The Tom Ireland Interview on StarTopia Post features this programmer at Mucky Foot talking about his work on the AI, among other things, on StarTopia, their space station simulator. The conversation talks a bit about how the game turned out, and goes into some detail about StarTopia's artificial intelligence, discussing some of the coding issues involved in its creation. One topic of interest was a discussion of the classic perception by lay people that every game's AI should be better than it turns out. Here is the answer that touches on this perception:

One difficulty is that raw data and valuable information are completely different things. Going back to chess, knowing what squares the pieces are on isn't enough to make you a grand master - in fact it puts you about level with a photo. To get anywhere worthwhile, you have to concentrate on far sketchier things - risk, flexibility, weaknesses or what the human player might do five seconds or five minutes down the line, and that's exactly the sort of vague, ill-defined, wishy-washy stuff that computers have trouble with. Computers need absolutely everything spelled out for them, and describing most AI problems accurately is a pretty tall order, so AIs usually fall short. Once enough clever AI methods are out there the standard will pick up, but it'll take a while! As faster processors arrive, slower techniques will move into reach too, so the outlook is good.