Out of curiosity, does anybody know which (if any) XML file(s) give(s) the numeric details of aggressive AI? Since the discussion has mentioned pleased leaders declaring war, I've been looking through the leaderhead files and one of the categories is "<NoWarAttitudeProbs>".
So, for Alex and Catherine, for example (text truncated somewhat):
<AttitudeType>ATTITUDE_CAUTIOUS</AttitudeType>
<iNoWarProb>20</iNoWarProb>
<AttitudeType>ATTITUDE_PLEASED</AttitudeType>
<iNoWarProb>80</iNoWarProb>
<AttitudeType>ATTITUDE_FRIENDLY</AttitudeType>
<iNoWarProb>100</iNoWarProb>
Julius Caesar and Tokugawa, but also Mansa Musa and Lizzie:
Annoyed - 10%
Cautious - 50%
Pleased - 90%
Friendly - 100%
Gandhi, Augustus Caesar and others:
Annoyed - 20%
Cautious - 70%
Pleased and Friendly - 100%
Now, what exactly "50%" means I'm not sure, but what that does seem to show is that if an AI is furious and comes to the conclusion that it should war, it will. (This is good, IMO.

) It also shows that the AI won't war when it's friendly, ever. I'm wondering - does aggressive AI actually modify these values? Or, perhaps, does it simply make it more likely that the AI will come to the point where the modifier kicks in?
As an example: Perhaps on average, on normal AI, Julius Caesar hits the point where war is considered by the modifier 10 times in a game; if every opponent considered is pleased, he can be expected to only declare 1 war. If Julius Caesar can be expected to declare twice as many wars on aggressive AI does that mean that his pleased modifier has dropped to 80, or war is considered 20 times?
If it's the latter (which I think it is), then this is the key point for many of us arguing in favor of aggressive AI over normal AI (in terms of how it should be, not necessarily difficulty): The AI considers war a more viable option, and in so doing, reduces the viability of war for the player. War tends to be the strongest option on normal AI, so weakening it for the player makes for a more balanced game.
(It can, of course, be argued that the military emphasis on aggressive AI makes it easier to out-tech and win peacefully, but I've not found that to be entirely true - usually one or two more AI than normal lag behind, but enough keep up to maintain tech parity by trades and the like; also, peaceful strategies tend to work better on less military which is a greater risk when the AI has placed more emphasis on it.)