learned distinct cue-outcome associations, but prior to outcome devaluation, disrupted subsequent model-based inference, confirming that the OFC is critical for creating new cognitive maps. However, OFC inactivation surprisingly led to generalized devaluation. Using a novel reinforcement learning framework, we demonstrate that this effect is best explained not by a switch to a model-free system, as would be traditionally assumed, but rather by a circumscribed deficit in defining credit assignment precision during model construction. We conclude that the critical contribution of the OFC to learning is regulating the specificity of associations that comprise cognitive maps.