The Quality Of Quantity In Video Game AI

It’s a joke I heard first from the guys on The Totally Rad Show, and it kind of always stuck with me. Since that year, I’ve always thought about the winners and nominees of the Best Directing Academy Award: is it really the best directing or is it just the most directing? It is usually called and presented as the Best Director award, but that presents its own set of issues. What qualities of the director are you judging? The ability to connect with the actors? How much time is spent covering non-directorial aspects of the film? There’s no clear-cut answer and is open to a great many interpretations.

Of course, the other way doesn’t help much, either, but it’s still an improvement because it’s really just the one question of best versus most. “Best” directing is easier to quantify than director since you can see what scenes or shots were most effective in any given film and see what ended up as the most cohesive product, but that once against circles around to whether that is because there is more directing or better directing. Over the past couple of years, it’s most likely that the former has taken the reins.

These two attributes tend to conflict in video games as well. Sure, some publishers tend to put out more games over better games, but I’m talking instead about artificial intelligence (and, to a certain extent, simulation) systems and whether quantity or quality is valued more. They are two notions people often conflate, just as they do with the Best Directing Oscar, but the question remains: is it better AI or is there just more of it?

This is, obviously, an oversimplification of a very large facet of gaming, but it’s a lot easier to wrap your minds around than delving into the particulars of it all. AI is such a deep and wide foundational slab intrinsically connected with video games (you generally can only find AI-free games in pure multiplayer affairs, but even then, you will probably find some sort of basic level of interlocking systems) that to talk about them superficially is just about the only way for both of us to get out of here with our sanity.

But let’s start from the beginning and define what Best AI really is. To me, it is the absolute appropriate systemic response to the current state of the world. The more accurate the game can replicate your expectations, the better. For the most part, the state of the world will be subject to your input (otherwise, as Harvey Smith puts it, it’s just simulation) and thus as long as your input is changing, the world should be changing accordingly. Pac-Man had simple but effective AI; if you got a power pellet, the ghosts would run because they knew the shoe was on the other hand at that point. That is good AI while, very simply, bad AI is when those reactions don’t exist, propagate, or seem appropriate. Crysis 2, for example, had enemies that wouldn’t even shoot at you if you were punching them right in their dumb, blank faces.

So what is Most AI? That is when there are plenty of reactions to your actions, but they are not necessarily appropriate or effective the given context. By and large, when people say AI, they mean enemy behavior, but this can also extend to other game systems like wildlife and information dissemination and the like. All of that can and should interact, but sometimes AI will do things that simply don’t make sense regardless of the current world state, or would only make sense if you or another system had provoked it to do so.

Take a look at Borderlands 2. For the most part, people lauded the improved enemy behavior given the Serious Sam-esque battle encounters of run backwards, circle strafe, and never stop shooting that the first one nailed and exhausted within the first hour (though I still ended up playing days’ worth of it). However, while I do believe there were improvements made to the AI systems of the game, it doesn’t seem like the actual enemy behavior has gotten any better. There is simply more of it. For instance, instead of running at you in a straight, unwavering line, Psychos will now every once in a while step off to the side in an effort to dodge your incoming fire. If that sounds familiar, that’s because that’s basically how the zombies in Resident Evil 4 and Resident Evil 5 worked, and they were mindless undead.

Psychos and the rest of the bad guys you face in Borderlands 2, as far as anyone can tell, still have some sort of self-preservation, but instead of counting for intelligence when you’re fighting them, you are just trying to adjust for random fluctuations in the pattern. This does not make for mentally engaging encounters because you are not engaging with systems that offer commensurate response to your input; you are fighting against a list that gets checked off every 10 paces or so.

Compare that to just about every stealth game that came out this year. All of them are driven by their systems and don’t just ask them for advice every once in a while. You feel as though if you hadn’t become involved, there would still be a very functional world operating. You just happened this one time to muck about in it.

Dishonored probably achieved this best this year. Three-dimensional sound propagation made it so that you were very conscious about running too close to oblivious enemies and watching out for rogue bottles / other clink-inclined items. Vision cones showed how far and wide suspicion is liable to turn into full-on awareness. And the ensuing reactions always seemed proper. Guards would yell for help or search nearby hiding spaces after discovering a dead or unconscious body, and you can throw bodies to distract swarms of infected rats to avoid getting your health all chewed up. It seems like whatever you do, your actions will yield a fitting response.

Far Cry 3 also succeeded because you can watch as things you have no part in unfold, and then change directions as soon as you intervene. Or maybe it was about you to begin with, prior to sinking its hooks in and the AI systems take over. Both would happen frequently to me. While scoping out an outpost, a tiger may linger around me after giving up on a deer chase. I’ll take cover in some foliage and pray to god it doesn’t see me when it suddenly runs into the enemy camp to kill a barking dog/everything else in there. I then step in and kill the tiger and the last sniper myself.

Or maybe I’ll be running from the tiger to begin with. I’ll be sprinting down a dirt road when I happen across two patrolling enemies. They open fire as I run literally in front of their faces. A rogue bullet of theirs strikes the oversized cat, and it soon stops chasing me to maul those poor dolts while I commandeer a nearby jeep to rundown that foul feline. The AI responses are expansive and expected in a very real-life-reflective sort of way. It achieves a breadth that you normally only experience while interacting with the real world.

Then again, that’s exactly the same predicament of More AI instead of Best AI. I would argue that Call of Duty games have the trappings of better AI where things (enemies especially) react as you expect to your actions. They will take cover when you fire, flank when you don’t, knife you if you get too close, and grenade you when you’re far. It is the breadth of reactions that you would expect from those in the thralls of battle, but it still somehow comes across as insufficient engagement. It feels like pure if-then statements and function calls.

Halo games have the same wide array of possibilities based on similar premises but with Elites ducking behind Grunts and Jackals scattering at the sight of a Needler. And sometimes reactions are…imperfect but are infinitely more believable than Call of Duty AI. There are many occasions when your foes and allies will take cover against lesser threats while a bigger one looms over them. There are cases when grenades are haphazardly thrown and blows up some friendlies. Halo games react believably but imperfectly, which kind of mirrors the natural, emergent course that reality often finds itself on.

So then maybe it’s not a question of Most versus Best. Maybe it’s about effective versus ineffective. When people say “better AI,” they often mean “more AI,” but that’s not to say that more of it is bad or to say that cold, calculated reactions are more affecting than imperfection. It’s about what is best for the situation—what is convincing and relatable—over what is alienating, which I guess is the great takeaway from every lesson in life.

, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,

Powered by WordPress. Designed by Woo Themes