In the 1970s, the roboticist Masahiro Mori speculated about a threshold of realism up to which humans empathize with robots. However, when the robot’s appearance and movement improve beyond that, humans are repulsed instead. This problem is known as the Uncanny Valley. But as the realism increases further, reactions slowly match human empathy levels.
Now, the Uncanny Valley has become a problem for computer graphics also, although in some pre-rendered cases, animation is good enough to fool humans. The question is, does this really cause problems for video games? Aren’t these reactions to AI characters perfect for creating an emotional experience for the player?
For example:
In Bioshock, Irrational/2K managed to create a very spooky environment, particularly using the “little sisters” in the game. The strange feeling of familiarity together with the horrific setting brings out strong emotions in the player.
In Half-Life’s Episode 2, Alyx plays an important role throughout the game, accompanying and interacting with the player. Valve did a great job in building a positive emotional bond between the player and Alyx.
What is there to learn from these examples for designing AI characters? How do you exploit the uncanny valley when creating negative emotions, but yet avoid it when aiming for positive emotions? What do you expect to see as developers take the levels of realism in animation even further?
Next week: Your favorite AI moments in gaming.
3 Comments ↓
The uncanny valley effect will be here with us in game for while I’m afraid. Graphics and even physics are very close to real life levels of detail. There are even competitions to be able to identify which screenshots are from GT5 and which ones are from reality, the same with 3D models of people’s faces.
However if you get an animated guy, even in the latest Pixar movies or Polar Express you can absolutely tell they are not real and there is something unnatural to their movement. Not to mention animation in interactive games which is more difficult to nail and have less inverted time on them.
I think that before we even think about building realistic enough behaviors for the characters to dodge the uncanny valley, we have to obtain realistic enough representation of those behaviors, which includes mainly animation in all it’s forms: movement, gestures, reactions, etc.
Leaving aside the controversy that Masahiro Mori created in the robotics’ world with his theory (some of his colleagues did not approve with it saying there is proof it works but also proof it doesn’t), I believe that we can look at computer games characters from this perspective also.
I never really was aware of this effect until I read your article and I thought a little better about the reactions I had to some of the characters in the games I played. So, from my own experiences analysis, I can say that this theory can be applied to the experiences human players get in a game. I did not play Bioshock, but I watched the trailers, and indeed the atmosphere in the game seems to come straight out of the Uncanny Valley itself. I don’t know if the designers for Bioshock really thought about this when they designed the levels and characters for the game, but they really did a good job in creating that creepy effect :). On the other hand, Half Life 2 I played, also Episode 1 and I believe that the designers of the game did a great job with Alyx also. The same goes for Tomb Raider’s Lara Croft, as you go on an play the games in the series and follow the graphical evolution of Lara, there is a point in the middle games of the series where you notice some of the human characteristics that tend to stand out, but the feeling goes away with the last games of the series.
As to how this can be improved in 3d games, I agree with Diego, the starting point needs to be the animation itself, then it can be improved by modeling the AI behaviors to be realistic beyond the Uncanny Valley.
I think, knowing what I know about this theory, we might be able to take it up a notch. If I may be so bold; consider the lack of emotional response we feel when we become overly critical at the more advanced AI in video games. They can look so realistic that we start looking for imperfections especially in facial features, probably because our brains are wired as such to be able to pick up on facial recognition. We have no problem killing murdering or otherwise maiming a character in a game probablyu because we really aren’t feeling any dispotion about that character at all, that is to say we know subcontiously that that character is’nt real even though he/she LOOKS so real. Now here’s the punch line. Could it be possible to take this same theory and apply it those said individuals (whom I never would have agreed with before) whom blame violence against people soley based on video games. Could it be possible that the Columbine massacre if it was video game influenced that the two individuals had a condition that not only aplly the uncanny valley to AI but when they look at real humans they had the same lack of emotional response?! Is it possible that some individuals can’t tell the diffrence between AI and a human being? This concept would be the exact opposite of the unccany valley or more apt an add on to the phenomenon. If it is possible, then psychologist need to look into this because it could provide a new avenue for treatment before another tradgedy like that would occur. Just my two cents.
Leave a Comment