Do bees understand zero?

What was found

Bees trained to prefer the card with fewer elements were shown an empty card for the first time. They chose it above chance. Crucially, the preference was stronger when the alternative carried more shapes, which is the pattern you would expect if the empty set were being placed at the low end of a magnitude scale rather than treated as a novel object.

The simpler explanation

A bee trained to prefer fewer elements has, in effect, been trained to prefer less of something. An empty card is the extreme of less by every low-level measure at once: fewest shapes, least ink, least edge, lowest density. Distinguishing an empty-set concept from a monotonic preference for less is genuinely hard.

That objection is not a blogger's. It was published in 2021 by MaBouDi and colleagues in Proceedings of the Royal Society B, who re-examined the stimuli, argued that the choices track continuous properties such as spatial frequency, edge length and convex hull, and then built a model of nine neurons containing no numerical representation at all that reproduced the reported behaviour.

The reply, and where the argument stands

In 2026 a paper in the same journal answered it. Zanon and colleagues argued that the critique's stimulus analysis excluded the low spatial frequencies that dominate how a bee actually resolves an image, and that this introduced spurious correlations between visual features and number. Recalculating the stimulus properties across the range a bee can genuinely see, they report, leaves a residual sensitivity to numerosity that the low-level account does not explain.