FREE counter and Web statistics from sitetracker.com
collision detection
content | discontent
send me yours
December 05, 2005
The Xbox 360 and the Uncanny Valley: My latest Wired News gaming column












Wired News just published my latest video-game column -- and this one's about one of my fave topics: The "Uncanny Valley" effect. It came out of me spending a week playing the lineup of Xbox 360 launch titles, and realizing that even though the humans are more realistic than ever, they're also creepier than ever.

To read on, you can see the whole piece at Wired News, or via the copy archived below!

Monsters of Photorealism
by Clive Thompson

My hat is off to whoever designed the new King Kong game for the Xbox 360, because they've crafted a genuinely horrific monster. When it first lurched out of the mysterious tropical cave and fixed its cadaverous eyes on me, I could barely look at the monstrosity.

I'm speaking, of course, of Naomi Watts.

Not the actual Naomi Watts. She's heart-stoppingly lovely. No, I'm talking about the version of Naomi Watts that you encounter inside the game.

In some ways, her avatar is an admirably good replica, with the requisite long blond hair and juicy voice-acting from Watts herself. But the problem begins when you look at her face -- and the Corpse Bride stares back. The skin on virtual Naomi is oddly slack, as if it weren't quite connected to the musculature beneath; when she speaks, her lips move with a Frankensteinian stiffness. And those eyes! My god, they're like two portholes into a soulless howling electric universe. "Great," I complained to my wife. "I finally get to hang out with a gorgeous starlet -- and she's dead."

What's the culprit here? Ironically, the blame falls partly on the Xbox 360 itself, and its bleeding-edge graphics engine. Sure, the 360 can generate the most photorealistic human avatars of any game console in history. But that is precisely why they look so creepy.

This paradoxical effect has a name: the "Uncanny Valley." The concept comes from the Japanese roboticist Masahiro Mori, who argued that simulacra of humans seem lively and convincing so long as they're relatively low-resolution. Think of history's best comic strips: With only a few quick sketches on a page, Bill Watterson can create vivid emotions for the characters in Calvin and Hobbes. When an avatar is cartoonish, our brains fill in the gaps in the presentation to help them seem real.

But when human avatars approach photoreality? Something weird happens. Our brains rebel, and we begin focusing on the tiny details that aren't quite perfect. The realism of our avatars suddenly plunges downward into a valley -- and they begin to look like zombies.

The telltale flaws are almost always in the skin, because as many animators have told me, the physics here are damnably hard to master: When light hits real-life skin, it penetrates a tiny bit and bounces back out. The eyes are an even nastier hornet's nest of math; you have to replicate subtle nuances of moistness and "all these tiny movements of the eye," as animation expert Henrik Wann Jensen notes. Even today's best Hollywood animators -- who have far more processing power at their disposal -- have shattered upon these shoals.

That's why the Xbox 360 -- and the whole oncoming new generation of superpowered consoles -- kind of alarms me. With all this hot new rendering power, game publishers are more than ever indulging their Hollywood envy and trying to produce increasingly photorealistic people.

Yet all they're doing is tramping deeper and deeper into the muck and grime of the Valley. Don't believe me? Just play any of the Xbox 360's release titles. Many of them are quite terrific games, in terms of play. But aesthetically they are A Land Where The Dead Walk Among Us. The gang members and guards in Perfect Dark Zero look like an army of cadavers. In Quake 4, my fellow Marines looked like the victims of thoroughly botched face lifts.

Don't get me wrong: I don't hate the Xbox 360. It produces sheerly gorgeous graphics so long as it's rendering things that aren't human faces -- such as scenery. At one point in Perfect Dark, I peered through a zoom scope and noticed that tiny imperfections in a distant brick wall were casting microshadows from a hazy phosphorus lamp, while ships bobbed gently in the nearby bay. It was a perfect postcard of noir menace.

Indeed, this is why the most aesthetically un-jarring 360 games are those that explicitly avoid attempting realistic human faces -- and stick to a cartoony style straight out of Disney or Nintendo. My favorite 360 title so far is Kameo, where the star is a magical elf rendered in anime-lite, with big eyes, a teensy nose and soft-focused skin. Sure, the backdrop environments are masterpieces of gothic realism, but Kameo herself is drawn in broad strokes -- a blend of styles that, not incidentally, was innovated by Japanese manga. Surreality, as it turns out, is more seductive than incomplete reality.

At least for now, anyway. It's possible that one day we'll climb out of the Valley. The game industry is increasingly aware of this problem (and many writers, myself included, have been harping on the subject for years now). Maybe someday home consoles will approach Deep Blue supercomputing power, and designers' understanding of skin and eye physics will vastly improve.

And someone will finally do justice to poor Naomi Watts.


Postscript: That image above is actually taken not from the Xbox 360 game -- I couldn't find a Watts screenshot -- but from the PC game. (Technically, that ought to mean it looks even better than the one in the Xbox 360.) But really, it's hard to get a sense of the true creepiness of the Uncanny Valley effect unless you see the avatar in action; that's when the ghastly slackness of the face truly leaps out.

Posted by Clive Thompson at December 05, 2005 12:30 PM

Trackback Pings

TrackBack URL for this entry: http://www.collisiondetection.net/mt3/mt-tb.cgi/1383

Comments

You know, when you have mentioned the Valley before, you have mentioned skin and eyes. But for me, it's always the mouth that looks the creepiest. The interplay of shodow and delineated teeth are always creepy in digital, to me.

That Watts capture is a good example. That's a mouth only a Lovecraftian horror could love.

Posted by: MoXmas [TypeKey Profile Page] at December 5, 2005 2:20 PM

The latest Call of Duty commercial ("I told her I was a general... and she BELIEVED ME!") really freaks me out for the same reasons.

Interesting as always, Clive.

Posted by: kickslop [TypeKey Profile Page] at December 5, 2005 3:05 PM

"When an avatar is cartoonish, our brains fill in the gaps in the presentation to help them seem real."

Not sure what you mean by this--I don't think cartoon characters seem "real" or "convincing", they just seem ... non-creepy. The problem certainly isn't due solely to the closeness to "reality" either: there's never before been a problem with an uncanny valley in the history of art, from cartoons to photorealistic portraits to everything in between. (And the same psychological effect applies to the physical analogue: dolls, puppets, masks, etc. vs. robots.) For some reason the problem arises only when you attempt to use computers to make things look more and more realistic.

Posted by: Michael S. [TypeKey Profile Page] at December 5, 2005 3:17 PM

MoXmas, yes, the mouth -- migod. I really do wonder what Naomi Watts herself thinks of this ...

I haven't seen that commercial, kickslop. Is is available online anywhere?

Michael, what I meant is that they seem lively and perhaps emotionally convincing: I'm drawn in, I care about them, they feel like genuine characters with real emotions. Your point about creepy vs. non-creepy is interesting. Perhaps if I were to study virtual Naomi long enough, I'd feel empathy for her virtual feelings as I do for Calvins in Calvin and Hobbes ... perhaps the problem with creepiness is that I get so freaked out while interacting with her that I shut down?

Posted by: Clive [TypeKey Profile Page] at December 5, 2005 5:25 PM

Oh, and, that's a great point about animation vs. other media!

Though it's not true of robotics at all. Indeed, Mori developed his theory specifically by noting that little, unintelligent, highly stylized robots seemed more "alive" and funny and cute to viewers than highly realistic, human-looking automatons; the latter looked like zombies, or wax sculpture brought to life, and made kids cry. As I noted in my previous column on this, my Roomba seems highly "alive" and cute partly because it's so endearing low-rez.

However, you could argue that the Valley effect certainly does not hold for human portraiture. Stylized Picasso portraits can seem insanely creepy, while a photorealistic Victorian oil painting can be perfectly lovely and inviting.

I think most of this has to to do with movement. In any static medium, the algorithm of increasingly-realistic = increasingly-creepy does not necessarily hold. But in any medium where the depictions of people move -- animation, robotics -- we'd hardwired to notice zombified, slack, or just-plain-"wrong" human faces.

Posted by: Clive [TypeKey Profile Page] at December 5, 2005 5:32 PM

I have a background in CG and ever since I started reading about the Uncanny Valley I've just felt there's is only one way to get to the other side. So, I'm all for the ghoulish bleeding edge. I've no doubt we'll eventually see the facial photorealism of the last Matrix movie - and more - on a console or pc.

Posted by: Anthony [TypeKey Profile Page] at December 5, 2005 5:47 PM

I think the real key here is how well humans can tell the difference between someone who is alive and dead. It's not the rendering, a dead person is exactly the same physically as a live one, its the internal motion that makes us alive. The breathing, the twiches and subtle pulses, the glimmers in eyes. Perfectly rendering a person just gets exactly what it looks like, a moving corpse. To make them look alive you need to get the blood pumping and the nerves firing...

Posted by: Abe [TypeKey Profile Page] at December 5, 2005 5:58 PM

I think you're getting closer to the truth when you talk about movement rather than skin texture. Skin rendering is pretty much a solved problem in the high-end world of CG feature films, but that didn't help Final Fantasy or Polar Express!

I find that most CG humans fall into the Valley primarily because of the way their faces move (or rather, fail to move). It's a problem we animators face on a daily basis. "Keep it alive" is a commonly heard phrase in animation dailies-- it's shorthand for a whole collection of little movements of the eyeballs, eyelids, cheeks, brows, mouth and jaw. Leave any of those little movements out, and the character dies or "goes puppety". And as subtle as they are, they require a surprising amount of thought to do properly. That's because they're almost always dependent on the character's social, mental and emotional context, not just their physical environment.

As long as movies and games keep getting made without paying attention to those details, they will keep falling into the Valley.

Posted by: otherthings [TypeKey Profile Page] at December 5, 2005 6:02 PM

If you haven't yet read "Understanding Comics" by Scott McCloud put it at the top of your to do list. He explains very clearly how and why the brain processes cartoony VS realistic images in the beginning of the book. In a nutshell, the less precise the image is, the easier it is to imagine a character as ourselves or people we know. Specifics add barriers to empathy because they distinguish a character enough to make them read as strangers.

But aside from that, it's esily the best book I've ever read on making art (whether you're in comics or any other media). The guy is a total genius.

As for the present "uncanny valley" effect, couldn't we solve it by just making sure all the games for now involve undead adversaries? I mean, if you're fighting zombies in the first place it seems like an asset all of a sudden.

Posted by: johntunger [TypeKey Profile Page] at December 5, 2005 7:35 PM

Nice piece, Clive. I thought the same thing when I played Quake 4 -- and Doom 3, and Black and White 2 and just about any other game I've played in the last little while. The scenery is amazing, the people... less so. They're so wooden and puppet-like that they almost make Keanu Reeves seem human.

Mathew

Posted by: mathewi [TypeKey Profile Page] at December 5, 2005 11:18 PM

You need to be very careful with design in order to avoid the uncanny valley. I think that's the key. What you want, is that you want to avoid "normalcy".

So I didn't really get that feeling, for example, watching Advent Children the way you get it watching The Spirits Within. Because one is a fantasy setting, the other tries to be uber-realistic. Huge difference.

So what do you need? Big, bold designs I think are key. It distracts your mind away from "normality", which is the cliff that they tend to fall off of pretty quickly.

Actually. I look at the posters I have in my room (big beautiful wall hangings..), one of Squall and Riona, a very romatic looking..and kinda creepy looking picture...(I mostly like the other effects of the picture), and the Auron bio pic from FFX (The one with the two Zanarkands in the background). That doesn't look creepy at all. It looks dignafied..again, because of the big, bold designs.

There's a secondary issue, and that's the issue of smoothness. I'll give an example. Legolas jumping off the Cave Troll. It was WAY too smooth for the scene, and it was obviously fake. But this has to do with motion, and not imagry.

Posted by: Karmakin [TypeKey Profile Page] at December 6, 2005 1:47 AM

To echo JTU's sentiment, at least we can look forward to some amazing & realistic zombie games over the next few years...
Also, I'm with Anthony on this one - we can talk about the Valley all we want, but the most likely way to get through it is to blast ahead full steam and hope for the best. 10 years from now we may well covet games that fall in "The Uncanny Valley" genre precisely BECAUSE they're so weird.
Graphics are great, but what the gaming industry could really stand to benefit from is more innovative gameplay and design.

Posted by: garthbreaks [TypeKey Profile Page] at December 6, 2005 10:29 AM

What about Half-Life 2? Whereas at games like Doom 3 all of the animations are pre-manufactured in a animation editor, HL2 leaves much of it to the physics and their character animation and face poser.

Valve tackled the problem well already with the HL1 when they started using bone system to animate characters. If you want the animation to look natural within the giver reality of the game (that doesn't mean realistic in the first place) it must rely on the system of the game itself, not on the external editor that brings the character to life before it is incorporated into the game. This then causes imbalance of the graphical elements of the game which result as eerie characters or other models bland enviroments, etc...

Posted by: Martin [TypeKey Profile Page] at December 9, 2005 7:54 AM

I was going to mention Half-Life 2 as well. I thought one of the freakiest scenes in HL2, not because of the Uncanny Valley but because it was so extremely realistic, was the part early in the game in which you see the Metrocops bust in a door while you're exploring a hostel in City 17. It was like watching an episode of Cops unfold right in front of you.

Posted by: Young Freud [TypeKey Profile Page] at December 10, 2005 12:50 AM

It's interesting, I just found out that VALVe wants to incorporate another effect into their games:
http://www.halflife2.net/forums/showthread.php?t=97324&page=2
Just read it your self. It's just another proof of what I was saying.

Posted by: Martin [TypeKey Profile Page] at December 10, 2005 6:14 AM

Post a comment

Thanks for signing in, . Now you can comment. (sign out)

NOTE: If you posted a comment and you can't see it -- try refreshing your browser.


Remember me?