Tuesday, April 29, 2014

Our Relationship with the Uncanny Valley

By Rachel Greenberg**, AutomationGT 
Edited by Frank Tobe, The Robot Report 

Three humanoid robots (sitting) and their human counterparts. Photo: Gadgets & Gizmos Show
At present, the development and creation of humanoid robots accounts for only a small part of the robotics industry, yet humanoid robots capture the imagination in a particular way and as such, when many people talk about robots, they are really talking about humanoid robots. How and when and to what degree we choose to make our robots look more human demonstrates some interesting things about those of us who make, and those of us who buy, robots.

There is a well-known phenomenon in the design of human robots known as the “uncanny valley,” so named by robotics professor Masahiro Mori, and based on the Freudian concept of the “uncanny.” Generally, we tend to like things better as they become more like a living thing, as long as we still find it cute or goofy, really anything that makes it obvious that the robot isn’t too human. Professor Mori’s “uncanny valley” refers to the massive drop in user approval when robots get too close to looking human. However, on the other side of the valley, approval spikes again when we get to non-robotic humans of average appearance.

Prof. Mori's Uncanny Valley caricaturized by Carmen Rodriguez PeƱamaria
Some researchers think that we respond negatively to these robots because they remind of us dead or diseased humans, both of which we have biological reason to avoid. However, if movies and science fiction are any indication, it’s evident that we are interested in bringing robots as close to humanness as we possibly can, regardless of the revulsion that many people feel. In part, we may just be interested in the questions and implications that this kind of pursuit of knowledge and development of technology would raise, but as a result we will likely end up producing robots that are fairly, if not totally, similar to human beings in some pretty remarkable ways.

The three biggest factors that appear to lead to a robot’s placement in the uncanny valley are physical appearance, movement, and speech. We tend to be repelled by robots that have false skin coverings, markedly human facial features and expressions, stilted movements, or stilted speech that may remind us of zombies. No matter how advanced we get with our humanoid robots, we can’t quite make it out of the uncanny valley. For an example, check out this video of the Geminoid DK. The Geminoid DK is one of the better examples of a robot that does a pretty good job of looking human, but as you watch it move, it’s easy to notice a bit of creepiness, even if it’s just for a minute, just because of some little motion or expression that isn’t quite right.

Until we have achieved robots that are perfect imitations of humans in appearance, movement, and speech, it seems that we will likely not be comfortable with the less perfect humanoid robots along the way. For now, we seem to be more interested in making robots that seem lifelike in only minor ways. We regularly see robots that are inspired by humans or some aspect of humans (like the industrial robotic arm, a design inspired by our own bodies), but we feel more comfortable with these human elements when they are for function, rather than form.

However, we do seem very interested in making robots more human in their intellect. There are lots of robots on the market now that in some small ways imitate humans in their behaviors and social abilities, though not in appearance. This seems to suggest that we want robots to be comprehensible to us as individuals that can act and react in seemingly natural ways, but become uncomfortable when they are intelligent but look too human. This drive to develop social robots suggests that we want robots with which we can develop maybe sentimental relationships, if not necessarily substantive relationships, and that to achieve this, it is more important that robots act human than look or sound human.

Some researchers suppose that our increased exposure to robots and to highly digitized faces in video games and movies may ultimately make us more comfortable with false human appearances, and as a result, it may ultimately be easier for us to accept robots that are near or in the uncanny valley. This may take some of the immediate pressure off of humanoid robot developers to develop perfectly human-looking robots, which will allow some leeway for experimentation and error as we work on the appearances of our social and interactive robots.

It’s apparent that we have a lot to figure out about ourselves and about humanoid robots before we accept them completely, but it may be an impossible task to ever push humanoid robots completely up out of the uncanny valley, or to jump over the uncanny valley altogether. It’s a daunting task to program a robot with enough capability just to be able to respond to people in a consistent way. It’s unfathomable how much care, detail -- and time -- it would take to create a robot that could be substantially closer to a human in all ways including appearance, movement, speech, and behavior.

The general progression of human technology tends to follow the pattern that if we can find a way to do it, we will do it, except in cases of obvious ethical or human consequence. And we live in an age in which technology progresses more quickly than we can readily adjust to it, and there’s no reason robots won’t follow this same pattern. Engineers will continue to work towards increasingly human robots and androids, and it will be fascinating to see how these robots will and won’t fit into the roles we designate for them.


Masahiro Mori   Photo: IEEE Spectrum

Additional resources on the topic of Prof. Masahiro Mori's theory of the "Uncanny Valley":

  1. Robots Podcast interview with Hiroshi Ishigura
  2. New Russian entrant into creepiest robot face contest
  3. Professor Mori at IROS 2013 describing how he came to develop the uncanny valley theory
  4. Masahiro Mori on the Uncanny Valley and Beyond


** Rachel Greenberg writes technical and marketing content for Automation GT in Carlsbad, CA. She is a graduate of Johns Hopkins University, currently working out of the San Diego area.