[NFP] [1.0.11.16] Predictable AI voting logic for Border Control Treaty resolution

[Pfeffersack]

Intial report with 1.0.0.341:

I have never seen any AI not voting here "A" for themselves - and to make it even worse, also not a single time any AI investing more then their free vote. The serious consequence is that a human player just needs to vote "A" for him-/herself as well with spending 10 DF for a 2nd vote - and both the culture bomb and a DVP are earned.

I can dig up a save if needed, but I'm not sure if it really that helpful, as it would just show that behaviour in one situation, but not the fact it is occuring always (or is there anyone with different experience? Please correct me then, thanks)

Three possible fixes (1+2 best combined):

1. If AI continue to go for A and only themselves, make them spend more then the free vote from time to time.
2. Teach them that B on a disliked civ is an option as well...
3. Rework the entire resolution. IIRC, it is the only targeting a specific civ in a purely positive or negative way - likely thats responsible for the AI Voting behaviour (understandable for each AI to an extend, but disastrous on the whole). It could be replaced by attaching culturebombing to a special district as A resolution variant (and ban to build a certain kind of district as "B" variant) for example, which would surely cause are more varied voting behaviour.

 

[Pfeffersack]

I tried to reproduce and ironically can only come up with a near-perfect example of this - in fact, I saw the first time something contradicting my initial statement slightly: Beside the usual behaviour, I saw one AI voting B, but with 1 vote only, too. However, even if this is possible under rare circumstances, there is still the big problem left of the AI only casting one vote on A directed to themselves, while the B votes seem to happen only very, very rarely (too rarely) to make any difference.

To reproduce:

1. Load save
2. End turn
3. WV voting comes up:
- 1st voting decision is about grievances . The human player vote here should not affect the problem discussed here, but I casted my free vote on A one myself (Catherine de Medici)
- 2nd Voting is the one discussed here. Spend 10 DF to place two votes one A for yourself (Catherine de Medici)
4. confirm, get the result 2nd picture shows

 

[Qwan]

AI voting is often predictable, although you have to play a lot and keep track to learn this. Although they still sometimes surprise me.

As for Border Control in particular, usually by the time it is available, most AIs are done building most of their districts, or you and they are building districts farther from the city, and the bomb from this helps less (you don't gain tiles that are outside the City Radius). So what the AIs are doing is not spending any diplomatic favor, and forcing whoever gets the bonus to spend theirs.

However, I agree with you that ALWAYS spending one vote for themselves is too predictable. Someone with a lot to gain by winning ought to, like the human, expect that one extra vote would be worth spending on themselves.

 

[Pfeffersack]

(…) As for Border Control in particular, usually by the time it is available, most AIs are done building most of their districts, or you and they are building districts farther from the city, and the bomb from this helps less (you don't gain tiles that are outside the City Radius). So what the AIs are doing is not spending any diplomatic favor, and forcing whoever gets the bonus to spend theirs. (…)

I have that resolution in nearly every game and it can come up in the middle age, where there is still room for culture bombing.

But even if it comes later - wouldn't it still make then more sense to try voting B then and punish someone? No natural gaining of tiles is a severe punishment IMO, even if its a bit later in game - it hits both when someone is doing late colonization or if bordering nations compete over tiles when cities are close, because it means you are forced to buy tiles (if that is possible at all if the ban is enacted - I can't even say if you are still able to buy tiles then...because I have never seen the B option getting enacted at all, not vs. me nor an AI )

---

Below is another example, marathon speed game turn 480, just end turn - this time the "perfect case" of all AIs voting A for themselves with the free vote:

 

[Qwan]

I browsed thru your game with the Empire lens. I found 4-5 places where districts were being built where culture bombing would have gained tiles, like 2-3. Not counting your dominant civ. In contrast, cities would grow 2-4 tiles from culture for the duration of the resolution. Since most cities would not be building districts, particularly not on borders which could expand, Civs have a LOT more to lose if they get hit by B than if they win A. More cities means more possible district bombs, but also more tiles through culture expansion, i.e. the valuation of A vs. B is (more or less) independent of number of cities. So it's not clear who to target with B. Or even if there is much value in winning A -- after all, you could just buy those tiles, if you really wanted them.

I guess if I were the AI, and weren't allowed to prognosticate how other Civs were thinking, I'd play it safe by choosing A for myself. It decreases the likelihood that B will win, and risking being hurt. And if A wins, it will cost someone, and benefit them hardly at all.

However, I can see your reasoning that IF the AIs knew who to target, B would be a better choice. If someone had both the highest score and the most cities, I'd target them. In your game, that's you, and really, everyone else should be ganging up on you. (Although Russia has more civics, implying more culture, and so should be hurt more.) But if everyone gangs up on you by voting B, your best counter is ALSO to vote B, and with your overwhelming diplomatic favor really hurt one of them. Which again leads to the conclusion its safer to vote A.

Maybe if you didn't dominate so easily, you might see different voting? Try a level of difficulty where you end up being one of the weaker ones!

I have that resolution in nearly every game and it can come up in the middle age, where there is still room for culture bombing.

But even if it comes later - wouldn't it still make then more sense to try voting B then and punish someone? No natural gaining of tiles is a severe punishment IMO, even if its a bit later in game - it hits both when someone is doing late colonization or if bordering nations compete over tiles when cities are close, because it means you are forced to buy tiles (if that is possible at all if the ban is enacted - I can't even say if you are still able to buy tiles then...because I have never seen the B option getting enacted at all, not vs. me nor an AI )

---

Below is another example, marathon speed game turn 480, just end turn - this time the "perfect case" of all AIs voting A for themselves with the free vote:

I have that resolution in nearly every game and it can come up in the middle age, where there is still room for culture bombing.

But even if it comes later - wouldn't it still make then more sense to try voting B then and punish someone? No natural gaining of tiles is a severe punishment IMO, even if its a bit later in game - it hits both when someone is doing late colonization or if bordering nations compete over tiles when cities are close, because it means you are forced to buy tiles (if that is possible at all if the ban is enacted - I can't even say if you are still able to buy tiles then...because I have never seen the B option getting enacted at all, not vs. me nor an AI )

---

Below is another example, marathon speed game turn 480, just end turn - this time the "perfect case" of all AIs voting A for themselves with the free vote:

 

[Pfeffersack]

Nice analysis and good aspect about possible fear of getting hit by B leading to A votes having the additional benefit of being on the safe side. You are also right that this game (Emperor difficutly, but Marathon and Terra map type) with my domination might to be taken with a grain of salt, however as said in my OP the described behaviour is occuring (nearly - and that nearly is just the one lonely B-voting civ in the first example, the first abberation I witnesses since playing with the GS expansion at all!) always, independent of game settings like difficulty, progress of the game or my position. You might be even right with your assumption that a low effeort A voting is often or even always a individual "effective" voting behaviour, but then the entire resolution needs a facelift IMO (because I think a resolution which design makes Voting always in the same way "right" needs change - I made a proposal with 3) in the OP).

 

[Qwan]

I'm curious how an all-human game plays this resolution. Anyone want to chime in?

I think AI voting (and AI play) could be made more competitive by observing all-human games, and deducing the "rules" they use.

 

[Pfeffersack]

Voting behaviour here is still the same in NFP 1.0.5.11.

This time the same is from a Deity game. Out of my 19 opponents (Highland map script just invites to overcrowd with so much space), 17 go with a single "A" for themselves - which allows me to seize the culture bomb bonus for myself for just investing 10 DF for a 2nd vote:

Spoiler :

Many of them have plenty DF and it is the very first WC voting (begin of MA) in this game, so tile grabbing with districts is still something valuable. So what I would like to see changed...well, for my taste the entire design of the resolution is a bit problematic, since it is the only targetting a special civ and the outcome either only benefitting or harming the target (I made a suggestion for change in the OP) - that inevitably leads to a natural tendency for voting for yourself positive ("A") But if we stick with that design and if still many of the AIs follow the "Vote for yourself"-route, they should invest at least a bit more. You just can't expect to get something out of this style of a resolution with a single A vote. If a civ isn't willing to spend DF for extra votes, it should go "B" - than there is at least the chance that some votes from different civs on a "bad boy" come together and have in impact (both on the target and for a diplomatic victor point).

To reproduce with the attached save:

1. End turn
2. In the upcoming WC session, cast 2 "A" votes for the players leader Ambiorix in the Border Control treaty voting (the choice in the other voting shouldn't matter; I voted "B" for Cultural CS)
3. See that you easily grab the win here, because most AIs just cast a single "A"-vote for themselves

 

[Pfeffersack]

This hasn't changed with 1.0.6.9 - the example below is one that shows IMO clearly that there is something with the weighting of the Border Control Treaty resolution wrong. As all examples in this thread show, the AIs are constantly refusing to put extra DF into votings about the BCT resolution (regardless if they do the most common "A" vote for themselves, but also in case of the occasional B-Vote against someone else). In some cases, this low spending might be justified, if the other vote is particularly important for a civ, but a general "only-free-votes"-strategy feels wrong...especially because the AI even strictly do that if the other voting has fairly limited impact - the example of this post is out of Marathon game, where currently several GPs types an unavailable, because all GPs have currently been taken:

Spoiler :

So now the Border Control treaty and the one about doubling/disabling GPP generation common up and this is what the AI makes out of this situation:

Spoiler :

So all the AI pour DF into a voting, where only 4 GP-types can be voted about. I'm ok, when particularly cultural or industrial civs do that, but for me it is highly unlikely that this matters for all AIs that much (even if the common goal is to hit Peter with hs Lavras, it is questionable IMO to put that much effort into it)


To reproduce with the attached save:

1. End turn
2. When the WC votings come up, vote as shown in 2nd pic and confirm
3. See the Ais vote like shown in the 3rd/4th pic
4. Examine if the voting behaviour shown under 3) is a good strategy for the AIs, especially when taking in account the GP situation (1st pic)

 

[Pfeffersack]

Unchanged in 1.0.11.16:

Spoiler :

To reproduce with the attached save, just end turn, vote with 2votes for the player in the BCT resolution and see above voting behaviour of the AIs.