It's the year 2053, and you're visiting the doctor's office. Moments after you enter the examination room, a young woman in a white coat steps confidently through the door, smiling and greeting you by name. You're impressed by her efficiency and friendly demeanor. She describes several possible causes of your lingering shoulder pain, but when she places her hands on you to manipulate the joint, her skin is oddly cold. That's when you notice that her eyes have a glassy sheen and her gaze never quite meets yours. When she turns to ready a machine, you hear the faint whir of electric motors and you come to the eerie realization that you're being examined by an android. She seems so human, but her few details that are inhuman disturb you. Welcome to life in the uncanny valley.
The uncanny valley is a hypothesis proposed in 1970 by Masahiro Mori, a major figure in the field of robotics in Japan. Mori proposed that we feel greater affinity for artificial humans as they become more realistic, but when they are almost perfectly human, slight differences creep us out, and our affinity for them drops. Should they appear indistinguishable from true humans, we would again feel affinity for them. On a graph of affinity versus realism, the drop of affinity resembles a valley. Mori called this pattern bukimi no tani, translated loosely as "the uncanny valley."
In the decades since the uncanny valley was suggested, it has transformed from a philosophical theory to an effect with real consequences. Computer-generated human characters appear in movies, and engineers are constantly developing hyper-realistic androids, the official term among roboticists for humanlike robots. The day that a humanoid robot helps you at the doctor's office or even lives in your house draws closer every year, and how we interact with and react to simulated humans is quickly becoming more important. Let's take a closer look at Mori's ideas and the factors that cause the effect, and find out how creators can develop artificial beings that fall outside of the uncanny valley (if it even exists at all).
The Uncanny Valley's Origin
Masahiro Mori's original description of the uncanny valley is relatively straightforward: We like artificial beings more when they appear more humanlike, but we really dislike them when they appear almost perfectly human with a few subtle flaws. But looking deeper at this phenomenon reveals a lot of complexity in the concept, not least of which is simply figuring out exactly what Mori meant in his original publication on the theory.
Mori's essay "Bukimi No Tani" was first published in the Japanese journal Energy in 1970, but the uncanny valley didn't become a popular concept in the West until the article was translated to English in 2005. Even then, the essay wasn't rigorously translated until 2012 (the first translation wasn't intended for publication), which meant our initial understanding of the uncanny valley concept wasn't completely correct [source: Hsu]. First, bukimi might be translated more accurately as "eerie"instead of "uncanny," but the "uncanny" terminology is too well-known to disregard. More importantly, the term Mori used for how much we like artificial humans, shinwakan, is not easily translatable. The original translation, "familiarity," doesn't fully capture Mori's intent. Researchers have instead adopted the term "affinity," suggesting that shinwakan is a blend of both familiarity and likability [source: Kätsyri].
Mori placed several examples of objects with human appearances along his uncanny valley graph, including industrial robots, toy robots and prosthetic hands. Many are particular to Japanese culture, such as bunraku puppets and Okina masks. While these may be perfectly valid examples, they can be difficult for Western researchers to study and understand.
Finally, Mori suggested a possible reason for the uncanny valley effect: He placed corpses and zombies at the bottom of the valley on the graph, and suggested that we experience unease at not-quite-accurate human simulations because they remind us of corpses and we naturally fear death.
Mori's concept of the uncanny valley was based on his own ideas of how humans might interact with humanlike robots, not on statistical studies of human/robot interactions (which would have been difficult to conduct in 1970). But his hypothesis, which we'll break down in the next section, set off a wave of further research into the phenomenon.
Mapping the Uncanny Valley
Let's use some examples from real life and pop culture to map out the uncanny valley more clearly. At the low end of the chart lie industrial robots, which are not humanlike and don't inspire much affinity. An android like C-3PO from "Star Wars" would be in the middle: His build closely resembles a human's, and he speaks and acts like a human, but his metal exterior and robotic face clearly show he's not a human. Yet, we feel some affinity for him.
Further along the uncanny valley graph are computer-generated humans from Disney animated films such as "Frozen" and "Moana." While these characters obviously portray humans, animators intentionally exaggerate their features so they don't appear too realistic. Based on the success of these films, audiences feel a high level of affinity for them. And then there are simulations like the computer-generated version of Tom Hanks in the 2004 animated movie "The Polar Express." The film's creators attempted to make a perfectly lifelike character but fell short, resulting in many critics describing the film as creepy or nightmare-inducing instead of charming [source: Zacharek]. That eerie Tom Hanks? Right near the bottom of the uncanny valley. And according to Mori, the intensity of the uncanny valley effect heightens when simulations move rather than remain static.
So, when features that characterize humans — like voice, proportion and texture — are inconsistent in replicas, it throws us off. Mori's theory that slightly flawed human replicas are reminiscent of corpses and death may be partly valid, but doesn't encompass the complexity of the uncanny valley. It's likely the phenomenon is the result of several different reactions. Here are some reasons humans might be freaked out by almost-perfect human simulations [sources: Hsu; Looser and Wheatley]:
- Humans tend to identify potential threats in our surroundings. A shrub that is clearly a shrub is not a threat, so we feel at ease. A lion that is clearly a lion is a threat and we react appropriately. A shrub that looks like a lion creates a sense of unease, since we aren't sure how to react. This pattern could hold true for realistic robots that make us unsure whether they are humans or androids. (This is similar to pareidolia, our tendency to notice familiar patterns where there are none — like when we see a face in a cloud.)
- Human perception is attuned to human faces, a vital skill in recognizing friends and family members and noticing outsiders who might pose a threat. This close attention to faces suggests the uncanny valley effect would be stronger for artificial human faces versus hands or legs.
- We recognize the slight differences in a not-quite-human android as deformities, which we instinctively associate with disease, causing revulsion.
Researchers have been hard at work studying how and why the uncanny valley occurs. Let's take a look at some recent studies that have tested the uncanny valley effect and uncovered data about its potential causes.
Studies on the Uncanny Valley
A concern in studying the uncanny valley effect is that it's tough to quantify affinity and lifelikeness. But researchers have conducted experiments to detect and parse the uncanny valley effect, and even attempt to provide mathematical explanations for it. One key finding is that the uncanny valley does not appear in every study that looks for it, and when it does appear, it does not always do so with equal intensity. This suggests that the effect does exist, but is caused by specific factors and therefore doesn't show up in studies that don't include those factors. For instance, one study found that people are better able to discern real from artificial humans when they're looking only at eyes (as compared to looking at just a nose or a mouth), indicating that getting the eyes right is an important step in creating realistic human replicas [source: Looser & Wheatley].
Something as simple as an unnatural pose or expression on an android's face might invoke the uncanny valley effect, as demonstrated in research that showed people were most disturbed by humanlike virtual characters who did not display adequate facial responses when startled [source: Tinwell et al]. Replicas also dip farther into the uncanny valley when they try to "deceive" the viewer into thinking they're human rather than simply portraying very realistic androids. A 2012 study revealed that people are the most creeped out when humanlike robots seem like they have minds and the ability to feel and sense [source: Gray and Wegner]. And one study discovered that the uncanny valley effect only occurs when people are looking at faces that are familiar to their ethnic group [source: Hsu].
The phenomenon extends beyond people — another fascinating experiment measured the responses of monkeys to a range of real and unrealistic and realistic artificial monkey faces. The researchers found that the monkeys experienced a clear valley when viewing the realistic artificial faces [source: Steckenfinger and Ghazanfar]. Taken together, the research suggests that the uncanny valley does exist, but that it elicits many human responses. That means that overcoming the uncanny valley would be a difficult task, and an artificial human that transcends the uncanny valley for some viewers may not do so for others.
Mori's theory on conquering the valley? Don't even try. He suggested roboticists keep their androids on the left side of the valley, using exaggerated features to increase affinity and avoid the uncertainty and creepiness a more realistic android might exude (like the Disney strategy). Other researchers suggest that it is impossible to get through the valley because it is actually a wall — humans' ability to detect subtle differences in human replicas increases alongside technological gains that make artificial humans more realistic [source: Tinwell].
There are uncanny valley examples outside of academia, though. What are some familiar figures that live in the uncanny valley? Let's find out.
Visiting the Uncanny Valley
Objects that mimic humans have a much longer history than the concept of the uncanny valley. Consider incredibly realistic sculptures like Michelangelo's Pietà. We have admired the skill of artists and the realism and emotion depicted in works of art for centuries, but perhaps because they are static and often uncolored, they don't seem to attempt to replicate humans exactly and don't unnerve us. Contrast this to the works of artist Ron Mueck: sculptures of humans that are fully painted and appear astonishingly lifelike. Although he often creates fantastic figures that are enormous or have fantastic features, even his works depicting mundane scenes (two old women talking, a couple lying in bed) have a disturbing effect. Mueck seems to intentionally exploit the uncanny valley effect in his art. Other examples of this intentional horror effect can be found in some Japanese horror movies, such as the unnatural movements of humanlike figures in "Kairo" and "Ringu."
Yet, when Mori first proposed the uncanny valley, there weren't any real-world examples of hyper-realistic robots or computer-generated characters. Since then, advances in robotics and computer graphics have made the uncanny valley a phenomenon we've increasingly encountered. The 2001 movie "Final Fantasy: The Spirits Within" featured ultrarealistic computer-generated images and characters, attempting to make them as lifelike as possible. And the aforementioned "The Polar Express" and 2007's "Beowulf" both were criticized for the creepiness of their not-quite-human characters. (They also helped popularize the uncanny valley concept in the U.S.)
More recently, the Star Wars movie "Rogue One" used computer-generated characters in place of the actors Peter Cushing (playing Grand Moff Tarkin) and Carrie Fisher (Princess Leia). They were generally more well-received than earlier CGI characters, possibly due to their limited screen time, and possibly because advances in computer animation helped pull them partly out of the uncanny valley.
Then there are androids. Advances in software, materials and electronics mean they're no longer confined to science-fiction movies. The most realistic androids today are art projects like "Nova Phil," a very realistic replica of science-fiction author Philip K. Dick by Hanson Robotics, or as technology demonstrations, like the Actroid robots built by Japanese company Kokoro. These androids have been tested as automated guides at public events and as telepresence robots, allowing for semi-direct interaction with someone who is in another location. For instance, roboticist Hiroshi Ishiguro's android replica can represent him while he teaches his university classes from a distant place. The verisimilitude of these robots drags them closer to the uncanny valley.
Will we be treated by android doctors tomorrow? Probably not, but that future is likely decades away rather than centuries. The implications of the uncanny valley will become even more important as robots and androids play larger roles in our lives.
Author's Note: How the Uncanny Valley Works
It was fascinating to unpack the complexity of the uncanny valley concept, but what really struck me was the "what if" of overcoming the uncanny valley. Imagine an android indistinguishable from a real human. It's chilling to ponder. It's also why I'm such a big fan of the "Westworld" series. There are so many ethical and philosophical concepts tied into the notion of creating artificial beings.
One of the most interesting things that came up in the research is that, in his later writing, Mori suggested that robots with artificial intelligence had the potential to embrace Buddhism — to see themselves as a part of a greater whole, unified with their environment and the humans they interact with. He collected these essays into a book called "The Buddha in the Robot." These ideas were a natural extension of his belief that robot designers should stay on the left side of the uncanny valley, creating robots that are aesthetically pleasing but not identical to humans. In fact, his experiences working with students participating in robotics competitions suggested to him another way to overcome the uncanny valley: people who create robots invariably feel like they have imparted some portion of their "soul" to the robot, much like an artist feels with his or her artistic works. This notion that robots (even unintelligent ones) have their own nature that is connected to the humans and the world around them is a pretty profound way to think about the future of technology.
More Great Links
- Gray, Kurt and Daniel M. Wegner. "Feeling robots and human zombies: Mind perception and the uncanny valley." Vol. 125, issue 1. October 2012. (June 15, 2017) http://www.sciencedirect.com/science/article/pii/S0010027712001278
- Hanson, David et al. "Upending the Uncanny Valley." AAAI'05 Proceedings of the 20th National Conference on Artificial Intelligence. July 2005.
- Hanson Robotics. "Robotic Philip K. Dick Head Rebuilt." (May 27, 2017) http://www.hansonrobotics.com/wired-robotic-philip-k-dick-head-rebuilt/
- Herman, Barbara. "The 10 Scariest Movies and Why They Creep Us Out." Newsweek. Oct. 30, 2013. (May 28, 2017) http://www.newsweek.com/2013/11/01/10-scariest-movies-and-why-they-creep-us-out-243852.html
- Hsu, Jeremy. "Why 'Uncanny Valley' Human Look-Alikes Put Us on Edge." Scientific American. April 3, 2012. (May 25, 2017) https://www.scientificamerican.com/article/why-uncanny-valley-human-look-alikes-put-us-on-edge/
- Hsu, Jeremy. "Robotics' Uncanny Valley Gets New Translation." Live Science. June 12, 2012. (May 25, 2017) http://www.livescience.com/20909-robotics-uncanny-valley-translation.html
- Kätsyri, Jari et al. "A review of empirical evidence on different uncanny valley hypotheses: support for perceptual mismatch as one road to the valley of eeriness." Frontiers in Psychology. April 2015. (May 25, 2017) https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4392592/
- Lay, Stephanie. "Uncanny valley: why we find human-like robots and dolls so creepy." The Conversation. Nov. 10, 2015. (May 26, 2017) https://theconversation.com/uncanny-valley-why-we-find-human-like-robots-and-dolls-so-creepy-50268
- Looser, Christine and Thalia Wheatley. "The Tipping Point of Animacy: How, When, and Where We Perceive Life in a Face." Psychological Science. December 2010. (June 15, 2017) https://www.researchgate.net/publication/49628902_The_Tipping_Point_of_Animacy_How_When_and_Where_We_Perceive_Life_in_a_Face
- MacDorman, Karl. "Androids as an Experimental Apparatus: Why Is There an Uncanny Valley and Can We Exploit It?" Cognitive Science Society. January 2005. (June 15, 2017) http://www.androidscience.com/proceedings2005/MacDormanCogSci2005AS.pdf
- Mathur, Maya and David Reichling. "Navigating a social world with robot partners: A quantitative cartography of the Uncanny Valley." Cognition. January 2016. (May 25, 2017) http://www.sciencedirect.com/science/article/pii/S0010027715300640?np=y&npKey=33d0de0a91ebc177bc6f6bc2fa711066f1fae5c628e11fb2889ed8a6d70ae424
- Moore, Roger. "A Bayesian explanation of the 'Uncanny Valley' effect and related psychological phenomena." Scientific Reports. November 2012. (June 15, 2017) https://www.nature.com/articles/srep00864
- Mori, Masahiro. "The Uncanny Valley." IEEE Spectrum. June 12, 2012. (May 22, 2017) http://spectrum.ieee.org/automaton/robotics/humanoids/the-uncanny-valley
- Schaub, Ben. "Meet my android twin." New Scientist. Oct. 11, 2006. (May 28, 2017) https://www.newscientist.com/article/mg19225731.200-meet-my-android-twin/
- Singal, Jesse. "A New Study of Robot Faces Suggests the 'Uncanny Valley' Is Real." New York Magazine. Sept. 25, 2015. (May 27, 2017) http://nymag.com/scienceofus/2015/09/robot-faces-and-the-uncanny-valley.html
- Steckenfinger, Shawn and Asif Ghazanfar. "Monkey visual behavior falls into the uncanny valley." Proceedings of the National Academy of Sciences of the United States of America. Oct. 27, 2009. (June 15, 2017) http://www.pnas.org/content/106/43/18362.full.pdf
- Tinwell, Angela et al. "Perception of psychopathy and the Uncanny Valley in virtual characters." Vol. 29, issue 4. July 2013. (June 15, 2017) http://www.sciencedirect.com/science/article/pii/S0747563213000113
- Tinwell, Angela et al. "The Uncanny Wall." International Journal of Arts and Technology. Vol. 4, no. 3. 2011. (May 25, 2017) http://www.bolton.ac.uk/StaffBiographies/Angela-Tinwell/Tinwell-Grimshaw-Williams-2011-The-Uncanny-Wall.pdf
- Zacharek, Stephanie. "The Polar Express." Salon. Nov. 10, 2004. (May 28, 2017) http://www.salon.com/2004/11/10/polar_express/