I'm not arguing simulationism for the sake of simulationism. I'm arguing increasing options available to the player because that almost always make games more fun.
I don't really care that the AI can't handle it either. 1) it's possible to improve AI, 2) good AI isn't necessary for enjoyment see : Deus Ex.
Actually, 1) is not really true : You can improve AI, but you could improve it too in a simpler system in which it would have performed better. The AI of the most complex game will always be weaker than that of the simpler one. Go, which has no hidden information, and a relatively small game space compared to any skirmish game, is already hard for the AI (it should beat all humans soon, but we are not there yet).
So a system with too many viable options (if there are a few dominant ones, the AI will always select them, making the real complexity of the system lower) will always result in a weaker AI for the same amount of effort than a relatively less complex system.
That's why it is impossible to program an AI that will not suck for a game like civilization 5 (heh, we are back to the topic!) :
It is possible to have the AI handle formations correctly (which for some weird reason has not been done), and the tech tree and build order should not be too hard for the AI to max, but choosing on what to spend most efforts, when to build military unit, where to expand, guessing the positions of opposing units and managing diplomacy in a non retarded way are very hard problems.
A simpler game, like Armageddon Empires, manages to give a proper challenge without having the AI cheat at all.
2) I agree with this. You can make the game around a mediocre AI if it is asymmetrical enough. It actually worked well in old X-Com, and it's the way they fixed Warlock in Armageddon mode.