From the lobby screen, you can determine just how difficult you wish to make your AI foe
Co-Optimus: There’s been a lot of talk of the AI portion of AI War and how you took a different approach than some of the more traditional RTS games, which you go into in great detail on your blog. Were there challenges in creating an AI that would have to respond to up to 8 players all targeting it instead of just one or two?
Christopher: I think that playing in co-op just brings a general problem with RTS AI into sharper relief -- in other words, I think the problems are already there, but that co-op play tends to expose them more heavily. The main problem is that of having the AIs be able to handle enough fronts at once to respond to 8 players, since those 8 players might even be managing 2-3 fronts each if they are all very expert. So that could conceivably be 24 different areas the AI is having to pay attention to at once if it is faced by a very formidable opponent, and there's not any other AI around that I have seen that really handles that well. I tend to play 4-player co-op in every RTS I play, and in most of them the AI is only doing one or two things at once, which makes it a bit too easy for us to defend. Even if the AI is attacking with multiple groups of guys at once, they all tend to bunch together in a way that works sometimes, but which becomes increasingly uninteresting over the long haul.
The core problem, there, is how to divide units up, and how to make them both pursue their own agendas and respond to player actions. The more players you have, the more of a problem that is, although a single player who is clever and uses multi-frontal attacks can run into the same issues. With AI War, I therefore went with a really hybrid AI, with a lot of unit independence and emergent behavior at the tactical level. So that way, however many fronts the players can muster, the AIs can easily respond to all of them at once. By having the individual units largely act independently, but with some flocking-style behavior that leads to emergence, you wind up with a situation where they can combine and divide their forces as needed. This can be pretty unsettling, and really makes players feel like they are facing off against an AI or alien opponent, something from sci-fi novels but not just a human opponent with similar abilities to their own. Paired with heuristic tendencies to make the AI do multi-frontal attacks of its own (with more fronts as more human players are added), this makes for an experience that gets a lot more interesting the more players you have. I'm actually kind of surprised no one has ever asked me this question before!
Battles like these are typical, as the game can support 30,000 ships onscreen at once
Co-Optimus: Many RTS titles have presented players with the opportunity to do some form of co-op, what some might call "comp stomping." What did you set out to do with AI War to make it different?
Christopher: I feel like most RTS titles are either focused on solo play, or competitive multiplayer, and co-op comes as an afterthought. And it generally shows. Usually the "comp stomp" involves playing teams on the skirmish mode, with AI players against the human players. This works pretty well in the main, and my play group has been enjoying that sort of play on a weekly basis since 1998, but the idea for AI War basically came out of my group's growing frustration with the weaknesses of that approach. Often those games are really short and repetitive -- an hour or two at absolute most -- whereas with co-op often you have a stable group of players and it is nice to be able to play a longer game over more sessions. There are more opportunities for growth and variety that way, and having long-term consequences to every decision is really cool. But many recent RTS titles have not even had multiplayer save games, which pretty much kills the longer co-op experience.
The other big problem we were having with the standard model was that the AI simply wasn't good enough to hold my playgroup's interest for longer than 6 months to a year (and even with that, we would be getting pretty bored with it by much sooner). Part of that was the AI, but part of that was that the variety just wasn't enough. Competitive multiplayer games are often touted as having near infinite replayability because you never know what the other team will do, and that provides variety even on non-random maps. I used to be big into competitive FPS play, so I know that largely does hold true. But with co-op games, they largely have the same replayability challenges of single player games.
My goal with AI War was to try to provide that same sort of longevity found in the competitive multiplayer scene, only in a solo/co-op context. That meant having random/procedural scenarios with a large amount of variety, and having an AI that would be much more interesting even at an expert level of play. Part of that involves having AIs that play really differently from human players, rather than just having them be a poor-man's stand-in for a real human opponent, which is what most RTS AI seems to be designed around. That makes for a really asymmetrical system, with a lot of scenario design possibilities that are foreign in the RTS space, but utterly common in many other genres: FPS, Platformer, etc. AI War is one of the few games where it would not be more interesting if there was a human opponent controlling the opposition instead of the AI, which I think is a key differentiator from "comp stomps."