Uncertainty-Guided Decision-Making in Bumble Bees

Abstract

The evolutionary origins of cognitive monitoring remain contested. Metacognition, the capacity to monitor one's own cognitive states, has long been linked to complex vertebrate brains, yet whether a miniature brain can evaluate the reliability of its internal representations is unknown. Here we demonstrate that bumble bees engage in uncertainty-guided decision-making. Bees dynamically adjusted opt-out choices according to perceptual difficulty, settling for a smaller guaranteed reward to avoid errors, and actively paid a reward cost to seek information under uncertainty. Without any retraining, bees transferred this 'opt-out-under-uncertainty' rule to novel tactile and working-memory tasks under an all-probe design. This strategy was stable across individuals and accurately captured by a confidence-based decision model. Our findings suggest that a brain of approximately one million neurons can support an abstract domain-general uncertainty-monitoring policy, indicating that the neural substrates for uncertainty-guided decision-making may be far more ancient and widely distributed than previously assumed.