By Brian Reynolds
Even in this new "Age of Multiplayer Gaming," one of our most important goals at Firaxis is to create games that have (pardon my buzzwords) lots of depth and replayability in single-player or solitaire mode. The Internet is cool, multiplayer games are cool, and Firaxis plans to include full multiplayer support in all of its games, including Sid Meier's Alpha Centauri. So many companies are turning away from single-player play or think of it as mainly a "tutorial mode" to be used before graduating to the "real" multiplayer experience, but we think a vast and mostly silent majority of gamers still spends most of its time playing solitaire - and we plan to be there.
So what gives a game the coveted replayability that turns a single-player game into a classic? Well, a lot of things, of course, but one of the most important is good AI - smart computer opponents. One of the most common reasons gamers give for preferring multiplayer games is that human opponents "are smarter and make more intelligent, unexpected moves" or conversely that in single-player games, "the AI is crappy." Knowing this, some developers are ready to throw in the towel on single-player AI and go all multiplayer. As a designer, though, I take it as a challenge to try to create the kind of algorithms that keep players coming back for more.
I've often been quoted as saying that when designing AI "I try to make the computer play the way I do." At first glance, this may seem like a meaningless statement, since by definition artificial intelligence is an attempt to emulate human decision-making. But what I am really describing is the process, not the end product. Now I'm (blush) a pretty good strategy gamer, so I begin by playing the game myself, keeping track of the decisions I am making and even more importantly why I am making them. I also turn on the omniscient view and watch what the computer players are up to, making notes when they do something "dumb." The trick with AI is to find the right "angle of attack" for a particular problem - when a computer is faced with a choice between several attractive alternatives, I want it to be able to think through the problem the same way I did.
Next: An example of AI fine-tuning