Traversing The Uncanny Valley
How Can We Design Robots We Trust?
Realistic humanoid robots no longer exist solely in the world of science fiction. Since her activation in 2015, Hanson’s Sophia robot has been granted Saudi Arabian citizenship, a United Nations title, and appeared as a guest on CBS 60 Minutes, Good Morning Britain, and The Tonight Show With Jimmy Fallon. Scripted or not, interacting with Sophia often appears seamless, with humans relaxing into natural conversation within minutes of their first exchange. Sophia’s ability to put her human counterparts at ease raises a lot of questions around the idea of the uncanny valley, and perhaps offers a lot of answers as to how to overcome it.
The uncanny valley is the hypothesis that as robots become more human-like in their appearance, humans exhibit more positive responses towards them. However, once robots reach the stage where they are almost identical to humans, these positive responses dramatically drop into the negatives, with users reporting discomfort, eeriness, and even disgust. Having a robot with no discernable differences to a human pushes these responses back up to the positive again. The theory was originally stated in 1970, by Japanese robotics professor Masahiro Mori. However, the uncanny valley remains just that; a theory. Little empirical testing has been conducted into the effect, or even existence, of the uncanny valley hypothesis, however with recent renewed interest in realistic androids this is rapidly changing.
Why Is The Uncanny Valley Relevant Now?
As we get closer to everyday integration of humanoid robots, such testing is required to ensure design mistakes do not render advancements critically flawed in their interactions with humans. The role of the uncanny valley is still widely debated, so understanding user interaction and response to different representations of humanity in robotics is incremental to both recognising how to interface these robotics experiences and how our world will shape itself around those interfaces. By taking a look at humanoid robots that do not elicit the uncanny valley effect in human users, and studies that have explored those that do, we can begin to understand how to traverse the uncanny valley and create robots we can naturally interact with.
How Have Others Overcome The Uncanny Valley?
Oh Jun-ho is a professor of Mechanical Engineering at KAIST and, alongside Hanson, developed Albert HUBO, a humanoid robot created in the likeness of Albert Einstein. They explain;
“I modelled after the old man’s face, so as to overcome the uncanny valley… because old men’s faces are imperfect, people feel less discomfort”
This presents two potential ways of avoiding the uncanny valley in humanoid robot design. Sophia and Albert HUBO are considered non-threatening because they are based on real faces (Sophia is modelled after actress Audrey Hepburn) humans are used to seeing, and comfortable seeing in a representational manner. Most of the population have only seen Audrey Hepburn and Albert Einstein in representational formats anyway, so a robotic representation doesn’t inspire so much unease; they feel natural and more familiar.
In addition to this familiarity through oft-represented interfaces, Albert HUBO used the face of an old man so that any imperfections added to, rather than detracted from, the robot’s facial expressions. The imperfections are familiar ones, rather than glaring indicators of construction. Psychology Today has suggested that;
“Minor flaws and imperfections in appearance can give us a feeling of seeing dead matter impersonating humans, much like watching a zombie. This can elicit fear or disgust while reminding us of our own mortality”
Another factor that may be contributing to our acceptance of Sophia especially is the fact that negative reactions to humanlike robots may be more down to aesthetics than their likeness to humans. In 2005, one of the earliest studies to question the uncanny valley hypothesis found expected uncanny valley effects when robots moved from unrealistic to realistic representations. However, this all changed when the images used were made more attractive. It seems we are more willing to trust a representation of a human that aligns itself with other representations we aspire to or respect, or from an evolutionary preference for the aesthetically pleasing.
What Can We Learn From These Methods of Overcoming The Uncanny Valley?
Minor Imperfections Speak Loudest
It’s the nuanced differences between an android’s face and that of a human that ring alarm bells for those experiencing the uncanny valley. Slight mismatches between emotional response and behaviour, or the expressions of the eyes and mouth, for example, immediately lead to distrust due to years of social conditioning to closely perceive microexpressions. However, as we have seen with Albert HUBO, implementing any imperfections into the robot design and ‘essence’ normalises them and makes them more relatable and recognisable as human. We, in turn, accept them as more ‘alive’.
Representation Is Easier Than Presentation
We view representations of living beings every day in the media. We know these people exist, and we know we are not looking directly at this person, but we may still be able to empathise with an awkward paparazzi photograph, for example. It seems that robots that are designed as representations of humans we have seen before slot neatly into this category of perception. Many argue that this is because we are familiar with the face already, yet are also trained in seeing such a face without being in the presence of the human person. We accept the apparent sentience of the android with more ease because we aren’t asking too much of our already trained perception skills. To sit across from a robot with a brand new face takes an extra step of perceptive processing.
We still have a way to go before we can fully call an android identical to a human being. Nevertheless, it’s becoming more and more obvious that that day will soon arrive and before we get there, it’s important to understand the potential roadblocks that will slow our progress. Human reactions to less-than-perfect human representations can have the power to halt such progress if the appropriate ways to avoid such negative reactions are not in place. It’s an exciting time for robotics and AI, but there’s plenty of research to suggest it doesn’t have to be a frightening time.
