A Dog’s View of Optical Illusions

In the Ebbinghaus-Titchener illusion, the two central circles have equal diameters. To humans, the circle surrounded by smaller circles appears larger. According to one study, the effect is reversed in dogs.

Byosiere decided to dig into the question for her doctoral research. She moved to La Trobe to work under the supervision of dog researcher Pauline Bennett, and devised a setup that would enable dogs to communicate what they were perceiving. The team created a small testing room with a touchscreen inside that would show various optical illusions and that dogs could interact with using their noses. Getting the first few animals—Lagotto Romagnolo dogs brought in by owners who had volunteered to work with the researchers—familiarized with the apparatus took months, Byosiere says. But the initial results, published in 2016, were eye-opening.

The team used various versions of the Ebbinghaus-Titchener illusion, in which two same-sized circles appear to humans to be different sizes due to arrangement of other circles around them. Each dog had been trained, using a generous supply of dog biscuits in exchange for right answers, to select whichever of the two center circles it perceived to be larger by tapping its nose to that part of the screen. The team had collected data using a range of control images, too, in which both the center circles and the overall stimulus size varied, to test whether dogs were biased towards choosing one of the images over the other.

To Byosiere’s excitement, the data revealed that dogs did indeed show susceptibility to the Ebbinghaus-Titchener illusion; they consistently selected one of the two identical circles in the illusory image more than the other. But rather than falling for the trick as humans do, picking the circle surrounded by smaller circles as being larger, the dogs did the exact opposite.

As the first step in training dogs to interact with visual illusions, researchers have to get the animals to associate tapping their noses to an object - in this case, a red bobble on a wand - with a reward.

COURTESY OF SARAH BYOSIERE

The findings were a striking illustration of how little science knows about canine perception, even though humans have for millennia had close relationships with dogs, says Byosiere, now the director of the Thinking Dog Center at Hunter College in New York. She notes that most research on dog cognition—up to three-quarters of published studies, according to one estimate—relies on visual-based tasks, many of which are adapted from studies on primates. But “we don’t necessarily know if [dogs] process the world in the same way that we do,” she says, so it’s sometimes unclear exactly what the findings from such experiments tell researchers about canine cognition. Understanding where canine visual perception differs from our own “could have a big impact on how we interpret these results.”

Visual tricks, for dogs

Researchers can test animals’ susceptibility to illusions by training them to interact with a screen or other apparatus, as Byosiere did in her 2016 study. Or, they can opt for what are known as spontaneous preference experiments, in which researchers present dogs with the same optical illusions using variously sized plates and portions of food, which dogs display interest in without any training.

In the Delboeuf illusion, the two central circles have equal diameters. To humans, the circle surrounded by the smaller ring appears larger. Two studies have found that dogs don't appear to perceive any difference.

Taking the second approach a few years ago, Christian Agrillo and colleagues at the University of Padua in northern Italy tried to test the susceptibility of dogs of various breeds to an illusion known as the Delboeuf illusion using dog treats bunched into circles. In the typical version of this illusion, the relative sizes of two identical circles are distorted by a ring encompassing each of them. Humans typically perceive the circle in the smaller ring to be larger than the circle in the larger ring—an effect thought to come from a visual quirk that causes us to round up the size of the former, because it’s almost the same size as the ring, but round down the size of the latter, because its size is so much smaller than the ring.

In Agrillo’s study, the researchers offered each of 13 dogs two plates of food positioned a meter apart. In the control condition, the choice was between two identically sized plates containing different quantities of treats. In the test condition, dogs had to decide between the same portion presented on plates differing in size. The team assumed that in both cases dogs would want to choose the portion they perceived to be larger. So if dogs are like humans in their perception of this illusion, the team hypothesized, they should choose the smaller plate in the test condition, on which the treat pile would seem larger.

They didn’t. While the dogs did go over to the larger portion in control trials, when it came to deciding between identical portions on different-sized plates in test trials, “their performance was basically random,” says Agrillo. Whether this means dogs are not susceptible to these visual illusions or just that the test conditions weren’t appropriate to detect it isn’t clear, he adds. Dogs in this experiment were rewarded with food whichever plate they chose, so there might have been little incentive to choose a portion that seemed ever so slightly larger. 

© KAILEY WHITMAN

Christian Agrillo and colleagues at the University of Padua in Italy carried out an experiment using this setup a few years ago, and assumed that dogs would choose the portion of food they perceived as larger. They found that dogs didn’t appear to perceive a difference in the portion sizes, even though humans would tend to perceive the food on the smaller plate to be more plentiful—a finding that could suggest that canine perception isn’t susceptible to this kind of illusion. See full infographic: WEB

Ponzo illusions use lines or grids to distort the relative sizes of two shapes. Humans typically perceive the rectangle on the left in this image to be larger. Several studies by one research group suggest dogs don't perceive a difference.

Findings that Byosiere and colleagues published at the same time that used their trained dogs and the touchscreen apparatus showing the image version of the Delbouef illusion seem to support the idea that dogs don’t fall for this optical trick. Byosiere has also more recently reported null results for dogs trained on various versions of the Ponzo illusion, which for humans distorts the size of identical shapes by surrounding them by lines or grids. 

Although “absence of evidence isn’t evidence of absence,” Agrillo says, the results so far suggest that dogs may not be susceptible to these particular visual tricks—something researchers may need to bear in mind when designing future experiments on canine cognition. 

Interpretation challenges for visual illusion experiments

For some types of visual illusions, researchers have found seemingly similar responses in dogs and humans, but a couple of studies also underline the need for caution in carrying out and interpreting experiments on canine perception. A few years ago, researchers at the University of Lincoln in the UK conducted a study where dogs interacted with a touchscreen showing the Muller-Lyer illusion, in which two identical lines seem to have different lengths due to the direction of arrowheads positioned at either end. The team found that dogs trained to pick the longer line consistently selected the stimulus with the arrows pointing inwards, just as humans performing the same task tend to do—offering a possible signal that dogs share humans’ perception of this particular illusion. 

In the Muller-Lyer illusion, the lengths of two identical lines appear distorted by arrowheads positioned at either end. Humans typically perceive the upper line as longer. Dogs trained to pick the longer line also pick the top line, but these findings might be explained by the animals choosing the larger stimulus overall, rather than perceiving the horizontal line as longer, as humans do.

But extra control trials and careful analyses of the data by the researchers performing the study revealed an alternative explanation for the findings: Dogs weren’t choosing the inward-arrow line based on perceived line length, they were choosing the biggest stimulus overall.

Experimenters try to eliminate this sort of alternative scenario in visual illusion studies. In their Ebbinghaus-Titchener experiments, for example, Byosiere and colleagues used multiple images with circles of various sizes to rule out the possibility that dogs were selecting images based on which was larger overall—a behavior that would also cause them to show the reverse effect to humans—instead of choosing images based on the central circle in the illusion. 

Confounding factors are a persistent concern in behavioral choice experiments, particularly in studies where researchers don’t know exactly what dogs are seeing or paying attention to, says Gregory Berns, a neuroscientist at Emory University. “When we design experiments, we can’t help but design them from a human perspective,” says Berns, who began investigating canine cognition nearly a decade ago and was one of the first researchers to train a dog to sit inside an MRI scanner. In studies that measure canine behavior as an outcome, “if a dog doesn’t do what you want it to do, it’s often not clear whether the dog doesn’t understand what you’re asking of it, or it understands but just doesn’t want to do it, or has a different motivation.” Studies often assess dog behavior at the group level in order to detect subtle trends in perception, he adds, but this may overlook substantial variation in perception and cognition among individual dogs.

© KAILEY WHITMAN

Sarah Byosiere and colleagues at La Trobe University in Australia reported an experiment using this setup in 2016. They found that dogs trained to pick whichever black circle they perceived to be the larger of the two consistently selected the one surrounded by larger blue circles—the opposite effect from that seen in humans—suggesting that dogs’ and humans’ visual perception may differ for some aspects of this illusion. See full infographic: WEB

Researchers who study illusions using behavioral experiments counter that, even with these challenges, the approach offers at least a preliminary way to explore how dogs see the world. As Byosiere explains, these studies “can give us a good first glimpse into the ways that they’re visually processing their surroundings. It can help us figure out where the parallels are and where they aren’t,” and potentially design better research studies in the future. 

She and other researchers have already started to consider perceptual explanations for dogs’ responses to certain illusions. In cases where dogs and humans show different responses—that is, dogs choose the opposite stimulus to humans or don’t display susceptibility at all—it may be that dogs’ visual systems are responding to different sorts of visual stimuli, she says. Humans are known to be particularly good at perceiving global patterns in images made up of smaller elements, for example. Canine perception, by contrast, may be more geared toward picking up the local stimuli in those images, notes Byosiere—a phenomenon that might explain why dogs respond differently than humans do to the Ebbinghaus-Titchener and Ponzo illusions, both of which need to be perceived at the global scale in order for the illusion to work in the intended way. 

Navon figures such as these can be perceived either globally—in this case, as an E and R— or locally, as an arrangement of Ns and Ss here. Some research suggests that humans are better at perceiving global stimuli than dogs are.

Such interspecies differences could reflect differing evolutionary pressures in dogs and humans, she adds. Berns agrees that there’s some evidence in the scientific literature that dogs don’t have as strong a visual preference for global stimuli as humans do, although he cautions that there have been only a handful of studies on the topic. 

Chouinard notes another way to understand perceptual differences between dogs and humans: how likely an animal is to perceive similar stimuli as identical to one another, rather than notice subtle differences among them. His and Byosiere’s work could suggest that dogs are less likely than humans to perceive differences among similar stimuli, he adds.

Studies that find shared responses in dogs and humans require careful interpretation, too. Elias Garcia-Pelegrin, a PhD student in psychology at the University of Cambridge studying cognition in crows, explains that if an animal shows susceptibility to an optical illusion known to trick people, shared neural machinery might be the cause—but equally, it might not. “Just because they have a reaction that is similar doesn’t mean that it’s underlined by the same cognitive processes.”

A little bit of magic

Behavioral experiments on visual illusions in dogs aren’t the only way to explore canine perception. Functional MRI studies, such as the ones carried out by Berns, involve imaging the brains of dogs that passively view images on a screen so that researchers can observe the neural signals associated with various types of stimuli. His team recently posted a preprint suggesting that, based on fMRI experiments, dogs don’t generalize easily from three-dimensional objects to two-dimensional images of those objects, or vice versa—a finding that could itself have implications for how scientists design vision-based tasks to measure canine perception. 

Emory University neuroscientist Gregory Berns has used fMRI to study dog perception.

EMORY UNIVERSITY

Other researchers are exploring how to present different types of illusions to animals. Garcia-Pelegrin, for example, was one of several coauthors on a recent perspective article arguing that magic tricks—even something as simple as using a sleight of hand to make an object seem to disappear—can be a powerful research tool in studies of animal perception. “Magicians use very elaborate techniques of deception that capitalize on some of our blind spots in attention and perception,” he says. “If other, nonhuman animals have analogous or homologous mechanisms of attention and perception, magic tricks should work on them.”

Magic tricks or other dynamic optical illusions are clearly more complex than static 2-D images, acknowledges Garcia-Pelegrin, who is experimenting with sleight-of-hand tricks to make objects disappear for the crows he studies. Observing a magic trick invokes not just visual perception but also other cognitive factors such as violation of expectation (when something surprising or unpredicted happens) and object permanence. It’s also hard to design control conditions for a magic trick experiment, notes Byosiere, though she adds that her team had a magician visit the lab at the end of 2019 to discuss exactly these sorts of studies. 

For now, Byosiere and her team are testing another dog-fooling trick, one that made its way around the internet a couple of years ago. In it, a dog owner raises and lowers a large blanket like a screen in front of themselves a few times while their pet looks on; then, on one of the raises, the owner disappears behind a wall and drops the blanket, giving the dog the impression its owner has disappeared into thin air. 

Byosiere’s group recently launched a citizen science project titled “What the Fluff!?” to study how animals respond to this illusion. “We’re asking owners to do this at home with their dogs,” she says. “We’ll be analyzing the footage and seeing if we can make any conclusions about object permanence and violation of expectation in that kind of magic trick.” She’s continuing her work with trained animals, too, to disentangle the basic questions on dog cognition that first motivated her back in Australia, she says. “Sometimes the best ideas, I guess, come from barbecues.”