Visual semantic tuning across the cortex shifts between tasks

Attention is a powerful mechanism that dynamically optimizes brain representations to prioritize behaviorally-relevant information. While previous studies focusing on single tasks have demonstrated that attention shifts tuning towards attended targets, real-world behavior often requires humans to switch between tasks with different demands. How does visual semantic tuning in the human brain shift to support the behavioral demands of different naturalistic tasks? To answer this question, we used voxelwise encoding models to compare visual semantic tuning across the human cerebral cortex between two distinct naturalistic tasks, movie watching and navigation. Results show that visual semantic tuning in the cortex differed substantially between tasks. Principal component analysis reveals that during navigation, tuning shifts increase the representation of vehicles and traffic signs compared to movie watching, and that these shifts were localized to distinct functional networks. These tuning shifts reconfigure the human brains representations of object categories based on their behavioral relevance. These findings demonstrate that visual semantic tuning in the brain dynamically shifts towards task-relevant information across naturalistic tasks, optimizing functional representations to achieve diverse behavioral goals. Significance statementIn real-world situations, we switch between many tasks with behaviorally distinct goals that require us to attend to different targets. Here we show that visual semantic tuning across the human cortex shifts between passive movie-watching and active navigation. These tuning shifts increase the representation of behaviorally-relevant objects, such as cars during navigation. Tuning shifts towards different object categories are distributed across distinct functional networks, suggesting that they engage different cognitive mechanisms. These results show that visual semantic tuning in the human brain is highly task-dependent and is optimized to represent behaviorally-relevant information.