The Fault in The Uncanny Valley
Let’s start with a question;
What is your first thought when you see the picture below?
Is “creepy” or “uncanny” a good word to describe it?
Okay, but if you said it is creepy;
What do you mean by it?
Does it mean that you are afraid of it?
Do you want to run from it?
Is it disgusting to you?
Do you see it as a threat?
Many would say that they aren’t afraid of it and it isn’t disgusting, although it is eerie. It is creepy. It is uncanny. People around it are uncomfortable.
But can we accept that there is a feeling of unease that we don’t fully understand? — Francis T. McAndrew, a social psychologist and professor at Knox College, couldn’t.
What is Uncanny? What is the Nature of Creepiness?
“There wasn’t a single study on it.”, McAndrew said while explaining the reasoning of his research.
He studied 1341 participants, from age 18 to 77. They all rated the perceived creepiness of various behaviours and physical characteristics — for instance, “why does being approached by a stranger in the night lead to feelings of creepiness” or “how creepy someone with greasy hair is”.
Other than finding what creeps us out, he wanted to know why and to what degree. He asked the participants to rate how creepy certain occupations and hobbies are. The participants also had to agree or disagree with the statements that can uncover why they felt that way, like “I am uncomfortable because I cannot predict how he or she will behave.”
The fun fact is that the most creepy things to the participants were clowns. In the second place of creepiness were taxidermists and then the sex shop owners.
It’s not a secret that clowns are both eerie and joyful — but why were they the creepiest things in this research?
McAndrew explained it with the following;
“They’ve got all of the things going on that would put us on our guard. You can’t tell what their emotions are. They have this painted smile and exaggerated features, the hair, the nose, the shoes. And they’re designed to be mischievous. If you go to a circus and a clown pulls someone out of the audience, you know nothing good is going to happen. They’re unpredictable. They don’t play by the rules. And if they don’t understand those rules, what other rules might they break?”
The results of the research showed that being “creeped out” is an evolved emotional response. It manifests through the high state of anxiety to the ambiguous presence of a threat.
To simply put it, the key to understanding creepiness is uncertainty.
Let’s imagine we are sitting on a bench. It is three in the morning and the night is quiet. There is a bus station on the other side of the street. A tall, dark man is standing there, seemingly waiting for a bus to arrive. Out of nowhere, he turns to us and watches. The shadow is hiding his head from us, but he’s turned in our direction. We do not know how he’s feeling. We can’t see his eyes, his mouth, his expression.
We are still sitting on the bench, trying not to give him much attention, but we could still see him in the corner of our eyes. He’s not moving. If we didn’t see him move when we came, we would think it was a creepy statue — but it’s not.
It is a creepy human.
The bus comes.
We take a sigh of relief. We won’t be uncomfortable anymore. The bus leaves, but to our surprise, the man didn’t get on it. The uncanny feeling comes back again. Chills run down our spine.
He moves and the shadow no longer covers his face. We can see a macabre grin and a deranged glare.
It got scary quickly from then on.
When we are sure of the danger, we are scared. If it might seem like danger can be near with no valid evidence to support it, we are creeped out.
So this must be connected to the Uncanny Valley.
What Is The Uncanny Valley Theory?
Everything started with Masahiro Mori, one of the pioneers in robotics and automation. While doing his work, he noticed something strange in humans’ responses to non-human entities and decided to write a seminal paper on it;
“I have noticed that, in climbing toward the goal of making robots appear human, our affinity for them increases until we come to a valley, which I call the uncanny valley.”
In other words, the theory states that the more human-like robots become, the more familiar they appear — until a point is reached at which subtle imperfections of appearance make them look eerie.
He also stated in his paper that human-like movement in non-human entities is something that can cause an even deeper valley, as it's showed on the graph below:
Even though “Uncanny Valley” was something that was “discovered” in 1978, the phenomenon is a mystery to this day. The scientific research on it began more than 30 years after, in the 2010s.
“This is one of those cases where we’re at the very beginning of understanding it,” a cognitive scientist at the University of California, Ayse Saygin stated.
Scientists did and are still trying to define the things they do know:
“If you look humanlike but your motion is jerky or you can’t make proper eye contact, those are the things that make them uncanny. I think the key is that when you make appearances humanlike, you raise expectations for the brain. When those expectations are not met, then you have the problem in the brain,” Saygin said in one of her interviews.
A modern example of this is Shrek the Movie. In early test screenings of the film triggered intense feelings of anxiety in children when Princess Fiona was on the screen. Many kids cried whenever she appeared and were frightened of her.
The reason? She was too lifelike.
The outcome? Filmmakers edited her so she’d look more cartoonish.
Another example is the film adaptation of the musical Cats. It featured human-looking, cat-like creatures that most of the world found eerie. Some found it funny, some found it revolting, but most of them said that it was downright creepy.
The reason? Human-like movements and cat-like motions in one character.
The outcome? Global cancellation of the film.
And although there is clear proof of something going on with almost human-like things, many are still sceptical about uncanny valley even existing in the first place;
“We still don’t understand why it occurs or whether you can get used to it, and people don’t necessarily agree it exists,” Saygin added.
But why?
Does it Exist?
Masahiro Mori said himself that;
“Pointing out the existence of the uncanny valley was more of a piece of advice from me to people who design robots rather than a scientific statement.”
A 2013 research on the relationship between human-likeness and eeriness ended up providing somewhat of evidence that supports the existence of the uncanny valley. The researchers discovered a linear relationship between creepiness and human-likeness when manipulating realism and facial proportions. In other words; human likeness depends on the quality and quantity of the facial and realism manipulation.
There was also research in 2014 about children’s responses to human-like virtual characters. It was found that kids between the ages of 9 and 11 are prone to feelings of creepiness when responding to virtual characters that resemble humans. They described them, in opposed to humans, as less friendly and stranger. The uncanniness grew if the facial expressions were startling or if the character lacked the upper facial expression.
There are a few more researches that argue on whether or not their results support the uncanny valley theory. Most of them publish the results but keep conclusions open for debate.
Is it Wrong?
In 2008 a research was done at Indiana University that studied responses to images of people, androids and CGI characters with varying facial proportions. They wanted to know how our feeling of uncanny changes based on different human-like aesthetics. They noticed that the more humanlike and alive images were, the more sensitive to changes in the image participants were.
In other words, the most unsettling image change to view was of a person with minor flaws. Images of a cartoon, robot or animal followed behind.
Karl MacDorman who is an associate professor of human-computer interaction at Indiana University studied the uncanny valley for a long time;
“I believe his theory is instead expressed by his examples, which show that a mismatch in the human likeness of appearance and touch or appearance and motion can elicit a feeling of eeriness. In my own experiments. I have consistently reproduced this effect within and across sensory modalities. For example, a mismatch in the human realism of the features of a face heightens eeriness; a robot with a human voice or a human with a robotic voice is eerie.”
But let’s get to the uncanny valley graph again.
No, no, the other one.
That’s not the one either.
Yes, that one!
But, wait. If we can pick and choose what we will put on the graph, how can our data be unbiased?
Well, for one, we can take random robots on the internet and that’s exactly what researchers in a recent study of the journal Cognition did. Maya Mathur and David Reichling decided to use 80 pictures of social robots that have actually been built. Also, they purposefully took only those robots that were meant to interact with users, robots that weren’t marketed as toys and that were capable of physical movement.
They decided to rate how friendly each robot seems. The participants of the research were also given 100$ and were told to decide how much money they would give to each robot in hopes of a return on investment. The economic investment game shows how much subjects trust the robots and it shows our willingness to interact with one another.
The results showed that there is an uncanny valley effect with both measures. In other words, as robots faces became more human, their likability rating increased up until they looked almost like a human. Then, there was a point in which ratings dropped drastically — something we can consider a valley. The same was with the other rating; the amount of money participants would give to robots was increasing up until they significantly dropped, only to increase again when the robots looked identical to humans.
But the graph below only shows cherry-picked extremes of the research, hence the big valley in the likability and trust. Also, the researchers could only show still images to participants so the movement factor of the robots was out of the picture.
If we take all 80 robots that were part of the research and put the results of their likeness on the graph, it would look more like this;
The graph shows there is a valley, but it’s drastically less curvy.
The uncanny valley has become a mystery, not only for robotic engineers and animators but for psychologists as well, and although there’s evidence that supports the uncanny valley theory, some say that it’s not a valley, rather a wall or a cliff.
Do we have anything concrete on it?
The answer to this is no, we don’t have anything that will with 100% certainty prove that uncanny valley does or does not exist. At least not yet. What is certain is that the more we research, the closer we are to finding answers.
During the research on this topic, there were a lot of things that stuck with me throughout writing this article, but nothing as much as the words of Christopher Bartneck, an associate professor at the University of Canterbury in New Zealand.
“We’re trained to spot even the slightest divergence from ‘normal’ human movements or behaviour,” he says. “Hence, we often fail in creating highly realistic, humanlike characters. We find the likability to increase and then crash once robots become humanlike. But we have never observed them ever coming out of the valley. You fall off and that’s it.”
In other words, the goal is not to avoid the uncanny valley but to avoid bad character animations or behaviours, stressing the importance of matching the appearance of a robot with its ability.
But the question remains;
Are non-human entities in the uncanny valley creepy because they look nearly human or are these non-human entities that lay in the uncanny valley creepy because they just resemble a creepy human?
