Skip to main content

Insects Recognize Faces Using Processing Mechanism Similar to That of Humans

Conventional wisdom holds that the ability to recognize faces requires a complex mammalian brain. But some insects are surprisingly adept at this task

The wasps and bees buzzing around your garden might seem like simple-minded creatures. They build nests, forage for nectar, raise their young and then die, their lives typically playing out over the course of a single year or less. Some of these species rival humans and other primates in at least one intellectual skill, however: they recognize the individual faces of their peers.

More specifically, members of a species of paper wasp can perceive and memorize one another's unique facial markings and are able to use this information to distinguish individuals during subsequent interactions, much as humans navigate their social environment by learning and remembering the faces of family, friends and colleagues. Further, even certain insects that do not normally memorize faces in the wild can be trained to do so—and can at times even learn to tell human faces apart.

A popular theory of intelligence holds that the exceptionally large human brain evolved to cope with the challenges posed by having to learn and remember many individuals in complex societies. But the discovery that creatures with a brain less than 0.01 percent as large as our own can also identify individuals is forcing scientists to consider how this startling ability evolved and which features of insect brains make facial recognition possible. The answer to the last question, in particular, could help software designers to improve facial-recognition software.


On supporting science journalism

If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.


A Fortuitous Discovery As is often the way with scientific discoveries, the revelation that wasps see one another as individuals resulted from a lucky accident. As a young graduate student in 2001, one of us (Tibbetts) was working on a project focused on detailing the social lives of Polistes fuscatus paper wasps. The project involved painting colored dots on their backs, then videotaping colonies and tracking the interactions among the insects. One day Tibbetts accidentally videotaped a colony with two unmarked wasps. The data would be useless unless she could figure out a way to differentiate between the two insects. As she watched the tape, she suddenly realized that she could tell the unmarked wasps apart by looking at the yellow, brown and black stripes and spots that make up their natural facial markings. Could wasps, she wondered, do the same?

Tibbetts couldn't resist checking. She spent the next few days documenting the fantastic diversity of facial patterns in wasps and then tested whether the creatures could use these patterns as a guide to recognizing individuals. Using an extremely high-tech method—applying modeling paint with toothpicks—she changed a wasp's facial features and then observed the social consequences. Aggression is rare in wasp colonies, so if the wasp was treated more aggressively by nest mates after the makeover, the behavior shift would be proof that wasps pay attention to faces. As a control, she also applied paint to some insects without changing their appearances to rule out the possibility that the wasps were reacting to some aspect of the paint other than its visual effect. She found that the nest mates displayed far more aggression toward the visibly altered wasps than to the control subjects; for the control wasps, interactions with their nest mates proceeded in a business-as-usual fashion. The results showed that wasps do indeed use variation in facial patterns for individual recognition.

Tibbetts was amazed. To appreciate just how astonishing this finding was, consider how humans recognize faces. First, we must perceive a specific arrangement of unique facial features—such as nose, mouth, eyes and ears—and, in our mind, link the arrangement to more abstract information about the person, such as that he or she is our boss or our neighbor. Then we need to quickly recall that pairing each time we see that particular individual.

Interestingly, we learn faces faster and more accurately than we grasp many other types of complex visual information. For instance: if you go to a party, you can quickly memorize the faces of the attendees without any great effort. It would take substantially more time and effort to learn multiple unique, but visually similar complex patterns such as those used in Chinese script. Both faces and Chinese script are composed of multiple elements that come together into a larger whole, but we are better at learning faces than Chinese script because evolution has furnished us with brain adaptations specifically for learning faces. Indeed, in the human brain a region called the fusiform face area is dedicated to face processing. This processing mechanism is so highly specialized that it fails us if an image is simply turned upside down. Likewise, even small modifications to critical regions of a face, such as the eyes, can hamper our ability to recognize a familiar visage.

Although humans as a species excel at face learning, approximately 2 percent of us have some form of deficit. Most face-learning deficits are thought to be hereditary, although they can also emerge in adulthood after injury to the fusiform face area. Given the importance of social recognition in human societies, such disorders can be debilitating. In some extreme cases, people may even have trouble recognizing their spouses and their children, likening the task to trying to learn the identity of different rocks in the garden. In addition, the impaired social development seen in individuals with disorders such as autism may stem at least in part from face-processing deficits.

Given the importance of specialization to human face learning, Tibbetts pondered whether paper wasps could have independently evolved a similar specialization or whether they instead process faces in a different way. To find the answer, she first needed a reliable method for training wasps to attend to images she deemed “correct” while bypassing incorrect ones. Researchers typically train social insects such as honeybees to perform certain tasks by rewarding them with sugar when they make correct choices. Honeybees are eager to work for sugar because collecting food to share with nest mates is one of their primary jobs. Wasps, however, can survive for weeks without eating, so initial efforts to train them using sugar rewards failed. Tibbetts and her then graduate student Michael Sheehan, now at the University of California, Berkeley, eventually found that they could train the creatures to go to the correct image in a choice between two by delivering a small electric shock to the wasps when they selected an incorrect image.

With the new training technique, wasps learned to differentiate between respective pairs of five different types of images: normal wasp face images, simple black-and-white geometric patterns, whole caterpillars (the wasps' natural prey), wasp face images with the antennae digitally removed, or scrambled wasp face images. The wasps rapidly and accurately learned to select specific normal faces within only 20 trials but had greater difficulty learning to discriminate among images in the other four pairings. Most strikingly, simply removing the antennae from a wasp face image or rearranging the face components dramatically reduced their impressive face-learning capacity.

The difference in the wasps' ability to learn normal face images versus antennaeless ones provides very strong evidence that wasps have neural systems specialized for wasp face learning. Antennaeless faces are composed of the same colors and patterns as normal faces, but the wasp visual system does not reliably process and recognize the altered image as a face. The effect of antennae removal on learning indicates that wasps, like people, perceive faces through some kind of holistic processing mechanism. That is, instead of learning each facial feature separately, element by element, the wasp perceives and processes a face as a whole. Thus, for proper learning to occur, the elements must be intact and correctly arranged. The effect of antennae removal on the wasps' ability to learn faces parallels the way human face learning falters when an image of a face is upside down, image brightness is inverted or facial features are scrambled.

The occurrence of face specialization in both humans and wasps suggests that this mechanism could be more widespread in the animal kingdom than initially thought, evolving when social conditions favor it. In the case of P. fuscatus, nests are started by groups of queens that work together to survive. But the queens also compete with one another for reproductive dominance. In such situations, it probably pays for cooperating queens to be able to recognize one another and remember how each individual ranks in the dominance hierarchy, hence the evolution of face specialization in this species. By the same token, the face-specialization mechanism should not be present in animals that do not usually need to differentiate individuals.

[break]

To test this hypothesis, Tibbetts and Sheehan measured face learning in a close relative of P. fuscatus—Polistes metricus—which has a different social structure. P. metricus nests are typically founded by a solitary queen. With only one queen reproducing, group members have little to gain socially by being able to recognize individual faces. Tibbetts and Sheehan had previously shown that P. metricus wasps do not vary in their facial markings and do not naturally recognize individuals. They then posited that P. metricus also lacks a special cognitive mechanism for processing faces, unlike the more socially complex P. fuscatus. Their findings upheld this supposition.

When given the opportunity, P. metricus wasps can learn faces, but it is difficult for them, and they learn faces only about as quickly and accurately as other kinds of images. Furthermore, antennae removal has no effect on the speed or accuracy of their face learning, indicating that this species lacks a specialized holistic-type face-processing mechanism. Instead they process faces as they do any other images: as a collection of independently processed features—perhaps like we humans might learn Chinese script.

How Honeybees See Humans Given that P. metricus can slowly learn faces when trained to do so, despite lacking a specialized mechanism for this activity, one might wonder whether insects with small brains have some capacity to also learn faces of an entirely different type of creature: humans. Inspired by the early paper wasp findings, one of us (Dyer), who studies how bees process visual information, became interested in figuring out whether these insects could learn to tell people apart. In this study, he trained common honeybees to distinguish a target face from a so-called distractor face, presenting them with human visages from a standard neuroscience test. The faces are similar enough that human subjects typically make some errors in judgment. The bees received sweet-tasting sucrose solution for making correct visits to a target face and bitter quinine solution for incorrectly choosing a distractor face. Although it took a while for them to catch on, the bees learned to reliably discriminate the pair of target and distractor faces over the course of 50 trials. The bees also learned to select a target face from a group of novel human faces.

Other experiments using this training procedure revealed some striking similarities between honeybee and human face-processing strategies. First, although the bees learned faces slowly as compared with P. fuscatus wasps and humans, they were able to develop some ability to process faces holistically, even though they are not hardwired to do so, as P. fuscatus wasps and humans are. Second, honeybees were able to learn multiple viewpoints of the same face and interpolate based on this information to recognize novel presentations. For example, after a bee learns the front and side view of a particular face, it will be able to correctly choose a picture of the same face rotated 30 degrees, even if it has not previously seen this particular image. The honeybees' ability to learn faces is particularly unexpected because honeybee society is far simpler than that of the wasps, consisting of a single queen and a mass of essentially identical workers that do the same tasks. Honeybees do not have distinctive facial markings, and their interactions in the hive often rely on complex pheromonal signals rather than visual cues.

The results suggest that this line of research could be a boon to efforts to develop automatic face-recognition systems. Identifying the same face from different angles is often thought to be a difficult task and is one of the major challenges of machine face recognition. But the tiny brains of these bees are much simpler than primate brains, so once we figure out what processing tricks the bees use for solving such complex problems, we should be able to apply them relatively simply to machine vision to enhance face-recognition software.

[break]

Taken together, these studies in insects tell us something fundamental about how face recognition may have evolved. The simple system that P. metricus wasps and honeybees have that allows them to learn to recognize faces through experience even though they do not normally distinguish among individuals in their everyday lives may be based on the general pattern-recognition abilities these species use during foraging. It may also function as an intermediate step in the evolution of face specialization. When the forerunners of today's P. fuscatus wasps found themselves in a novel social environment in which telling individuals apart aided their survival and reproduction, the wasps could learn to identify individuals. Over time natural selection most likely built on that foundation, changing their brain to produce the face specialization that enhanced the wasps' ability to sort friend from foe quickly and reliably. The intermediate system allowed this adaptation to evolve rapidly: P. fuscatus and P. metricus are closely related, and their last common ancestor would have had the comparatively primitive face-learning system seen in P. metricus. Thus, P. fuscatus's biological adaptation to processing faces efficiently must have evolved recently, after its lineage diverged from that of P. metricus.

So the next time you are out in your garden, take a minute to appreciate the resident wasps and bees. There is ever so much more going on in their teensy brains than we could have imagined possible.

MORE TO EXPLORE

Visual Signals of Individual Identity in the Wasp Polistes fuscatus. Elizabeth A. Tibbetts in Proceedings of the Royal Society of London B, Vol. 269, pages 1423–1428; July 22, 2002.

Individual Recognition: It Is Good to Be Different. Elizabeth A. Tibbetts and James Dale in Trends in Ecology and Evolution, Vol. 22, pages 529–537; October 1, 2007.

Insect Brains Use Image Interpolation Mechanisms to Recognise Rotated Objects. Adrian G. Dyer and Quoc C. Vuong in PLOS ONE, Vol. 3, No. 12, Article No. e4086; December 31, 2008.

Configural Processing Enables Discrimination and Categorization of Face-like Stimuli in Honeybees. Aurore Avargues-Weber, Geoffrey Portelli, Julie Benard, Adrian G. Dyer and Martin Giurfa in Journal of Experimental Biology, Vol. 213, pages 593–601; February 15, 2010.

Specialized Face Learning Is Associated with Individual Recognition in Paper Wasps. Michael J. Sheehan and Elizabeth A. Tibbetts in Science, Vol. 334, pages 1272–1275; December 2, 2011.

Elizabeth A. Tibbetts is an associate professor at the University of Michigan. She studies how evolution shapes animal behavior and cognition.

More by Elizabeth A. Tibbetts

Adrian G. Dyer is an associate professor at RMIT University in Melbourne. He conducts behavioral studies to investigate how the visual systems in different animals process complex information.

More by Adrian G. Dyer
Scientific American Magazine Vol 309 Issue 6This article was originally published with the title “Good with Faces” in Scientific American Magazine Vol. 309 No. 6 (), p. 62
doi:10.1038/scientificamerican1213-62