Visual learning in honeybees and humans

Unlike honeybees, humans automatically extract predictive statistical properties of visual scenes using a sophisticated type of probabilistic learning, a study finds. Despite their small size, honeybees have sufficient brain power for processing visual information. However, whether honeybees and humans share comparable computational strategies to extract information based on the statistical structure of visual scenes remains unclear. Aurore Avarguès-Weber, József Fiser, and colleagues combined behavioral experiments on honeybees and 142 humans with computational modeling. The authors presented a series of pairs of scenes, which consisted of different black shapes placed at various locations on a grid. During each trial, humans and bees selected the more familiar shape-pairs out of two options simultaneously presented; honeybees indicated their choice by flying to and landing on the correct scene. Only humans showed automatic sensitivity to the conditional probabilities of shape-pairs, learning that the appearance of one specific shape always predicted the co-occurrence of another specific shape nearby. A sophisticated type of probabilistic learning model that could account for human performance in visual learning tasks did not capture honeybees' performance, which could be explained by a simple counting-based model. According to the authors, the rich internal representation obtained by complex probabilistic learning may contribute to the superior cognitive capacities of humans.

###

Article #19-19387: "Different mechanisms underlie implicit visual statistical learning in honey bees and humans," by Aurore Avarguès-Weber et al.

MEDIA CONTACTS: Aurore Avarguès-Weber, University of Toulouse, FRANCE; e-mail: aurore.avargues-weber@univ-tlse3.fr; József Fiser, Central European University, Budapest, HUNGARY; e-mail: fiserj@ceu.edu; Adrian Dyer, RMIT University, Melbourne, AUSTRALIA; e-mail: adrian.dyer@rmit.edu.au