dh_epic
Cold War Veteran
The short answer is that it's damn hard.
The reason it's damn hard is because programming is like an argument: "If this then that, but if not, then this other thing..." Sometimes it's like a recipe "look at 10 things, and flag the three biggest ones."
So much of human cognition is not an argument or a recipe. What is the recipe for sight? What is the argument that tells you where one object stops and another begins? A monkey can figure out what an object is. But a robot can't even tell an object from its shadow. That's why robots on assembly lines have to work with the lights turned up all the way.
There's even math problems -- especially in the form of pathfinding -- that humans can solve relatively easily, but a computer can only find the best path with 100% certainty if it goes by trial and error.
Of course, we're not asking for Firaxis to build The Matrix here. And there's a lot of small things they could do that go a long way. The main issue is that it takes a lot of time and thus money to keep improving the AI, when all you really need to do to sell a game is write "better AI" on the back of the box. Moreover, the more things that the AI "thinks" about, the slower the game goes.
Everyone thinks that the AI is bad. But trust me, it could be much, much worse. The AI improved light years since Civilization 3.
