Mammals, including humans, are believed to have specialized neural circuitry for the problem of face recognition and detection. However, honeybees possess the remarkable capacity to identify human faces with 80% accuracy, despite a lack of any evolutionary need to do so. Further, these creatures do not need to be subjected to some horrible Ludovico Technique to perform this task; the only training needed is getting bees to associate the "target" face with a sucrose solution. This ability appears to be "zero-shot" and "out of distribution." Somehow, the honeybee seems to do this task by its own general-purpose visual reasoning.

This model is intended to compare a honeybee's remarkable visual intelligence with CNNs - can they be trained to do this task without even being trained on human faces? On the other hand, suppose CNNs are competitive with bees. Bees are also known to fail badly when the faces are flipped upside down - unlike humans, the knowledge of the "target" face when rightside-up doesn't seem to transfer. Possibly CNNs are superior at this.