Skip to main content

Do Androids live in The Uncanny Valley?

An AI-equipped ‘astronaut companion’ called CIMON is on its way to the International Space Station. With fictional androids starring in TV shows Westworld and Humans, just how close are we to creating a socially-acceptable companion robot? It’s the word ‘acceptable’ that’s the problem and from this point forward we leave the comfortable world of logic circuits and servomotors, and enter the uncertain, illogical world of human perception.

CIMON_robot_ade4398e5ee39762b19ba563df5060297a4211d7.jpgCIMON: Astronaut companion on the ISS Image credit: Airbus/DLR

First off, what’s the difference between a Humanoid robot, a Cyborg and an Android? Here are some working definitions:

  • A Humanoid Robot is a mechanical approximation of the human form. In other words, it has recognisable limbs, a head and a body. One of the earliest to be constructed was Eric in 1928, recently rebuilt for the London Science Museum. A modern example would be Boston Dynamics’ Atlas.
  • A Cyborg is an organic being, usually but not necessarily human, with mechatronic enhancements. This conjures up images of the fictional Robocop, but a hearing-aid wearer is also, technically, a Cyborg.
  • An Android is a very advanced humanoid robot designed to be indistinguishable in every way from an organic human. Needless to say, there are no known real examples of this ‘ideal’ at the moment. The best we have, are merely visually ‘realistic’. Androids in fiction have been around for nearly a hundred years.

The ‘Uncanny Valley’

In the 1970’s a professor in robotics, Masahiro Mori, postulated that as humanoid robots improved in appearance, an observer’s reaction would become increasingly positive and empathetic. At some point however, nearing about 80% perfection, the reaction plunges into negativity or revulsion. Further improvement yields a return to positivity with total acceptance as the robot becomes indistinguishable from a live human. This plunge in the graph (Fig.1) became known as the Uncanny Valley.

Android_blog_11_c865f22f8518e9596b53baf7a0abe42f26b841be.png

Mori’s idea stimulated a great deal of debate at the time, but it’s only recently that actual robots, as opposed to CGI-based simulations in the cinema, are starting to approach the ‘plunge’. The difficulty lies not in identifying the features of a particular robot that cause the negative reaction, for example the eyes, but pinning down that subtle change that brought it about. Maybe there’s a mathematical function like the Golden Ratio for artistic or photographic composition that can make the difference between a simple snapshot and work of art. If there is, it’s yet to be discovered for androids.

What makes a robot ‘acceptably realistic’?

The adjective ‘creepy’ is frequently applied to humanoid robots that are perceived to have crossed the line of acceptable realism. Although highly subjective, most observers will agree on the level of creepiness or otherwise when assessing a specific robot. Here is a (rather pretentious) video of a current state-of-the-art android: How realistic is it? Is Erika creepy or just unconvincing?

My own view is that Erika is slipping into the Unhappy Valley: in spite of the superb craftsmanship which provides a high degree of static realism, there is just nothing like enough complex facial movement. A major problem for the android designer is reproducing all the tiny movements of the human face and synchronising them with the voice. Frequently characterised as ‘Body Language’, it’s something the robot must master to make that leap over the valley. Another contender for the title ‘Most Realistic Android’ is Sophia from Hanson Robotics. Here is a video of Sophia in action; essentially a promo for the company, featuring Sophia being presented to an audience by her creator. It’s heavily staged, but impressive for her lifelike head movements when she is apparently ‘listening’ to Hanson speaking.

Ignoring the fact that the back of her head is transparent, I feel that Sophia is way more convincing than Erika and (just) manages to avoid any ‘creepiness’. It’s interesting to compare the android with the real human and see just how far technology has progressed; and how much further it has to go. So, what is acceptable realism in terms of robots? Robots are by definition active and not static: it’s possible to create a perfect picture of a human and indeed a colourless sculpture can be admired as a thing of beauty. Movement is the key challenge when attempting to construct an artificial human. Body language, particularly of the human face, is understandable even to a small child. A robot unable to ‘speak’ this language will be just a robot, and a slightly sinister one at that, no matter how clever.

Do humans dream of perfect androids?

The portrayal of artificial humans in fictional stories may give us a clue to our expectations for a perfect android. Interestingly, films that see a future of mistrust and conflict between humans and robots often feature perfect androids: the new Westworld and Humans series, Blade Runner and even the Terminator. This suggests that we would prefer imperfection; just so there’s no confusion or misunderstandings. Not that we’re worried they’ll take over the world or anything. Perhaps the whole idea of a 100% perfect human replica will always be unacceptable. The model provided by Data, the android crewman on Star Trek the Next Generation TV series may suggest a limit short of 100%.

It could be that Mori’s graph should feature a further plunge downwards at about the 99% point. Personally, I find Data acceptable, with his bland facial expression and slightly jerky head movements. He is unmistakably an android. At one point in the Star Trek series, we are introduced to Data Mark 2, an improved model with an ‘emotion chip’ fitted. This emotional android has a full range of facial expressions and is much more human-like. Unfortunately, he falls straight into the valley because his now-creepy appearance matches his all-too-human evil intentions.

Social Robots

It seems that it’s too easy to create a nearly-realistic humanoid robot that can make a human observer feel uncomfortable in its presence. This presents a severe challenge to the designers of the social robots now seen as essential if elderly humans are to be cared for, and the very young to be educated in a bleak future world. The current thinking involves backing away from the perfect reality goal, mainly to limit the development costs, but also to avoid the risk of a bad customer reaction. The next two video clips show a fictional ‘servant/butler’ robot called Kryten from the UK comedy science fiction series Red Dwarf, and a real product from Engineered Arts called Socibot. First, the incomparable Kryten:

Despite having the head of an ugly cartoon robot, Kryten is very definitely well clear of the uncanny valley; a perfect social robot just needing major advances in AI to make him a reality. Back in the real world, we have Socibot which uses a back-projection system to make the facial expressions limitless.

A clever concept which avoids having a head full of whining servomotors and jerky motion. Not perfection, but realistic enough, and judging by the public reaction, not at all creepy. A step further back takes us into the familiar world of the classic humanoid robot, in this case to another commercial product, Pepper.

Anthropomorphism

Anthropomorphism is the attribution of human traits, emotions or intentions to non-human entities (Wikipedia). It sounds like the process of creating an android, but it’s not. It is the uniquely human trait of interpreting the actions of an animal, usually a pet cat or dog, in terms of human emotions. The human brain can even do this for an inanimate object such as a car. The Pepper robot has the basics of the human form, highly simplified and androgynous, but hitting that sweet-spot where anthropomorphism kicks in.

Pepper is ‘cute’, small and non-threatening. The only emotions it has are those created in the observer’s mind. Human imagination can be harnessed to save an awful lot of coding effort and servomotors.

Robots in Space

The human body has evolved to work in Earth’s gravity, hence the development of legs and feet to propel the body over the land surface. In the ‘weightlessness’ of Space, these appendages are largely useless unless they have something to push against. The concept of a legless robot has appeared several times in fiction: for example, Nomad in the original Star Trek series and V.I.N.CENT in the movie ‘The Black Hole’. Vincent looked nothing like a human even with its eyes, grippers and vestigial legs, but it was cute and everybody loved it. A real space robot called Robonaut has been operating on the International Space Station for years. It is humanoid in form (See picture below) but, notice that its ‘legs’ are really arms and its feet grippers. This may be an ideal format for Space, but if they tried to ‘humanise’ it further with a recognisable face, I think the astronauts might freak out.

Robonaut_with_legs_db6ebaf122e15a0949363821a6c6b6393cab885e.png

Returning to the picture of the grinning ball at the top of this article, this is the latest robot to be sent to the ISS as an astronaut assistant. Called CIMON, which stands for Crew Interactive Mobile Companion, it’s propelled by jets of air sucked in from the cabin and is programmed to recognise and work with one particular astronaut. I wonder if Alexander Gerst will succeed in forging a bond with his attentive, and possibly flatulent new friend, or will CIMON end up being used for football practice. Just don’t give it control of life support systems or the air-locks.

In the movie 2001 A Space Odyssey the AI-based HAL 9000 spacecraft computer reacts to receiving conflicting instructions with paranoia and sets about resolving the contradiction by killing the crew. HAL was just a disembodied voice to the crew; the only sign of its physical presence an all-seeing expressionless red eye. Had HAL possessed a more physical human form, perhaps Dave Bowman and Frank Poole might have sensed the onset of its ‘nervous breakdown’ before it was too late. Still, CIMON might have been worse: it could have been a realistic human head floating around in the dark. The stuff of nightmares, not just the uncanny valley.

If you're stuck for something to do, follow my posts on Twitter. I link to interesting articles on new electronics and related technologies, retweeting posts I spot about robots, space exploration and other issues.

Engineer, PhD, lecturer, freelance technical writer, blogger & tweeter interested in robots, AI, planetary explorers and all things electronic. STEM ambassador. Designed, built and programmed my first microcomputer in 1976. Still learning, still building, still coding today.
DesignSpark Electrical Logolinkedin