Honeybees “cheat” on math tests: examining the use of numerosity in numerical cognition tasks

Recently, impressive numerical abilities have been claimed for animals like honeybees. Here we critique whether the learning tests used are sufficient evidence for numerical cognition, or whether there are alternative explanations for the behaviour seen. We replicated the methods used in honeybee studies in which bees were trained to discriminate visual stimuli that varied in number of elements. Bees performed similarly to those in the original study. Additional control tests showed, however, that bees had learned continuous (non-numerical) cues of the stimuli that covaried with numerosity and not number itself. Further, we created a simple model containing just nine elements that learned the continuous cues of training stimuli only, with no numerical processing. This model was able to reproduce all behaviour in learning tests that had been considered indicative of numerical cognition. Our results caution a revaluation of recent claims of numerical abilities of honeybees, as well as other works using similar methods.