I’m currently reading a book on game design. The chapter on AI speaks only of the good that will come with advancing computer intelligence, yet not a word of caution or hesitation is included. I quickly outlined a number of worries I have over advanced AI and decided to bring them to our resident computer science major, Christian. Keep in mind I’m not “against” better AI, I just think it may lead to issues designers will have to deal with. I also find game design discussion to be infinitely enjoyable.
What follows are the initial worries I had and then Christian’s cool headed reponses.
Jay: In an old interview, Warren Spector spoke of making the AI for Deus Ex 2. He said that he actually had to tone down the enemy intelligence in at least one specific spot because it led to the unavoidable death of the player. His point was that intelligent enemies do not automatically equal fun enemies.
Games are generally designed with enemies that are in better positioning than you and who tend to out power you with better guns and more health. This can be modified but if they have worse weapons than you it will be less satisfying to defeat them as much pleasure is gained from being the underdog. If it’s dealt with by making grunts dumb but bosses smart, the grunts will have to be significantly stupider to the point where it will be like fighting monkeys before fighting the human bosses.
Christian: The issue of how to balance good AI is an important one. Perhaps it is the biggest obstacle it faces. I really like this point, and in order to provide an answer to your worries, one of the things I’ve thought about is what exactly a “smart enemy” does. Is he deadly accurate? Able to move and find cover? Can they sneak up on you for a stealth kill?
Jay: Certain genres will not be able to use advanced AI without huge changes. Think of a shoot ’em up. If the enormous enemy ship were intelligent you’d have no chance in hell. If the ships were all toned down and smarter much of the enjoyment of being a single small ship defeating a huge army full of enormous monstrosities would be gone. I can think of many other genres that may also have problems implementing high level AI.
Christian: You’re right; we play games to have fun, and usually, to be the hero. It makes sense that you aren’t wiped out in the first five minutes, because that wouldn’t be much fun. What I’m getting at is this; AI encompasses many things. It isn’t just about being able to kill well. Good AI also means characters with smart pathfinding. Characters with the ability to know their surroundings and asses threats. It can be about a character interacting differently depending on a conversation or something that you do. It can be about NPC’s really and truly learning things and changing as a game progresses. Balancing good combat AI is going to be tricky, but there are still so many other things that can be improved that would enhance a game immensely without drastically changing the level of difficulty.
I’m rather tired of playing things like 24: The Game and having my backup being run over by a subway train. Or seeing my comrades in Call of Duty 2 cover their heads while being two feet from a grenade instead of running away from it, or telling me six times that enemy infantry is to the south. Simple improvements like this could do wonders.
Jay: People are pattern seekers. A game like Tetris isn’t intelligent yet people love it because we enjoy dealing with small logic systems (I use that as a laymen, forgive me if I’m butchering the term). A shootemup is very similar to Tetris in that beating them requires hand eye coordination and memorization of patterns. This isn’t an argument against AI necessarily, but it’s pointing out that most or all classics work very well being simply pattern recognition games without any fancy AI.
Christian: Pattern seeking works with old games because a) with Tetris, it’s basically a puzzle and b) games like Mario and Zelda are a fictional abstraction of the real world. You don’t question why Mario is huge and why he runs with a certain momentum, because the Mushroom Kingdom is its own entity. Also, I believe pattern matching games we remember as being so fun were fun because they are quality games made by good developers. I can’t count how many bad Bubble Bobble rip-offs I’ve seen or played.
Anyway, with modern gaming being so realistic (or at least striving for it), we have a different story. When we see a well rendered human model move and attack like a robot in a realistic game world, it looks and feels strange. If the goal of the modern developer is to make a game that exists in a modern setting (or a plausible Sci fi/fantasy setting), then good AI should enhance the feeling of immersion, and more importantly the believability of the game world. Suspension of disbelief won’t work forever if the industry continues this trend where beautiful worlds are populated by the equivalent of Disney Animatronics.