Why do humans anthropomorphize Non-player characters in games? Studies have shown that even a circle with facial expressions and the right motion can engage a human. Part of this anthropomorphism may stem from how we, as humans, naturally interact with each other. When you observe another person’s actions, you naturally attribute a sentient mental model to them. You have an idea of “why” they’re doing what they’re doing. They’re not simply exhibiting a bunch of random physical actions, but instead they’re attempting to achieve some more abstract, intelligent goal. We tend to assume that others have similar mental models to our own, and for the most part this gets the job done.
With the non-player characters modeled in games, our brains inject the same assumptions about motivation into our perceptions of them. While these aren’t correct assumptions, they bring up interesting issues regarding what kinds of things result in “realistic” behavior of non-player characters within a simulated environment.
The utility of capitalizing on this tendency of ours to attribute intelligence to other entities has been demonstrated in various commercial game successes. Bungie’s Halo series received acclaim from both critics and players alike for the apparent intelligence of their enemies, but many players cite behaviors such as flailing about and generally freaking out when a grenade got stuck to them. This type of behavior, while outwardly identical to what we would expect an intelligent being to do in that situation, differs from the intelligent entity in that the reason for doing it is different.
An intelligent entity may freak out in such a situation because he recognizes his own imminent demise, and panics. The NPC in this example is simply following a simple rule; if the entity has a grenade stuck onto them, then play the freak-out animation. But the result for the player observing the situation is the same, and the simulated experience is more rich.
This presents us with a question that we must ask ourselves when developing autonomous entities for insertion into training or gaming environments. Should we focus on developing truly intelligent agents, or should we focus on “faking it” through manifestation of actions which appear to be intelligent, but are driven by more simple heuristics.
The more intelligent that the agents are, the more robust their behavior. They become harder to “break” by figuring out their rules and exploiting the gaps. An example of exploitable behavior could be an NPC which is supposed to engage the player in hand to hand combat, but is more limited in his ability to traverse the terrain than the player. The player can then exploit this behavior by situating himself in a position which is inaccessible to the NPC, and attack him from range while the NPC mindlessly tries to run to the player. A more intelligent agent may recognize the futility of his actions, and choose a more productive course of action. If nothing else, he could choose to take cover and hide until the player moves to a location which the NPC can move. This added robustness of a truly intelligent behavior would certainly improve the realism of the interaction.
The other side of the coin is that there are elements of the NPC’s manifested behavior which are more dependent upon his “body” than his “mind”. Animating correctly and performing appropriate gestures within a particular context all contribute to the perceived realism of the character. If we took the intelligent NPC described above, and gave him extremely intelligent behavior which always chose exceptionally effective tactics for dealing with the player no matter what the player did, but gave him no animation while walking around the game world, it would break the immersion of the game. Aside from the most basic animations associated with simple movement, a particular environment may require very specific gestures based upon the culture of the people being represented. For instance, people in an Afghani village exhibit different body language than Americans living in a rural town in the United States. In order to correctly model the behaviors of NPCs representing individuals in a particular context, we must employ animations and gestures which correctly mimic the real-world behaviors of the humans being simulated.
Likely, the reality is that the end goal should be a combination of both. In order to create a truly immersive experience for the player, the NPCs of the world must have realism in both their “mind” and their “body”. They must act intelligently, so that they are able to adapt to novel situations that may arise within the simulated environment, but they must also exhibit manifested behavior which looks real.
The difficulty that this presents for behavior modeling is that it requires multiple types of experts to collaborate on the development. A computer or cognitive scientist may have expertise about how to represent intelligent decision making, and are able to represent elements of how the thought process of the NPC should work. A cultural expert may be required to specify what gestures the people being represented would use, and in what context they would use them. A 3D artist may be required to handle creation the 3D model used by the character, and create the actual animations used by the simulation.
In order to bring all of this together, some mechanism must be used to facilitate communication and coordination between these groups. In future posts, I’ll present elements of Discovery Machine’s solution: A visual behavior modeling language which can be used and understood by scientists, engineers, artists, and cultural experts alike, and which facilitates quick capture of complex behaviors that can be directly inserted into simulation and game environments.
Currently, we are working on developing behaviors for a training simulation where we are driving a variety of non-player characters (NPCs) representing the population of an Afghani village. Some of these behaviors are quite complex, while others are more simple; some characters just sit around and converse with each other, seemingly relaxing. However, even these seemingly simple behaviors, when developed correctly, add a large degree of realism to the overall environment. Despite the fact that the NPC’s are not actually exhibiting “intelligent” behavior, we have the tendency to anthropomorphize the characters.