Resize Text: 
A journey into the "uncanny" place where the not-quite-human can be found.
On two recent vacations, I found myself in two buildings as different as almost logically possible. In Philadelphia, the stately, brick Mütter Museum of medical abnormalities has a wrought-iron fence and marble foyer with the names of benefactors chiseled in stone. The museum itself looks fittingly like a giant antique curiosity cabinet. Visitors remained hushed, as if in reverence, as they peered at fetuses in jars; pictures of bulging, face-twisting tumors; even plaster casts of Siamese twins who shared one skull. Woozy when leaving the grand hall, I understood the need for what looked like red Victorian fainting couches in the foyer.
Before that, I had visited a video-game parlor in Tokyo, a dark, beepy hive of pink and green strobe lights. The building was utilitarian, with low ceilings and metal stairs, and teenagers and old men crowded around flashy boxes. Despite ten thousand distractions, I felt compelled to watch “Dance-Dance Revolution,” where people simulate dancing in a nightclub. Some gamers were amazing at Dance-Dance, some comically awkward (myself included). Yet some of my strongest memories came between dancers, when virtual club-goers appeared onscreen. For digital art, I could recognize how well done they were; but their grayish skin, warped faces, and glitchy pumping and thrusting creeped me out. Inexplicably, I felt hints of the same revulsion I later would in Philadelphia. But why? Why did I feel uneasy in front of digital creatures who had never lived, or suffered?
Because I’d stumbled into “The Uncanny Valley.” It’s a theory that says the closer virtual creatures (like robots and video-game characters) get to looking human, the more harshly actual humans judge them.
As a species, we find animal faces or cartoon faces cute because we enjoy finding human likenesses in things. But for quasi-human creatures, our brains flip things around: Instead of cooing over similarities, we magnify differences and fixate on flaws.
The uncanny valley has shaped work in robotics and video games for years, but scientists have recently realized that it similarly informs neurology, psychology, aesthetics, and ethics.
Humans almost certainly evolved this sense of the uncanny, and if humanoids, androids, and artificial intelligence begin to penetrate our daily life, the uncanny valley may also dictate how we evolve in the future.
***
Given the island nation’s love of electronics, it was inevitable that an engineer from Japan, Masahiro Mori, would first discover “bukimi no tani,” the uncanny valley. By 1970, Mori and colleagues had built a generation of robots that might pass for human at a glance. But only at a glance. In fact, Mori began to question the assumption that the more he “succeeded,” the more real humans would embrace his creations. Being an engineer, he decided to graph the tendency. According to his original paper on the subject, if you plot a robot’s human likeness on the horizontal axis, and its resulting familiarity, or likeability, on a vertical axis (below), the likeability increases only to a certain degree. At some vague point, as a clunky robot like the Jetsons’ maid takes on flesh and human qualities, our mind reclassifies it as a creepy android or zombie. When that happens, our regard plummets.
Mori felt his insight was a prescription: Don’t aim for verisimilitude with robots, he cautioned, because you risk overshooting, crashing into the valley, and alienating people. The idea became conventional wisdom, especially among Japanese government officials who have staked millions of dollars on developing androids to pick up the labor slack for the nation’s aging population. Designers, though, chafed at the idea that they shouldn’t even try to crawl up and out of the valley. Many denounced Mori’s theory as pseudoscientific because he laid it out intuitively, without hard data. However, data from recent psychological experiments buttresses the idea of an uncanny gulf in the human mind.
When scientists show people pictures of obvious humans and obvious robots, people generally approve. But human-android composite faces and android mug shots inspire disgust. In 2005, participants in one study called pictures of a state-of-the-art robotic female head “weird,” “frightening,” “scary,” and “very sickly.” Others added, “The woman figure kind of freaked me out,” and “The woman seemed to be dead ... given the chance I would have preferred not to see it.” Overall, results from both the United States and Japan suggest that two factors shove people into the uncanny valley, and the results are robust, varying little for differences in age or gender.
One trigger is mismatched parts. Humans can usually stomach a doll’s eyes on a doll. Seeing them digitally grafted onto humans, though, stirs alarm. The other trigger is features that don’t look right by themselves: synthetic hair, an unnatural patina, a droopy mouth. Awkward eyes are especially damning, says Thalia Wheatley, an assistant professor of psychology and brain science at Dartmouth College. “It’s hackneyed to say that the eyes are the window to the soul,” she says, “but we get so much information from the eyes.” Altogether, though, our minds can get snagged on any one of those features, and when it does, we’re surprisingly unforgiving. One paper calls the uncanny valley the “uncanny cliff” to emphasize the steepness of the descent.
These triggers become even more sensitive when robots or digital pictures move and interact with people. “When human muscles move, they usually influence more than one joint,” says Karl MacDorman, a leading android researcher at Indiana University who often collaborates with Japanese designers. But, he says, “robots may look strange when their joints move independent of each other.” Also, moveable parts like fingers or limbs cannot jerk too much, and any exposed skin has to be the right temperature and texture. Even tougher to get right are facial expressions because lips, tears, blushes, eyebrows, and words all have to coordinate. Note that on Mori’s graph, although the highs are higher for moving creatures than for still ones, the lows are lower, too. A corpse is unsettling, especially if the face looks plastic. But only a moving corpse scares the bejesus out of you.
So far the uncanny valley has influenced primarily robotics engineers and computer artists, but the theory provides insight into other disciplines as well, like art. “Painters tend to enlarge their subjects’ eyes, even in self-portraits,” says MacDorman, but “many of the rules taught in art school don’t apply to highly realistic characters.”
To take one example, one of the most influential portrait artists of the past half-century is Alex Katz. His pictures are instantly recognizable because he always uses broad planes of color to create faces. Darker splotches suggest shadows or the outline of a nose, but even those splotches are almost always uniform, and he rarely shades or blends much. Although not considered abstract, Katz’s paintings are abstract in some sense because he counts on our minds to add a human texture to monolithic patches of color. On Mori’s scale, Katz is perched on the first peak.
Now contrast Katz with a current darling of the New York art scene, John Currin, whose work slides into the uncanny valley. Though capable of rendering normal-looking people, Currin instead paints grotesques: people with painfully elongated necks, bulbous buttocks, huge foreheads, and starved limbs. For subjects, he often portrays high-society types we’re not supposed to like anyway, and Currin makes sure they repel us, as an over-earnest android might. Of course, we’re not really supposed to hate Currin’s subjects, who are, after all, human beings. But that tension defines his work.
The tension also highlights another area the uncanny valley helps make sense of, human interactions. As MacDorman writes, “a very humanlike robot may provide the best means of finding out what kinds of behavior are perceived as human, since deviations from a human ‘other’ are more obvious.”
In a simple case, Japanese scientists are using androids to study why humans break eye contact during conversation. On a deeper level, psychologists like Sherry Turkle, at the Massachusetts Institute of Technology, want to understand how robots push “Darwinian buttons” by making eye contact and responding during conversations with smiles or nods. Turkle also has discovered, as she described in the online science forum Edge.org, that people who spend a great deal of time with social robots betray “a certain fatigue with the difficulties of dealing with people,” who don’t push buttons so reliably. “A female graduate student ... told me that she would gladly trade in her boyfriend for a sophisticated humanoid robot as long as the robot could produce what she called ‘caring behavior.’” Disturbingly, Turkle added, a few people even began to view robots as more real and suspected that their lovers and friends were simulating emotions like affection. Most of those robots stopped short of the uncanny valley; but if designers ever do crawl out (and recent evidence suggests that they’ve started to, at least with flat pictures), those problems could deepen. Overall, Turkle has written that she worries artificial beings “might be good for us in the ‘feel good’ sense, but bad for us in our lives as moral beings.”
The uncanny valley can also help explain another moral dilemma—why humans shun people who look different. Try as we might to staunch it, uncanny disgust can spill out in our everyday lives. Sometimes this is innocuous. Most people don’t consider the shock of a bad facelift a moral quandary. But sometimes it digs into uncomfortable territory. What about seeing a stroke victim who cannot move one paralyzed side of his face? Or a person with microcephaly, a congenital condition that deflates the skull? Even a drooping eyelid or a lazy eye (remember how important eyes are) can push Darwinian buttons of distress. All of this, of course, makes any feeling person feel awful. But one reason to study why humans slip into uncanny valleys is to help us face, and overcome, those qualms.
***
To make progress, though, scientists will need to map out the causes of the valley—the plate tectonics, if you will, that tore open the uncanny rift in the first place. Likely, it’s a product of evolution: monkeys and other primates, when exposed to “monkeyoid” robots, exhibit the same fear and distress that humans do. But whether we evolved a sense of the uncanny for its own sake, or whether the sense arises as an unwanted byproduct of evolution, is an open question.
Some scientists argue the uncanny valley provides a selective advantage. Mori assumed it arose from a healthy fear of death. He wrote, “a feeling of strangeness ... may be important to our self-preservation.” A related idea is that humans avoid corpses and sick people (or at least sickly looking people) to limit exposure to the deadly microbes they harbor. Or, to turn this idea around, perhaps our brains are programmed to favor vigorous-looking people, who presumably make better mates and produce fitter offspring.
Other scientists seek a different explanation, based on the way the mind functions. They argue that, far from being a seamless computer, the human brain is a “kluge”—a jury-rigged contraption of independent parts (a language part, an emotion part, etc.) that evolution has patched together. Alone, the parts perform their tasks well enough. But they sometimes malfunction when working together. Wheatley, the psychology professor, surmises that the uncanny valley opens up when two parts, or “networks,” send conflicting impulses.
The first network anthropomorphizes anything it can, which explains why we weirdly find adorable faces even in lumps of coal, a carrot, a pipe, and a slushy round “head” of snow. The second network, a “living-thing network” lights up whenever we encounter live beings. Each helps us in its way, and in nature the two rarely step on each other’s toes. But, Wheatley explains, when we see an android or an unsettling face, “Parts of our brain are saying, ‘This is alive.’ But parts are saying, ‘Wait a minute!’” The torturous ambiguity causes us to shun the source because “our brains never evolved to process things like mannequins and dolls,” she says. “Or clowns or mimes for that matter.” (Which might explain some people’s seemingly irrational fear of Bozo.)
Other scientists look at the uncanny valley as something like an existential trial. Status and power are important to our self-worth, so perhaps machines that look too human stoke our fears of being replaced in our jobs, in society, or even on Earth. Moreover, MacDorman and coauthors note provocatively in one paper that “human-looking robots—especially if they could one day rival human intelligence—raise the question of whether we might not all just be soulless machines.”
Teasing out what contributes to our sense of the uncanny will occupy scientists over the next decade, if not longer. Wheatley’s lab plans to use functional MRI machines, which measure blood flow and brain activity, to scrutinize people wallowing in the valley. And steadily improving androids and graphics programs, especially in Japan, should provide thousands of “natural” experiments.
What’s certain is that after studying confrontations with androids and digital beings, we’ll know more about what it means to be a carbon-based human. As MacDorman has said in statements about his work, “We might be using androids, but what we’re really studying is ourselves—what motivates us and how we interact with one another.”
Sam Kean is the associate editor of Search.
Become a Subscriber