Technology has made making extremely human-like robots a reality. But is this a good thing? Tara Bautista delves into the unease that many people experience with things that are almost – but not quite – human. Welcome to the ‘Uncanny Valley’.
When realistic isn’t pleasantly real enough
Last weekend I started watching a new fantasy show on Netflix called The Dark Crystal: Age of Resistance. You could say it’s a cross between Game of Thrones and Jim Henson’s The Muppets.
It’s a great series but it took me three episodes to get over the weirdness of the puppets. The pairing of human voice acting with the puppets’ mechanical facial expressions made for an unsettling experience. (And spoiler alert: I never want to see puppets kiss again).
Turns out I’m not alone in feeling this way.
In 1970, a Japanese robotics professor called Masahiro Mori first predicted the relationship between how human-like something is and our feelings towards it. He argued that at first, we experience increased positive feelings the more human-like the object is. But as something approaches being almost but not perfectly human-like, he predicted a dip in positive feelings and we would experience feelings of discomfort and revulsion.
Professor Mori called the dip the “Bukimi no Tani” – literally translated as “Valley of Eeriness” but more popularly known as the “Uncanny Valley”. Into the valley would fall things like androids (human-like robots), Japanese bunraku puppets, prosthetic hands, corpses and zombies.
The science behind our heebie-jeebies
Up until 2004, the Uncanny Valley hypothesis remained controversial and little studied. Nowadays it’s gaining acceptance as more scientific evidence backs up its existence.
In 2015, a landmark study reproduced Professor Mori’s predicted rise-dip-rise when participants rated the likeability of 80 real-life robot faces (ranging from clearly robot to android) as well as digitally-composed robot faces. The researchers showed the Uncanny Valley effect even influenced participants’ decisions about robots’ trustworthiness.
(Fun fact: Pixar won its first Oscar for the its CGI short film, Tin Toy in 1988. However, audiences were so creeped out by the human baby character that it drove the film studio to initially focus on stories with non-human characters like Toy Story and Cars).
But why do have negative feelings surrounding almost perfect human representations?
Firstly, the existence of a similar phenomenon in monkeys suggests there’s an evolutionary basis for it. Macaques are frightened and avoid eye contact with realistic but not perfect digital images of other monkeys. Like humans, they are more at ease with real images of monkeys and representations that are real but not too real.
An early but largely discredited hypothesis about how the uncanny valley works was that we generally are uncomfortable about things we can’t easily categorise. In the case of androids: are they human, or non-human?
However, the most likely explanation is the perceptual mismatch hypothesis. Because they look so human, we can’t help but want to judge them by human standards. But our brains pick up from their movements and responses that they’re not human.
In a 2011 study, the brain activity of 20 participants were monitored as they watched videos of a human, android and robot performing the same everyday actions such as drinking water from a cup or waving. The largest difference in brain responses between the three types of videos occurred when watching the android.
The researchers interpreted their findings to mean that our brains have no trouble processing when a human moves like a human, or a robot moves like a robot. But the brain lights up with an error message when there’s a mismatch, as in with the human-like appearance but robotic motion of androids.
The expectation that androids would move like us may be why the internal brain regions that help us move smoothly are activated when watching the slightly jerky movements of an android. One such region is the subthalamic nucleus (STN), which is impaired in voluntary movement disorders such as Parkinson’s disease. In fact, it’s suggested that our brains mistakenly interpret androids as people with Parkinson’s.
Our discomfort with perceptual mismatches also applies to emotions in artificial intelligences. We’re spooked when something we know to be controlled by a computer appears to show spontaneous empathy like a real person.
In sum, scientists think that because androids fail to meet our expectations of a normal healthy human, we have an averse response.
Towards getting un(real)?
So, how should we account for the uncanny valley in an era where androids are being developed to become our caregivers, companions and even priests?
One solution is to simply to design robots that are not too human-like. Think adorable Disney robots like Wall-E, or Baymax in Big Hero 6. (Relatedly, Pixar finally got the formula right for non-threatening, animated human characters with The Incredibles).
Or perhaps it’s just a matter of exposure and time (three episodes?) before the valley becomes a plain.
Great, now I have a name for that feeling apart from herbie jeebies. Feeling a need to watch your new series and see how I go with it.