Investigation of AI reassessment upon reload

[roguebolo]

Pode,

When I tested from midgame, both with the savegame entitled "DefendThapsus" and with the sequence in which I did 80 moves without save loads (which is still running in the background), the AI never ceases its aggressiveness toward the human player. To be quite honest, the AI doesn't have much of a choice; in both situations, those AI factions had no one else to be aggressive toward; in "DefendThapsus" it was seeking to retake its territories from an aggressor - myself. (In general this seems to trigger aggressive behavior on the part of the AI.)

The difference was the way in which it manifests its aggression. The AI tends to attempt to maintain a siege without a save/load, or relieve one with save/loads, resulting in alternately establishing and relieving sieges when there is a load game each turn. In many cases, relieving the siege gives it a better chance of doing damage to my armies if I decide to attack it in open terrain, since it is clearly outmatched. "DefendThapsus" is particularly illuminating because if you play with save/loads for several turns, Brittania will start besieging two targets in the north (Lemonum and Condate Redonum), while Numidia is besieging two settlements in the south (Carthage and Thapsus). As a result, you can observe the behavior of four sieges simultaneuosly.

However, if enroute to a siege, a "save/load" does not cause the AI to forget its target. So the targets never really change, just the way it manifests its aggression.

In the 80-move test I did observe the AI choosing to relieve sieges despite the fact that there was never a load game. In "DefendThapsus" and similar savegames where sieges were in effect, I have not yet seen it elect to maintain a siege after a loadgame except on rare occasions wherein the target is a weak rebel territory. Bromley and others have reported that there are indeed such instances, but that they are limited to certain settlements which appear to be "immune" to save/load behavior. In the long run, though, I don't consider alternately relieving and reinstating sieges to dramatically effect gameplay. I still need to eventually oust the assailant from my territories.

After allowing the AI to play 80 turns without any load games, I also don't consider the general passivity of the AI to have any dramatic effect on the overall political landscape, with the exception of the fact that loadgames will tend to cause the AI to leave rebel territories unclaimed. Without them, the AI will snatch them up in what I've referred to as a "grabfest". After observing numerous tests in numerous starting scenarios, the behavior of the AI toward rebel territories is still my main complaint regarding loadgames.

80 turns is about 1/8th the maximum turns allowed in a game, and maybe 1/4th the moves I would actually expect in a game. A few minor exchanges of the AI factions' territories changed the way that I percieved certain factions, their "presence" relative to other factions, but did not provide me with any particularly strong factions. However, the takeover of rebel territories had an absolutely profound effect on two factions in particular, Pontus and also the Brutii after Germania went bottom-up. If one considers that Dacia is a Brutii protectorate, the combined alliance will be almost as strong as I am (at least in terms of number of territories) after the Brutii manage to establish civil order in their northern territories. (They've been trying to do that for over thirty turns, though.)

I don't think that pressing "End Turn" for 80 more turns without load games is going to provide much additional information. That will be nearly one half what I expect the entire length of the game to be, and i would expect to have won the game by that point, so it's a fruitless effort.

For me, the reasons why it's screwed are not really important. Even if we knew exactly what was wrong, that doesn't help us if CA refuse to consider it as a bug/screwed feature.

Well, an understanding of the AI's behavior is the only thing that's going to lead to a workaround. I can force the AI to behave in certain ways by the use of alliances, but I do not seem to be able to force it to change its behavior toward rebel provinces, where alliances have no effect. If a trigger can be identified for midturn-reassessment, then pulling that trigger might galvanize the AI into some action by forcing it to select the next "equally viable" option in the list.

Such information would undoubtedly have been useful to the developers at some point, but my guess is that they have already studied the situation and understand it more fully than any theorizing that we have done here; and that they have concluded (like myself) that this is not exactly a game-breaker and that any changes we see in the behavior of the AI is reserved for future releases.