< The Uncanny Valley


Friday, March 05, 2010

BROOKE GLADSTONE: Yes, it’s Oscar time. Avatar is the great blue whale the smaller films must beat. Among its qualities, the exquisite 3-D rendering of the Na’vi, the aliens we come to love. Because they are aliens, filmmaker James Cameron was able to leap over the dangerous chasm into which many computer-driven movies fall, never to recover. It’s a fear-inducing phenomenon called the “uncanny valley.” On the Media’s Jamie York explains.

JAMIE YORK: In 2001, DreamWorks was paying approximately 60 million dollars for an animated movie about a green ogre named Shrek. And the animation brain trust tasked with making this 60-million-dollar investment pay dividends was well on their way.

[MUSIC UP AND UNDER] Loveable, if grotesque, green ogre, check. Donkey sidekick that cracks wise, check. Fantastic world for everyone to inhabit, check. There was just one problem. In test screenings, the heroine, princess and motivating force behind the movie, was having a most unexpected effect.

LAWRENCE WESCHLER: They were so good at doing what they were doing with the princess character that when they showed it to audiences of children, the children started crying and freaking out because there was something wrong.

JAMIE YORK: That’s Lawrence Weschler, a long-time journalist who wrote about this incident for Wired Magazine. It’s true, the animators of Shrek were so good, so sophisticated that they were scaring their intended audience. Why? Their princess had fallen into what’s known as the "uncanny valley.”

LAWRENCE WESCHLER: Which was this notion by a Japanese roboticist named Masahiro Mori. The notion was that if you made a robot that was 50 percent lifelike, that was fantastic. If you made a robot that was 90 percent lifelike, that was fantastic. If you made it 95 percent lifelike, that was the best – oh, that was so great. If you made it 96 percent lifelike, it was a disaster. And the reason, essentially, is because a 95 percent lifelike robot is a robot that’s incredibly lifelike. A 96 percent lifelike robot is a human being with something wrong.

JAMIE YORK: Mori called it the uncanny valley, a play on Sigmund Freud’s idea of the uncanny: something familiar and yet foreign, at the same time. Science and technology writer Clive Thompson:

CLIVE THOMPSON: More or less around the early 2000s you started to get computer graphics inside games good enough that the people looked very people-like and they were hitting that valley point ‘cause you would look at them and you'd go, oh, wow, that looks really realistic. And you'd get up close and you'd see the face, and you'd realize that the skin wasn't quite moving [LAUGHS] the way it’s supposed to move and the eyes looked sort of fishlike. I mean, it was just aghhh! You know, I was - I would just shriek and shriek like a little girl, because this face looked so horrifyingly dead and dull.

JAMIE YORK: A computer-generated character climbs out of the valley when it finally looks and acts real. Create a character like Homer Simpson and you’re fine. A character like Tomb Raider’s Lara Croft is a little – unnatural? But so be it. The animated character based on Tom Hanks in The Polar Express, seriously creepy. The makers of Shrek, having learned their lesson, reimagined Princess Fiona as a slightly more cartoony-looking love interest, thus avoiding the valley. But Final Fantasy, a 135-million-dollar digital adventure which opened the same year as Shrek, fell into the valley and bombed. The studio that made it later folded. Similarly, Polar Express and Beowulf the movie left audiences unnerved by characters with dead eyes. On the other hand, Avatar -


ACTOR: It looks like you. This is your avatar.


JAMIE YORK: - uses awe-inspiring technology but avoids the uncanny valley because the aliens are distinctly alien, blue, stretched and unreal. The faces of the Na’vi don't trigger our uncanny reflexes because they're simply not human. Faces, despite ten years of refinement in computer rendering, they're still the Holy Grail. So many muscles attached to each other moving in tandem, so many complex and inimitable plays of light.

LAWRENCE WESCHLER: With faces, and especially with the whites of eyes, light behaves in a quantum fashion, exactly the way it does in milk, so that if you want to create a digital glass of milk and have a digital light go around it, unless you use quantum physics, you’re going to end up with something that looks like chalk.

JAMIE YORK: So much so, the animators practice first with bowls of digital milk. Weschler says that’s because just a glimmer of inauthentic light in the eyes is enough to sink a project. Pixar, a company born of digital animation, which featured a rather unintentionally frozen-looking baby in one of its first short films, Pixar learned it wasn't worth the risk. And it’s no coincidence that they've built their fortunes making protagonists out of those things that are easiest to light: shiny toys, cars, bugs, monsters, robots and old men with square heads who love balloons.


JORDAN NAGAI AS RUSSELL: Are you in need of any assistance today, sir?


JORDAN NAGAI AS RUSSELL Well, I’ve got to help you cross somethin’.




JAMIE YORK: What’s insurmountable in films is doubly so in video games, where the rendering has to happen on a second-by-second basis. Karl MacDorman is a professor of computer human interaction at the Indiana University Department of Informatics. Intent on bridging the valley, he’s been testing what precisely we feel when we look at near-humans on the screen.

KARL MacDORMAN: Out of perhaps, you know, 28 different emotions, there are about 5 key emotions that are highly predictive of uncanny reaction, and the most important one is fear.

JAMIE YORK: There are a number of theories as to why we fear when we gaze into the uncanny valley. One suggestion is that it’s existential because we see the potential for being replaced by nearly-perfect computers. I don't quite buy that. Theologians argue that we're seeing human imitators that lack a soul. For me, the most convincing explanation for the fear is evolutionary. We see in uncanny faces something not quite right, something unhealthy or unappealing, and our instinct is to recoil. Maybe they're contagious. Maybe they're not suitable mates. Asif Ghanzanfar is a professor in the Neuroscience Institute in the Psychology Department at Princeton University. He'd heard of the dread engendered by the uncanny valley, so he sought to test the reaction with his evolutionary indicator of choice, Macaque monkeys.


ASIF GHAZANFAR: Of course, we can't test extinct ancestors, so the only way you can test an evolutionary hypothesis in psychology or neuroscience or in any field is to do a comparative study, where you test a closely-related species.

JAMIE YORK: Ghazanfar created for his subjects three different kinds of what he calls synthetic monkey faces, one unrealistic, one realistic and one real. He flashed these faces on a screen for his five monkey subjects to watch, in random order for two seconds each, and he tracked where their eyes went.

ASIF GHAZANFAR: I mean, there’s good reasons to think that they increase their looking times to things that they prefer. All five monkeys – it was actually really remarkable – all five monkeys showed exactly the same pattern. When they'd look at the one that we predict would elicit the uncanny valley response, they looked at those images the least amount of time, which is also -- seems consistent with the interpretation that we're actually seeing a real uncanny valley effect.

JAMIE YORK: Ghazanfar’s research seems to suggest that the valley is real and the response is evolutionary, but we still don't know why. Ed Ulbrich knows the uncanny valley as only an explorer could. He was approached in the mid-'90s to create the technology for The Curious Case of Benjamin Button. He concluded that it was impossible, and he declined. But in 2002, he reconsidered, and with his fellow animators built the most sophisticated human likeness yet to appear on film. In the first 50 minutes of the film, we see a purely digital version of Brad Pitt’s face, aging backwards. Ulbrich says that portion of the film plants a flag squarely on the other side of the valley.

ED ULBRICH: To see the first motion of Benjamin now, a digital character that carried all the subtleties of Brad’s performance, it was a giant relief and it was kind of celebratory. It was like, okay, we - we've done it, we're here.

JAMIE YORK: But Lawrence Weschler and Clive Thompson argue that Ulbrich didn't scale the deepest part of the valley. Old, wrinkled skin reacts to light differently and is easier to digitally render. And the conceit of a human aging backwards is inherently unreal, a fantasy, so we're more likely to forgive the uncanny effect. They're waiting for a romantic comedy where the lead is, oh, by the way, a digital creation.

CLIVE THOMPSON: Do a CGI version of, you know, a well-known star, like Gwyneth Paltrow, in a way that I cannot tell it’s not her and I'm not freaked out by it. When you've got that, you know, I - I'll accept that you've crossed the valley but not before then.

JAMIE YORK: Clive Thompson says that when it comes to his beloved video games, he'd rather they give up than subject him to any more freaky, zombie-like characters. He believes that if game designers would only ease off the obsession with realism, they could find true art.

CLIVE THOMPSON: Think back to like the great age of deep realism in painting, and it sort of reached a point where like if you want to do a realistic painting of a road, there are only so many ways for that road to look because, frankly, there are only a, a few ways that roads are different in the real world. Whereas, when you get the Impressionists and the Cubists and that, that reaction, they're saying, there are a thousand ways we can paint a road, or a human face, right, because we're trying to get at something that is deeper than just what it looks like. And suddenly art, in my view, became much more interesting and much more expressive.

JAMIE YORK: For Lawrence Weschler, really bridging the gap between real and virtual comes down to something more ineffable than the algorithm for Newtonian light refraction or the blush on a cheek. For him, the eyes are important not simply because they're a problem to be solved, but because they have something more to tell us about what crossing the valley would really represent.

LAWRENCE WESCHLER: Sartre has this beautiful essay called The Face. He writes: “If I watch his eyes, I see that they are not fastened in his head, serene like agate marbles. They are being created at each moment by what they look at.” It’s a great notion, the notion that what we face when we face a face is the future, in some sense. It’s omnivorous with intent and with expectation and with desire and with fear, in a way that a thigh is not, which is why animating it becomes so much more complicated and so much more exciting.

JAMIE YORK: That future, of course, isn't so far away. Scientist Asif Ghazanfar of Princeton envisions the creation of a digital human face that can help treat autism. Karl MacDorman at Indiana University foresees a day when robots are companions for the elderly. And both are sure that time is coming, carried on a technological wave funded by Hollywood. But they are also sure that success will raise even more existential questions. MacDorman imagines the moment when an old woman first realizes that her beloved companion is synthetic and doesn't really understand her. That’s an uncanny shudder of a different kind, the kind we have when we stare in the face of what it means to be human. For On the Media, I'm Jamie York.