Note: Where available, the PDF/Word icon below is provided to view the complete and fully formatted document
Disclaimer: The Parliamentary Library does not warrant or accept liability for the accuracy or usefulness of the transcripts. These are copied directly from the broadcaster's website.
The uncanny valley in robot design -

View in ParlViewView other Segments


Robyn Williams: And now finally for another take on how those robots should look. Should they be like us or look more like machines? If they're too much like us will they freak us and will we lose track of who's a person and who's a device? This is what Ayse Saygin is studying at the University of California, San Diego. She's a cognitive psychologist, originally from Turkey.

Ayse Saygin: I became interested in the design of artificial avatars or robots. I was initially not planning to work on robots for the sake of figuring out how we should design them, I was mostly interested in the brain, the human brain, as we perceive each other, you know, when we are face to face talking right now, of course your listeners won't be able to see us, but I can see you and I can see your gestures, I can see your facial expression and these are very, very important things for humans, social primates. And we have brain systems that allow us to be quite good at these things, you know, body language and non-verbal communication.

So what happens to these systems when you have something that isn't a human but is clearly doing human-like things, like actions or talking? So there you have something that evokes a sense of a person but isn't really exactly looking like a person.

Initially I was interested in, well, how does the brain respond, what can we learn about the brain from this kind of research? But what happened was that I fell into this concept called the uncanny valley. So the uncanny valley was a concept brought up a very long time ago in the 1970s before we had much robots in our lives at all…is the idea that the design of the robot, if you make it more and more and more human-like, instead of people just being more positively disposed towards it…initially yes, you know, more human likeness is good, but if you really become close to the human appearance there comes a point where instead of people liking it and having a positive disposition towards this robot, they have a negative reaction to it.

Robyn Williams: They get freaked.

Ayse Saygin: Exactly. And the words that they use were, you know, 'zombie-like' or 'creepy' or 'eerie'. And this has been an idea that's been around since 1970. Actually it goes back even further, not about robots but Freud in an earlier writing…the uncanny concept has been brought up earlier in the very early 20th century and they're talking about dolls and marionettes and how clowns can look a little creepy. So it's not necessarily a concept that's specific to robots.

So I became very interested in this first as a practical problem because we were trying to work with these robots. But also, as you say, I'm interested in the design of things or the look of things. I started thinking, you know, people are working on these robots that are very expensive, they have very important application domains. They're telling us all the time that robots will be used in healthcare and education, but if we design them in such a way that people get creeped out by them, then they're not going to fulfil their intended purposes.

So there might be a few weird people like me who are fascinated by creepy robots, but most people would not want to have a creepy robot. And I personally am interested in them but I still wouldn't want one in my home.

So now our research has come to a point where we're like, how can we use what we've learned about human perception and human cognition to design robots or avatars, artificial humanoid characters that aren't creepy but are more or less, you know, one day I can have a conversation like this one where it feels fluid and smooth and so on.

Robyn Williams: One of our broadcasts a couple of months ago was talking about robots and how even ones which looked nothing like a person—it didn't have a head or arms or legs—nonetheless people were giving these robots names, giving them personality. And rather than having one if it was broken just replaced by the same model, they wanted theirs repaired, as if there was something going on personally between the person and the machine which we didn't know about. Does that surprise you?

Ayse Saygin: Oh yeah, I mean, this is one of the funny things about this topic, is at one level humans are very easy to get anthropomorphic or emotionally attached. It doesn't take that much actually. You think about children with their favourite toys, they can become really attached to them. Or even just two dots and a line, it just looks like a face and…

Robyn Williams: Yes, I remember, Castaway, Tom Hanks fell in love with Wilson, you know, the football.

Ayse Saygin: Yeah, perfect example. So at some level it is very easy for us to get that feeling. I always like to use this example of Furby…do you remember the Furbies? I have the original Furbies. They now have new models. See, I don't want to see the new Furby because I'm attached to the old design. And I know, I've opened the Furby up and I know that it's not really smart, there's not much computation going on in there, it has very basic capabilities, and I'm a scientist who thinks about this stuff all day. And I turn my Furby on and play with it, and at some point if you leave it alone it's going to go back to sleep. But before it goes back to sleep it starts going…yawns a little and says, 'Boring…' and wants you to go and engage with it. And every time I feel guilty. And I know that it doesn't think anything, I know it has no mind, I know it has no feelings that I'm hurting, but I'm like, 'Oh, I'm a bad friend to my Furby, I'm neglecting my Furby.' So it's so easy to press our buttons.

So why is it so hard, then, to design a very successful robot? And the problem arises when you start making your robot more human-like. It actually becomes harder rather than easier. And this is all kind of hypothesis at this stage that we're still working out with experiments, but what we're thinking about it right now is when you see a person-looking thing, your brain, through decades of experience or possibly through the influence of thousands of years of evolution, expects a certain behaviour that comes with a person. And this isn't a conscious thing that I'm talking about. Even when you know that this is a robot after a while, it's just the brain circuits that are saying, 'Oh, there's something that has a very human-like appearance.' So when that thing behaves in a way that's not what you expect, that dissonance is what we think has to do with this uncanny valley and this uncomfortable feeling that we get.

Robyn Williams: I see. So what you might do, if you want to have a living companion (should I say even a partner), it might be a good idea to build in something that is recognisably machine-like rather than human-like to go along with all the rest of it. Is this where you're leading?

Ayse Saygin: Yeah, well, this is one of the ideas. And I'm definitely not excluding the possibility that we will make human-like agents that work well. It's possible. But one approach is to make it look robot-like, which consciously and unconsciously creates a sense of 'it's not a human'. For instance, one of the things that our lab focuses on is body movement. And humans and animals have a characteristic fluid body motion. And robots are being designed to mimic that now, but many robots just have your typical clunky kind of…

So if you have a very human-like appearance but then you have that clunky motion it stands out. But if you look like a robot and you move like a robot, it's sort of okay. The same exact movement, and this is what we have found in our studies, the same exact movement is perceived to be more acceptable in a robot-looking version of the agent than a very human-looking version.

So one trick if you want to make a product right now is to make it less human-like. And this happens a lot in cartoons and animation. If you consider animated films, the really photorealistic human-like ones, we really are very discerning about realism, but if you stay away from there, look at Pixar films where they animate characters that don't look very human-like at all but they can really evoke positive emotion, we can empathise with them. Have you seen WALL-E, that movie? It's so simple yet so expressive and we feel for the emotion that this robot feels. So the design can be very simple yet still press our buttons.

Robyn Williams: Yeah, like Toy Story is a good example of the way it began.

Ayse Saygin: Yes. Early Toy Story, definitely.

Robyn Williams: Or maybe even The Lego Movie. Ayse Saygin is Professor of Psychology at the University of California, San Diego. And she has a fabulous sense of design, as her office reveals very quickly.

Next week The Science Show goes on the road to see what young drivers are risking and what we could do about it.

Production by David Fisher and Charlie McCune. And if you haven't seen The Lego Movie go for it. It's awesome. I'm Robyn Williams.

Ayse SayginAssociate Professor
Cognitive Science Department
Neurosciences Program
University of California
San Diego CA USA

Further Information
Ayse Saygin at University of California San Diego

PresenterRobyn Williams ProducerDavid Fisher