AI behavior question

[Andy06r]

I have a real quick question for the AI gurus here.

I recently completed a cultural victory as the mayans on a standard tectonics map. I ended up on an island next to the Khmer's and holkan rushed them, with the other 5 civs on their island - this included washington, shaka, and kublai khan and they were constantly warring with each other.

A bit of political background - their entire continent is christian and they have built the apolstic palace (sp). When I discovered them, all 5 showed "annoyed" with me and since I had abandoned my military during the pre-astronomy days my power graph was pathetic. I quickly fixed the power graph, but I still expected to have issues with their tech trading bloc (which I couldn't break into) and AP.

But what actually happened is that my empire generated more gold than all 5 of them combined and I coasted to a very easy cultural victory around 1940. I understand that KK and shaka were throwing hissy fits, but I still expected a 5v1 crusade at some point that never happened.

Is the AI incapable of noticing "this guy has a massive economy" and "is close to winning a cultural victory?" I don't want AI's to arbitraily dogpile, but with most of the AI's at cautious/annoyed I expected more action.

 

[Janov]

Hi Andy,

from what I gathered, the AI´s don´t really play to win, exclusively. That means they will let you win in a certain way, even if they have the power to stop you from doing so. For example, let´s assume they "know" that you are about to build a spaceship. They will not change their strategy and start building an invasion-army to STOP you from winning - like a human player would.

There has been considerable debate to wether this should change (while it wouldn´t be easy to implement) - but there are also some points against changing this. Jdog 5000 is actually changing some code to have the AI´s more actively pursue "victory" in game terms for the newest iteration of betterAI, up to now the AI´s just kind of "played" their game, only happening onto a victory condition by chance.

Jan

 

[MadmanAtW]

Part of the human game is manipulating the diplomacy so that, essentially, civs will "let" you win. If you've put the effort into keeping Ragnar friendly, your reward for this is that he doesn't invade when you launch the spaceship. I wouldn't mind seeing the chance that he invades when you've started the spaceship go up at the levels of friendliness where he can attack you, but I wouldn't want to see everyone automatically declare.

 

[NotSoGood]

Actually long time ago the AI used pretty sneaky strategy to stop me from building the spaceship. It sent lots of spies in my cities and destroyed all process and finished spaceship parts. I was close to finishing the spaceship but after the attack of spies I lost because the AI sent the spaceship first and won.
I was way behind in everything but still it was nice because it felt like the AI had some kind of intelligence...

 

[TheMeInTeam]

Part of the vanilla BTS AI difficulty is that it has the bonuses but its stupidity allows it to functionally tag-team the player:

1. Accepting a permanent ally vassal (aka peacevassal) that's pushing for a culture win, and GIFTING IT TECHS. This is an example of the AI actively attempting to lose, while propping someone else up to beat the player.
2. Peaceweights allow it to have the only viable war target in sight be the human...on turn 0!
3. Demands only placed on human
4. Cheating to detect trades with civs it doesn't know
5. AIs that underbuild units can become food for warmongers. My custom AI abused this to hell.

Some of these things remain in betterAI (and would require game rule changes to eliminate), while others are being addressed. Of the above, only #1,#2 and #4 irk me a lot, and IMO betterAI absolutely should avoid #1.

 

[Cybah]

"3. Demands only placed on human" I'm not sure. In the current developer version you can observe many more war declarations on players who are actually at war. Individual dogpile or demand? I don't know.

 

[Pfeffersack]

"3. Demands only placed on human" I'm not sure. In the current developer version you can observe many more war declarations on players who are actually at war. Individual dogpile or demand? I don't know.

I think "demand" is here used in a very strict sense - tribute demands or ones targetting ending trade or entering war. The AIs among themselves have beside dogpiling a bribe action to get others into wars. Details on this here (#121):

http://forums.civfanatics.com/showthread.php?t=286180&page=7

I would like to see the AI using the same logic towards a human; it would not only be fair, but also make the game more interesting. Cancelling trades or declaring wars just because an AI demands it is seldom a thing worth to do.