Structure in noise: Recurrent connectivity shapes neural variability to balance perceptual and cognitive demands in the human brain

Human behavior is remarkably complex, balancing the seemingly opposing traits of reliability and flexibility. In contrast, neural population activity is inherently time-varying, raising the question of how ever-changing activity supports both stable and adaptive behaviors. Using large-scale human intracranial electroencephalography (iEEG) covering the cortical hierarchy from sensory to association areas, we tested whether neural variability shapes perceptual and cognitive performance in a context- and demand-dependent manner. Our results show that neural variability is not random noise but hierarchically structured by a cortical gradient of recurrent connectivity, supporting the spatiotemporal unfolding from perceptual to cognitive processing. Reduced variability in sensory areas enhances the fidelity of sensory representations, whereas increased variability in association cortex reflects recurrent dynamics that support memory maintenance. In sum, neural variability captures the duality of human behavior-consistent during sensory processing, yet adaptable when needed.