Honeybee visual cognition: a miniature brain’s simple solutions to complex problems

Abstract

In recent decades we have seen a string of remarkable discoveries detailing the impressive cognitive abilities of bees (social learning, concept learning and even counting). But should these discoveries be regarded as spectacular because bees manage to achieve human-like computations of visual image analysis and reasoning? Here I offer a radically different explanation. Using theoretical bee brain models and detailed flight analysis of bees undergoing behavioural experiments I counter the widespread view that complex visual recognition and classification requires animals to not only store representations of images, but also perform advanced computations on them. Using a bottom-up approach I created theoretical models inspired by the known anatomical structures and neuronal responses within the bee brain and assessed how much neural complexity is required to accomplish behaviourally relevant tasks. Model simulations of just eight large-field orientation-sensitive neurons from the optic ganglia and a single layer of simple neuronal connectivity within the mushroom bodies (learning centres) generated performances remarkably similar to the empirical result of real bees during both discrimination and generalisation orientation pattern experiments. My models also hypothesised that complex ‘above and below’ conceptual learning, often used to exemplify how ‘clever’ bees are, could instead be accomplished by very simple inspection of the target patterns. Analysis of the bees’ flight paths during training on this task found bees utilised an even simpler mechanism than anticipated, demonstrating how the insects use unique and elegant solutions to deal with complex visual challenges. The true impact of my research is therefore not merely showing a model that can solve a particular set of generalisation experiments, but in providing a fundamental shift in how we should perceive visual recognition problems. Across animals, equally simple neuronal architectures may well underlie the cognitive affordances that we currently assume to be required for more complex conceptual and discrimination tasks.