Humans have more fun

Research article into the neuroaesthetics of playing games against robots suggests, unsurprisingly, that humans derive more enjoyment from playing with humanoid robots. The appearance of the robot was perceptual. The robots looked humanoid, but all cohort players (Lego Mindstorm arms, human facial robot, and actual human) adhered to a random selection of moves in what is hardly a game, but is the classic Prisoner's Dilemma with trivial AI. As the boring video linked at the end of the article shows the Lego Mindstorm pressing one of the two (and only two?) buttons: Cooperate or Defect, the game in question is close to half a century behind today's entertainment. The article's primary value is to lay an fMRI foundation under the claim that humans have more fun playing humanoid robots.

"Both regions correlating with the degree of human-likeness, the medial frontal cortex and the right temporo-parietal junction, have been associated with Theory-of-Mind. The results demonstrate that the tendency to build a model of another's mind linearly increases with its perceived human-likeness." (Can Machines Think?)

(I wonder what an fMRI of a viewer watching the distinctly nonhuman Wall-E would conclude?)

The secondary value of the article, is the speculated causal linkage to a Theory of Mind. The humans are believing (when playing humans) that their opponent has a set of beliefs, which are falsifiable. This is also a higher-order belief. What has interested Andreas Witzel and Jonathan Zvesper has been the behavioral distinction, by a special kind of human-like cognition, beliefs about beliefs. Can higher-order beliefs be more fun? Such epistemics often arise in Poker: I believe that you believe... Their article gives a (now hypothetical) example from Elder Scrolls IV: Oblivion, in which agents would behave plausibly if they would reason about the beliefs of other agents (rather than stupidly disregarding them has having no Theory of Mind). While it is speculative, I agree with their long-term projection.

I've been recently encouraged by the short-term opportunities for higher-order beliefs by use cases in Assassin's Creed. Suppose you Save a Citizen. Well in Assassin's Creed the only way I knew how to save a citizen was to kill the four or so ruffians that are grabbing or ogling the girl (and in the first impression of Damascus, it usually was a girl). I killed the ruffians. More often came. It didn't feel random. It seemed to me (and perhaps this is just an illusion) that the allied others had to be around, had to observe the dead bodies (they commented about the bodies) and seemed as if they were hardcoded to attack, knowing by your events that you killed them. They didn't suspect others. But what if they reasoned about your beliefs, as in whether you saw them. Might they not sneak up on you?

The problem to be solved, though, is not realistic epistemics, but entertainment. In some shooting games, the shots at you are less likely to hit you if you are not facing that way. It's not more realistic; it feels more fair, because you weren't looking that way. You had no chance to avoid the incoming fire. Also, the bullets are slower. Now, that does not invalidate a physics engine capable of simulating plausible collision or a behavioral simulation of plausible aiming. Rather it has been tuned for the player's entertainment (as have the explosions in a movie). Yet a designer can play with the physics and discover parameters that simulate fascinating interactions. I ask you, two, what are the parameters of epistemics that a designer would modify? And what are the nature of the algorithms upon which the epistemic engine is assembled?

The trick becomes, if you accept the hypothesis, that higher-order beliefs in computer games are desirable, is applying them to achieve aesthetically pleasing epistemics. User entertainment is the primary profit driver of commercial videogames. As in movies or music, emotions associated with the experience sell games. If you're not a professional, the most likely reason you are ever playing a game is to have fun right now. Accuracy of rendering, dynamics, physiology, or even of epistemics, is only secondary to entertainment.

Yet for rendering, versimmilitude if not the ultimate end does represent the end of subjective differentiation. Many non-photorealistic visual effects have been heightened by the same hardware and base software as the photorealistic effects. Once pixel pushing power exists, its feature set can adapt to user's needs. Modern high-end characters have bones, flexible flesh, carefully crafted or captured motions. Artificial intelligence fights the invisible war, to guide agents toward more entertaining behavior. It is common to desire soldiers to fight, to take cover plausibly, so that the user subconsciously associates the behavior with human behavior (as lectured at GDC by Killzone's programmers).

That degree of engagement almost always requires a minimum level of a consistent world. It can be cartoon (Legend of Zelda) or even abstract (Geometry Wars), but it shouldn't be inconsistent at its chosen level. If you are in Spider-man 2, you expect Spider-man to behave like Spidey. You expect thugs to be accosting people and to need your help. And you expect Mysterio (okay I didn't expect Mysterio), or some villain at any rate, to threaten Manhattan.

In Assassin's Creed, one could imagine a much richer tapestry of subterfuge, of hiding from enemy guards at least, and perhaps a touch of true intrigue, at most from implementing higher-order beliefs. Yet, before taking higher-order beliefs too seriously, remember that in Assassin's Creed, many of the first-order beliefs are implausible. Characters walk into each other, or sometimes into static poles (quite comical when you blending with a group of such absent-minded scholars!). When you fall onto them from three stories up, they get up and look at you, but that may be all. If you follow a mark and he sees you often, he doesn't seem to pick up that he is being followed.

/entertainment - permalink - comments