Prior expectations can be used to improve perceptual judgments about ambiguous stimuli. However, little is known about if and how these improvements are maintained in dynamic environments in which the quality of appropriate priors changes from one stimulus to the next. Using a novel sound-localization task, we show that changes in stimulus predictability lead to arousal-mediated adjustments in the magnitude of prior-driven biases that optimize perceptual judgments about each stimulus. These adjustments depend on task-dependent changes in the relevance and reliability of prior expectations, which subjects update using both normative and idiosyncratic principles. The resulting variations in biases across task conditions and individuals are reflected in modulations of pupil diameter, such that larger stimulus-evoked pupil responses correspond to smaller biases. These results suggest a critical role for the arousal system in adjusting the strength of perceptual biases with respect to inferred environmental dynamics to optimize perceptual judgements.

Animals infer latent properties of the world from noisy and changing observations. Complex, probabilistic approaches to this challenge such as Bayesian inference are accurate but cognitively demanding, relying on extensive working memory and adaptive learning. Simple heuristics are easy to implement but may be less accurate. What is the appropriate balance between complexity and accuracy? We construct a hierarchy of strategies of variable complexity and find a power law of diminishing returns: increasing complexity gives progressively smaller gains in accuracy. The rate of diminishing returns depends systematically on the statistical uncertainty in the world, such that complex strategies do not provide substantial benefits over simple ones when uncertainty is too high or too low. In between, there is a complexity dividend. We translate these theoretical insights into specific predictions about how working memory and adaptivity should be modulated by uncertainty, and we corroborate these predictions in a psychophysical experiment.