So I'm lining up my forces along the border while playing on, I believe, easy or second-from-easy difficulty. All of a sudden a pop up appears with the face of the leader that I was about to attack. The message said something along the lines of: "I notice you're lining up your forces along my borders. This is obviously a preparation for an attack of some sort. If the difficulty setting was higher I would take this as an offensive gesture and respond accordingly. However, due to current difficulty settings I am unable to react correctly to the situation."
That's a terrible example. The AIs in Alpha Centauri did that (and IMO did a better job of it too) (and Alpha Centauri came out well before GalCiv). It just doesn't qualify as anything special to me - for one thing, it's SUCH a lame thing for the player to do - hey look, we have these big armies next to all your stuff, but of course we're at peace, don't worry - the fact that the AI responds to it is sort of a basic requirement, it's just part of being able to RECOGNIZE that a state of war actually exists and to act appropriately, regardless of whether all the punctuation is correct on the paperwork. It would be just as bad for the AI not to respond to outright destruction of some of its units; the fact that it does respond to that is nothing big.
I'm talking about an AI that could actually interpret ongoing results in combat and draw correct conclusions, such as: "the matchup of my ship A vs his ship X tends to result in so-and-so, while my ship B vs his ship X results in this-and-that, THEREFORE since he has so many ship Xs I should be building more ships with these particular characteristics" ... or, "the last five times I sent troop transports loaded with soldiers to that planet, they got blown out of the sky. I should stop, and put a heavy combat force in orbit and keep it there for a while, and certainly not send it anywhere else until a set of heavily-escorted troop transports arrives". In other words, actual dynamic analysis based on relatively simple rules that actually changes behavior in important and challenging ways for the player.
The Alpha Centauri AI really is my baseline - anything that doesn't do as well as it did is unacceptable, and I certainly hope for improvements. For one thing they are more capable of recognizing when they're getting their asses kicked on the battlefield and of suing for peace before taking crippling losses of important things, like bases, as opposed to unimportant things, like troops. In GC the AIs would seem to look at the monumental number of ships they had and the paltry number I had and decide the war was going well; the fact that they were taking hundreds of times the losses never seemed to register.