Goal-directed action selection relies on conjunctive representations that bind goals, actions and expected sensory outcomes

Abstract

Goal directed actions are performed to obtain specific sensory outcomes. However, theories of action control differ in how the brain encodes action-outcome predictions and whether they are encoded as separable components or integrated into conjunctive representations that support behavior and learning. Here, we tested whether predicted sensory outcomes are integrated into the neural representations underlying the selection of goal directed actions. Participants learned novel action-outcome contingencies and selected actions to generate specific sensory outcomes, while their brain activity was recorded using EEG. Combining time resolved multivariate decoding with representational similarity analysis, we tracked both feature specific and conjunctive neural representations over time. We found that expected sensory outcomes were integrated into high-dimensional conjunctive representations that could not be explained by a linear addition of individual features. The strength of these representations predicted trial to trial reaction times, alongside compositional action representations. Over time, the strength of compositional feature representations weakened, suggesting that the brain more selectively relied on conjunctive representations over the course of learning. These results provide neural evidence that predicted sensory outcomes are incorporated into behaviorally relevant conjunctive representations during action selection, consistent with theories that propose integrated action-outcome representations.