Flex-Eye-Bility: Dissociable pupil-linked surprise signals during adaptive task switching

Abstract

Cognitive flexibility allows rapid behavioral adjustment to unexpected changes in task demands, reflecting distinct effects of surprise magnitude and belief updating. Pupil-linked arousal is known to track surprise, but whether and how it supports our ability to adapt to the environment remains unknown. To address this, we tested how explicitly cued versus implicit task switches shape behavior by tracking computational surprise and pupil-linked arousal in low vs. high volatility environments. Participants (N = 16) performed a probabilistic task-switching paradigm with occasional faulty feedback while pupil size was recorded. Behavior showed improved post-switch accuracy recovery in explicit blocks and during high volatility. A computational Bayesian belief-updating model yielded dissociable surprise signals: outcome surprise about the feedback was associated with poorer post-switch accuracy recovery, whereas Bayesian surprise reflecting belief updating was associated with better recovery. Pupil deconvolution suggests separable early vs. late phasic pupil responses related to outcome and Bayesian surprise, respectively. Together, these results link cognitive flexibility to distinct computational and arousal signatures.