Temporal dynamics of honeybee learning and decision-making are revealed by a novel automated PER system

Abstract

The proboscis extension response (PER) has been widely used for decades to evaluate honeybees' (Apis mellifera) learning and memory abilities. This classical conditioning paradigm is traditionally administered manually, and produces a binary score for each subject depending on the presence or absence of the proboscis extension in response to a stimulus - typically an odor which has been associated with a sucrose reward - to classify whether or not the bee has learned the association. Here we present a fully automated PER system which delivers stimuli in a more controlled manner, and thus standardizes the protocol within and between labs; further, the AI-facilitated behavioral scoring reduces human error and allows us to extract a richer meaning from the outcome. The automated frame-by-frame assessment goes beyond the binary classification of "learned" or "not learned", expanding the possibilities for many other measures. Using this method, we investigate the real-time decision-making processes of honeybees faced with difficult learning tasks. When posed with a quantitative (rather than qualitative, as in the case of different odors) PER association, honeybees show a pattern of rapid generalization to both the rewarded and non-rewarded stimuli, followed by a slowly acquired discrimination between the two. Our work lays the foundation for deeper exploration of the honeybee cognitive processes when posed with complex learning challenges.