The Uncanny Valley - Why More Realistic Characters Look Less Human - Extra Credits

Hey guys. Today we're going to talk about the Uncanny Valley. I'm sure some of you are probably already very familiar with this topic, but I'm always surprised to find how many people haven't heard of it before or who have heard of it but don't know what it is. So, enough beating around the bush, whatever that means, let's do this. Let's start off with a little history lesson. The concept of the Uncanny Valley was introduced in the 70s by a Japanese roboticist named Masahiro Mori. The story Mori liked to build robots a whole whole lot and over time he started building his robots to look increasingly human-like. And the more human qualities he gave to his bots, the more people liked them. Their vaguely human characteristics were charming. They looked like big bolted awkward children and everybody loved them. But as Mori continued to improve his humanoid robots, adding synthetic skin and rudimentary facial expressions and such, he discovered a strange trend. He was surprised to find that people didn't respond to these new robots so well. Oh sure, other members of his field were impressed with his advances, but just being around these robots made people

uneasy. This observation led Mori to come up with the theory of the Uncanny Valley. The premise is simple. If something that is clearly not human is given human qualities, we find those qualities endearing. But give it too many human characteristics and it starts looking like an imperfect simulation, which we find kind of disquieting or even revolting. But once you get past that and make the object look and act almost like a perfect human, we start to You know what? Hang on, I brought a chart. Where did I put that? Ah, here we go. This graph depicts the Uncanny Valley theory. It shows the relationship between how human something looks and how much we like it. Let's start at the beginning of the curve. At this point, the object isn't human looking at all and people respond to it with a big "meh". But if we add a few human characteristics to the object, suddenly it's a lot more appealing. Now it's got personality. It's cute or awesome, as the case may be. But then if we keep adding past a certain point, we hit this huge drop in appeal. Suddenly the object isn't cute anymore. Suddenly we're freaked out. But if we keep pushing past this point, eventually the object will become indistinguishable from human beings and we'll be okay with it again. That area, right down there, is the Uncanny Valley. So why does this happen? The idea is that if an object is clearly not human, then its human-like

characteristics will stand out and appeal to us. But if the object is almost, but not quite human, its non-human characteristics are all we're going to see. We know what a human looks like. I mean, we see them every day. So, when something is off, we know it, and it's unsettling. So, now that we're all on the same page, let's steer the sucker back on topic. What does this all mean for games? 3D artists and animators have been fighting the uncanny valley for some time now, and they made a lot of progress over the years, but the battle's not over yet. Graphically, games have advanced to the point where we're having to struggle with this problem, too. So, what are our options? Theoretically, we want our games and characters to be visually appealing, which means we're going to be aiming for one of these two peaks. That gives us two options: photorealism or stylization. Let's start with the first one. There's a lot of appeal to the idea of photorealism. I mean, it's amazing to look at. It lends itself well to simulating reality in a visually believable way, which many game concepts can definitely benefit from. But technologically, we aren't quite there yet, and it shows. One of the main problems with aiming for photorealism is just how easy it is to screw up. As I pointed out before, the main problem with a character that's almost, but not quite human, is that all the areas the character falls short in become

glaringly obvious. And it's not just an issue of model detail and texture resolution, either. Movement has a huge impact on photorealism. Think of all the times you've seen a beautiful screenshot from a game and thought, "Good god, this is going to be incredible!" And then you see the game in motion, and it all just falls apart. There are a host of other factors that can compromise the humanity of a character. Flat voice acting, sub-par animation, lame acting performance, smartass. Even environment interaction. I've seen a gamer play a hyperrealistic game, get stuck on geometry, and say, "What the hell? This is stupid. I never get stuck on that." And then play a Mario game and have the same thing happen without it bothering them at all. My point is, there are so many, many, many layers of detail required to create a completely believable human. If just a few of those details are missing or off, then the whole thing tumbles backward into the valley, and all that hard work is for naught. The fight to create believable photoreal characters is a huge money sink, and it's a very, very easy battle to lose. Let me give you an example. You remember that Final Fantasy movie, Spirits Within? Square tried to create a film with photorealistic CG characters, and it did look pretty cool. I mean, especially for the time. But, you remember how awkward and vaguely

inhuman the characters seemed? Now, think about The Incredibles. Even though they're clearly stylized, caricatures of humans, really, they look more lifelike and human than anybody in Square's ill-fated movie. And even though it's been over 10 years since Spirits Within, even with all of the technological advances we've made in that time, the same trend still holds true today. The characters in all of these movies continue to kind of just freak us the hell out. While the much more caricatured humans in these films somehow seem way more human and lifelike by comparison. On that note, let's have a look at our second option, stylization. Go ahead and make a quick mental list of your favorite game characters. Just the first five that pop in your mind. I'll wait. No, I won't wait. Let's just look at a few well-known characters. Let's say, Mario, Kratos, and some Final Fantasy people. These guys are some of the most recognizable characters in gaming. All extremely appealing, but not one of them photorealistic. Some of them may look more real than others, but you'd never mistake any one of them for a real-life human being. And since they're clearly not human, it makes their human characteristics stand out. That's what makes Mario so cute and funny, Kratos so

brutal and ruthless, and Final Fantasy characters so earnest and expressive. Those are their most human characteristics. We don't expect them to move perfectly or react to danger like we would, or when they get stuck on a 2-ft wall, we're willing to excuse it as an annoying idiosyncrasy of the medium, rather than being completely torn out of the experience. The beauty of sticking with this side of the valley is how open it is to aesthetic variety. You can go for any look you want to. You can go for a cell-shading look, try a painterly style, keep it simple, go retro, 50s illustrations. The number of style possibilities are endless. And unlike the colossal expense of pursuing photorealistic graphics, even small developers can create a visually memorable game this way. Of course, this option's not without its own little pitfalls and challenges. The task of creating characters that are human enough to resonate with an audience without pushing the caricature too far is a fine line to walk. And for games that thrive on a feel of simulation, a stylized look could undermine the authenticity of the experience. It's a delicate balance. So, as you can see, in the end, we have two options when creating game worlds and characters. One, we can strive for photorealistic models and try to increase fidelity by removing glitches, improving animation, perfecting voice acting, and you know,

pretty much ironing out every last kink. Or, two, we can try and perfect the craft of exemplifying human characteristics in non-human characters. I don't really think there's a wrong answer to this problem. I don't think pouring money into our graphics is a bad thing. That's the only way we're going to keep pushing the limits of the realism we can achieve. And maybe one day see what the other side of the valley holds. But, it's important to remember that graphical fidelity does not equal human fidelity. The graphics in me is as eager as anybody to see the first believable photorealistic character in motion. But, just don't forget how green the grass can be on this side of the valley. You may have noticed it's very brown-looking over there. See you next week.