Summary
A bioRxiv preprint reports that bumble bees adjusted opt-out choices to task difficulty and transferred the strategy to new tactile and working-memory tasks without retraining. The findings suggest that uncertainty-guided decision-making can occur in a brain with about one million neurons.
A bioRxiv preprint reports that bumble bees adjusted their choices according to uncertainty, sometimes accepting a smaller guaranteed reward rather than risk an error. The researchers also report that the bees transferred this opt-out strategy, without retraining, to new tactile and working-memory tasks.
Choosing and seeking information under uncertainty
In the tasks described by the authors, bees opted out more as perceptual judgments became difficult. An opt-out meant settling for a smaller sure reward instead of risking a wrong answer. The bees also paid a reward cost to seek information when uncertain, according to the abstract.
These behaviors point to two related choices: avoiding a difficult decision and actively seeking information. The researchers say a confidence-based decision model captured the bees’ choices, and that the strategy was stable across individuals.
Transfer beyond a single task
The authors report that the bees applied the opt-out rule to novel tactile and working-memory tasks without retraining, using what they describe as an all-probe design. The transfer suggests the behavior was not limited to one familiar perceptual task. The abstract does not give further details about the tasks or the design.
The finding bears on metacognition, commonly defined as monitoring one’s own cognitive states. The researchers interpret the results as evidence that a brain of about one million neurons can support an abstract, domain-general policy for responding to uncertainty. That possibility broadens the range of animals in which such decision strategies may have evolved.
The evidence is behavioral: choices and their fit to a confidence-based model. It is evidence for uncertainty-sensitive decision-making, rather than a direct measure of subjective experience or human-like conscious introspection. The report is a bioRxiv preprint, and its abstract gives no sample size or numerical effect estimates, which limits assessment of the scale of the findings.