Vox Populi Diplomacy Feedback

[Recursive]

Remaking this thread to collect additional feedback on Vox Populi's Diplomacy AI.

This is a companion thread to a series of tutorials/guides that I will be posting in this stickied thread: https://forums.civfanatics.com/threads/vox-populi-diplomacy-guides.683886/

These guides will explain how the Diplomacy AI currently makes decisions, which are intended to assist the community with proposing any improvements. They will be updated as any changes are made.

If you have any comments or ideas based on these guides or in general, this is the thread for that!

Edit:
In general, what I'm looking for is specific suggestions. Much like with the VP Congress, "X is broken" doesn't give me useful information I can use to improve the AI. As noted in the series of guides, that's why I'm working on explaining how the AI works.

 

[Slyceth]

It's very clumsy how the available resources by AI's are listed. You need to list ALL resources the AI can trade that are not impossible, and just give them a red tint if they're severely overpriced (x10 friendly price)
I don't want to constantly have to click through EVERY AI trading screen to find those hidden resources. That's 10-30 seconds depending on the AI playercount per turn extra, multiply by 500 turns that's 5000 seconds, 83 minutes of life wasted per civ game and I play about 3 epic speed civ games a day? :/

Also in the late game, where I might out of laziness just accept a deal where I overpay 12 gold per turn regular basis, I don't want to waste time refusing strategic resource deals for 5 gold per turn.

 

[CppMaster]

Isn't 5 gold per turn still high for a strategic resource?

 

[Recursive]

It's very clumsy how the available resources by AI's are listed. You need to list ALL resources the AI can trade that are not impossible, and just give them a red tint if they're severely overpriced (x10 friendly price)
I don't want to constantly have to click through EVERY AI trading screen to find those hidden resources. That's 10-30 seconds depending on the AI playercount per turn extra, multiply by 500 turns that's 5000 seconds, 83 minutes of life wasted per civ game and I play about 3 epic speed civ games a day? :/

Also in the late game, where I might out of laziness just accept a deal where I overpay 12 gold per turn regular basis, I don't want to waste time refusing strategic resource deals for 5 gold per turn.

Isn't 5 gold per turn still high for a strategic resource?

The resources being collapsed is actually a bug that's supposed to be fixed in the latest version - but as part of the UI, that's actually not an AI decision.

As for the prices, it's Deal AI, not Diplomacy AI, which decides trade prices. I'm going to work my way to Deal AI eventually but it's currently a mess (albeit improved compared to before after many reworks).

I should also note that I've only revised about half of the Diplomacy AI - the big part I haven't tackled yet is the dialogue system.

DIPLOAI_DISABLE_TRADE_OFFERS is an option in (1) Community Patch > Core Files > Core Changes > DiploAIOptions.sql. It can be toggled on midgame, and has an exception for offers to renew an existing deal. You can turn it on in the meantime to avoid spending hours on the trade screen.

 

[CppMaster]

That's 10-30 seconds depending on the AI playercount per turn extra, multiply by 500 turns that's 5000 seconds, 83 minutes of life wasted per civ game and I play about 3 epic speed civ games a day? :/

So you spend 4h a day just on the trading screen? Damm, boy, you may want to take a look at your priorities

 

[Recursive]

In general, what I'm looking for is specific suggestions. Much like with the VP Congress, "X is broken" doesn't give me useful information I can use to improve the AI. As noted in the series of guides, that's why I'm working on explaining how the AI works.

 

[CromulentRomulan]

One of the options in the diploAIoptions file is to disable Civs voting for another Civ as world congress leader. By default it is set to enable this behaviour.
This has on one occasion led to a Civ voting for me to be world leader and end the game. I think that maybe the default behaviour for this should be inverted and users can edit that file to enable it. It doesn't make sense for another Civ to make another win (I was playing with endgame aggression enabled also).
The cooperative wars setting in that same file doesn't appear to work. I play with that setting to disabled and frequently have AI's declare coop wars against me and others. This is both a source of frustration (for those of us that don't enjoy coop wars or bribed wars) and also can be a problem when it blocks you for paying off a Civ to make peace with another Civ (there's a cooldown number of turns before a Civ that agrees to a coop war will make peace)

 

[Anarcomu]

It doesn't make sense for another Civ to make another win

Well, in the same way vassals are dedicated to you, you could argue that a Civ that is too far behind to have any chance at winning would rather have another civ they like as winner.

 

[Stalker0]

Isn't 5 gold per turn still high for a strategic resource?

I generally consider the minimum trade value to me is 2 gp for 1 SR, aka one that I'm probably not going to use for a while. Anything lower than that is an insult. And note that is bare minimum, I often will not trade at those values if I think I will have any use for the SR.

Coal is in its own ballpark, it is SO much more valuable than any of the other SRs to me if I don't have 2 coal per city. Its easily worth 10x or more in value.

 

[saamohod]

I generally consider the minimum trade value to me is 2 gp for 1 SR, aka one that I'm probably not going to use for a while. Anything lower than that is an insult.

2 gpt for 1 SR in mid-late game is worse than insult. It's a joke. Usually at this time you're making hundreds or thousands gpt.

I press Esc in 99.9% cases of AI SR trade offers.

 

[Zuizgond]

The amount of gpt AI is offering for lux is at the right place in ancient/classical in my opinion but I still think they offer a bit too much in medieval and renaissance (don't know about industrial+). Sometimes they are willing to offer more than 25% of their income for one lux, which I think is a bit excessive.

 

[CppMaster]

2 gpt for 1 SR in mid-late game is worse than insult. It's a joke. Usually at this time you're making hundreds or thousands gpt.

I press Esc in 99.9% cases of AI SR trade offers.

2 gpt is better than 0 gpt if you wouldn't use the resource anyway.

 

[Stalker0]

2 gpt for 1 SR in mid-late game is worse than insult. It's a joke. Usually at this time you're making hundreds or thousands gpt.

I press Esc in 99.9% cases of AI SR trade offers.

I agree. I was stating my early game notions, with the assumptions those deals should scale up with time.

 

[Stalker0]

The amount of gpt AI is offering for lux is at the right place in ancient/classical in my opinion but I still think they offer a bit too much in medieval and renaissance (don't know about industrial+). Sometimes they are willing to offer more than 25% of their income for one lux, which I think is a bit excessive.

I agree with this (and think it gets worse in industrial). The trick is....the lux happiness is just much weaker at this poitn in the game. A few more happiness doesn't move the needle one bit, whereas in the early game it can literally be the difference between reasonable happy and rebellion. So people really shouldn't pay that much for them, they are mainly for WLTKD at that point.

 

[saamohod]

2 gpt is better than 0 gpt if you wouldn't use the resource anyway.

No, it's actually worse because I'm giving the AI (who is my real or potential competitor) strategic resource almost as a gift. I'm making them stronger and get nothing feasible in return.

 

[CppMaster]

No, it's actually worse because I'm giving the AI (who is my real or potential competitor) strategic resource almost as a gift. I'm making them stronger and get nothing feasible in return.

Not every AI is a real competition, though. A weak, farway civ won't harm you anyway. Selling them 20 unused strategic resources for 40 gpt is at least something. I wouldn't sell it to a strong competitor, though.

 

[CromulentRomulan]

Well, in the same way vassals are dedicated to you, you could argue that a Civ that is too far behind to have any chance at winning would rather have another civ they like as winner.

I don't think that argument holds water. For one vassals can and often do vote against your interests in the world congress and AFAIK they don't contribute votes to the world leader votes. In this particular game the AI that voted for me to win was the next most likely player to be win the game as they had made vassals of all other players except me and another AI, and so could have been reasonably assumed on course for a domination win. Thematically your argument feels like the AI behaving in a board game winmake manner rather than a grand strategy manner, which doesn't feel consistent with the rest of the game.
@Recursive another note on the cooperative war thing I mentioned - I've been sufficiently annoyed by it to learn that an AI neighbouring my territory and is extremely hostile on the opinion modifiers often won't declare war without having it be cooperative with another Civ, even when they could easily take me on. So there is probably something worth looking at in terms of this interaction with willingness to war behaviour of AI. It feels inconsistent that an AI will readily take a bribe to fight an unfavourable war but balk at a winnable fight unless they can make it coop war.
Last thing on diplo behaviour - world congress projects. It's less of an issue now that tourism has been nerfed a bit but it was really obnoxious that an AI could spam the tourism boosting world congress project. Is it possible that a cooldown can be placed on each of the world congress projects, similar to how a sanction cannot be applied to the same Civ in successive sessions (or how a religion cannot be proposed as world religion successively if it fails).
Another aspect of diplo/voting behaviour that was and remains jarring was that AIs will push through tourism boosting agendas even if another Civ is much better placed to capitalise on it to win than them. This would frequently assist tourism victories for AI and myself in games. I think the default behaviour should be to vote down those agendas unless the AI is leading in tourism influence

 

[N.Core]

This has on one occasion led to a Civ voting for me to be world leader and end the game.

Yeah, this is the problem with DV currently, AI can accidentally let someone win DV because they spent all of their votes on someone.
What AI should do instead is to make sure the vote is tied, by spending some of their votes and then abstaining some of it, in order to balance the votes and delay the DV for them to gain other VC types.

 

[Recursive]

another note on the cooperative war thing I mentioned - I've been sufficiently annoyed by it to learn that an AI neighbouring my territory and is extremely hostile on the opinion modifiers often won't declare war without having it be cooperative with another Civ, even when they could easily take me on. So there is probably something worth looking at in terms of this interaction with willingness to war behaviour of AI. It feels inconsistent that an AI will readily take a bribe to fight an unfavourable war but balk at a winnable fight unless they can make it coop war.

Thank you for the feedback.

On this particular point, AI gains +1000 coop war desire score towards players they're already targeting for war (the threshold to ask is 40). So it's less that the AI is balking at waging war without a coop partner and more that AIs which have friends and already want to go to war are highly likely to request aid.

 

[Rosete]

I'd like to suggest that weaker AI be more willing to band together against the top AI. Currently, in an average game of 8 players, you'll normally have 1-3 top AI, and 5 other AI trying to butter up to the top AI. I don't like this. The bottom AI buttering up the top AI only make the strong stronger. A weak AI partnering with a stronger AI to take down a strong AI isn't desirable to me, because now the surviving strong AI just "wins more". If multiple weak AI band together to take down a stronger AI, this now leave more opportunity for a weak AI to rise up.

I think it would be better if the bottom 5 AI be more willing to band together. Either forming a coalition of 5, or several groups of 2-3. The idea is that instead of the world splitting into 2-3 groups in direct compeition, now you'll have 4-5 groups.