The Effects of Culture, Religion and Cores on AI Tendencies: a controlled experiment

[dharper]

After you have tested and eliminated everything else then try making SPA and other nearby countries weaker. I have noticed that when I make SPA stronger then FRA just sits there as if waiting to be attacked.

Spain doesn't exist, much like the rest of the world. The only countries in existence are France and its several one-province neighbours, none of which has an army over 1,000 to start (some have leaders I couldn't be bothered to eliminate), none of which are in alliances since they can't see each other.

I actually went ahead and gave France a 'hitlist' of its targets - and still it sits there, for 20 years, without declaring war.

Given all the times I've wished the AI would declare war a little less, this is more than a little ironic.

One possibility is that there are too few countries in the world - yet in my initial tests the AI was declaring war as soon as it got a diplomat, and the number of countries is the same.

I'm stumped, folks.

 

[unmerged(16460)]

I would suggest that you change FRA's tag to U12 or something, as U12 is less likely to be hardcoded in anyway. I would suggest that you do the same thing with all the target nations.

Good luck, I really apreciate your initiative!

 

[dharper]

An update: I do get DOWs, but the time-to-DOW has increased from a few months to 11-20 years. I hex-edited my EU2.exe file to allow it to run in the background and sit back, checking every 15 minutes or so - that works, albeit slowly.

 

[unmerged(26889)]

Afterwards, I doublechecked the target provinces to make sure they all had the same income. But of course they didn't. Of course? The different state religions have different income penalties, so there's a slight difference between them.

You can eliminate these differences by editing religion.csv...but you probably know that, don't you?

 

[DSMyers1]

I thought I'd throw out the results of my first batch of tests, all on hard/coward. I suspect that the aggression setting probably has a huge effect on these results. Anyway, I documented which were force annexed and which were diploannexed. These may indicate which the AI likes to target with wars, and which the AI prefers to deal with diplomatically. It must be noted that the version I used did not have the core edited in, and it still took the AI a long time to do anything, likely from the aggression setting. It is assumed that everytime France DoW'd that it annexed. I edited the AI to be extremely aggressive, but not to target any. I will edit in some comments on the information tomorrow.

Reminder on what the countries are:

France (test subject): France owned many provinces, all of which were national provinces (had cores), were Catholic, and had French culture. All of them had tax collectors as well (in earlier tests France spent some time building them, which slowed down my experiment). France also had cores on the Huguenots, Brittany and Burgundy. The French AI was set to 'default.ai' and given a starting army of 50,000 as well as a starting navy of 20 ships (although they did not use them, France began building ships as soon as the scenario began, so I gave them some).

Burgundy (target): Burgundy owned Bourgoune province and was French Catholic, claimed by France. Culture & Religion + Core

Brittany (target): Brittany owned Moribhan province and was Gaelic Catholic, claimed by France. Religion + Core

Provence (target): Provence owned Provence province and was French Catholic. Culture & Religion

Navarra (target): Navarra owned Béarn province and was Basque Catholic. Religion only

Huguenots (target): The Huguenots owned Poitou province (to avoid them neighboring Brittany) and were French Reformed. Culture + Core

Lorraine (target): Lorraine owned Lorraine province and was French Reformed. Culture only

Flanders (target): Flanders owned Flanders province and was Dutch Reformed. Core only (NOT IN THIS TEST)

Switzerland (target): Switzerland owned Bern province and was Swiss Reformed. Control - no similaries to France

My data:

******Hard/Coward*******

WITHOUT CORE:

Trial 1:
France got:
Provence, Diploannex, 37 years
Navarra, Diploannex, 113 years
French Catholics, Force Annex, 143 years

Number of Years: 345


Trial 2:
France got:
Burgundy, Force Annex, 42 years
Brittany, Force Annex, 65 years
Provence, Diploannex, 82 years

Number of Years: 83


Trial 3:
France got:
Huguenots, Force Annex, 44 years
Navarra, Diploannex, 47 years
Burgundy, Force Annex, 80 years

Number of Years: 84


Trial 4:
France got:
Navarra, Diploannex, 53 years

Number of Years: 83


Trial 5:
France got:
Provence, Diploannex, 45 years
Navarra, Diploannex, 49 years
French Catholics, Force Annex, 131 years

Number of Years: 209


Trial 6:
France got:
Burgundy, Force Annex, 11 years
Brittany, Force Annex, 12 years
Huguenots, Force Annex, 55 years
Huguenots, Diploannex, 115 years
Provence, Diploannex, 120 years

Number of Years: 206


Trial 7:
France got:
Brittany, Force Annex, 21 years
Huguenots, Force Annex, 25 years
Burgundy, Force Annex, 43 years
Flanders, Force Annex, 43 years
Helvetia, Force Annex, 125 years

Number of Years: 207


Trial 8:
France got:
Brittany, Force Annex, 19 years
Burgundy, Force Annex, 74 years
Navarra, Diploannex, 89 years
Huguenots, Force Annex, 142 years
Provence, Diploannex, 164 years

Number of Years: 203



I must add that I was running these games simultaneously, as may be noted from the similar time intervals of some. I do not know if that causes the AI to behave differently than when running a single game at a time.

 

[unmerged(31425)]

What an interesting experiment. I’m hooked, please keep us updated.

As idontlikeforms mentioned you should equalize the provinces, so they are 100% identical. It does have a say in colonization, so it probably have a say in war too. Sea and landlocked province might also make a difference.
And location of the country, like BUG is inside France might also be a factor.

I think I will test along with you…

The way it has a say in war is by the economic strength of the AI that could possibly be DOWed. The French AI will consider it vs his own.

BTW Daywalker, do you intend to respond to my PM? Or is it a matter you're not knowledgeable on?

 

[unmerged(31425)]

Dharper, you should also key in on the relations. If those cores are making the core owned countries deteriorate in relations faster, we don't really know if France is DOWing them because of cores or because they technically have worse relations by a few points. Perhaps putting them all on -200 to start, like Norre suggested in the other thread? -200 should speed up the DOWing too and save you some time playtesting it.

 

[eon 2]

An update: I do get DOWs, but the time-to-DOW has increased from a few months to 11-20 years. I hex-edited my EU2.exe file to allow it to run in the background and sit back, checking every 15 minutes or so - that works, albeit slowly.

This may be silly, but: Are You sure that when You added Flanders you didn't introduce a code bug ? Have You checked the savefile for the cores ? (I don't know if You start from a savefile or from "scratch".)

 

[unmerged(34273)]

Would be interesting to see if its the state culture or province culture that matters.

LEts argue that culture does make the ai dow more. Would ai then dow a holder of accepted culture even thought the state has no culture. In that case the ai would have to check each province controlled by a nation.

 

[dharper]

Experiment 2

Hypothesis​

When an AI is given the opportunity to attack and all else being equal (casus belli, low relations, overwhelming economic/military advantages over the target's alliance, all targets with same number of CoTs, all targets with same income through taxvalue and goods production) and several targets that fit this criteria present itself, it will:

(1) ALWAYS attack a country that it has a core on; and

(1a) If more than one exists, it will give preference to ones with the same religion and culture (but not always choose them); or

(2) ALWAYS attack a country that has at least one provence of the same religion; and

(2a) If more than one exists, it will give preference to ones with the same culture (but not always choose them); or

(3) ALWAYS attack a country that has at least one province of a state culture.

Setup​

In this experiment I examined pairs of targets against each other to determine AI preferences. To do this I eliminated all but four countries from the world - the observer nation (Zimbabwe), subject nation (France), and two targets, chosen for their qualities (Core, Religion and Culture) and all being one-province minors neighbouring France with the same tax value and goods production. The only change from experiment 1 for targets is that they no longer have CoTs, as I was worried there was too much randomness (and exploration) induced by them. Now there is only one global CoT (in Paris), which seems to be stable (!).

Five tests were performed for each pair, and the number of times each target was chosen first was recorded. In cases where no target was chosen in the first 30 years of the game, the test was discarded.

Results​

I will post results by pair as I finish them - since some tests have to be thrown out and it takes a long time to run through 30 years, progress is slow and it should take some time to show all pertinent matches.

2.i Flanders (Core only) vs Provence (Culture and Religion)

Flanders
4

Provence
1

Conclusion: This invalidates hypothesis 1; the AI has a strong tendency to choose core provinces, but does not always choose them over other, more attractive options.

2.ii Flanders (Core only) vs Navarra (Religion only)

Flanders
5

Navarra
0

Conclusion: The results suggest that the AI usually or always chooses a core over a same-religion province.

2.iii Flanders (Core only) vs Lorraine (Culture only)

Flanders
5

Lorraine
0

Conclusion: The results suggest that the AI usually or always chooses a core over a state culture province.

2.iv Provence (Culture and Religion) vs Navarra (Religion only)

No results yet.

2.v Provence (Culture and Religion) vs Lorraine (Culture only)

No results yet.

2.vi Navarra (Religion only) vs Lorraine (Culture only)

No results yet.

 

[unmerged(17856)]

Nice work dharper

As someone who'll likely be running experiments for a living, I'll play the evil scientist for second here.

What about the tag variable? Your results are potentially confounded with the tag. In experiment 2, the core assingment seemingly elicited the most DoWs, but Flanders always had the core. One could argue it may have been tag choice that infuenced the DoWs. Running control conditions where you simply reversed the tag assignment within each of the three conditions would settle that nuance. Provided the results of experiment 2 were then replicated in this third experiment, you would then be armed with significant evidence that the core assingment elicits a substantial and robust main effect on DoWs!

Curiously, there may be some room for an interaction(s) here; when you combined religion and culture there was at least 1 DoW, whereas culture and religion, on their own, resulted in no DoWs. It would be interesting, in addition to the above controls, if you could examine the effect of combining culture, religion, and core against that of simply a core with wrong culture and wrong religion in a fourth experiment. For this you'd run a seperate group of comparitive control trials, where both tags retained a core with wrong cultures and wrong religions. You could then compare the distrubutions of DoWs between the two groups. If you can, run the experiment with as many trials as you can within each group, I'll run some stats on your results. The more trials the merrier, as the interaction may be subtle.

EDIT: Note that having established that the tag has nothing to do with the DoWs in experiment 3, you would be free to assign the conditions of experiment 4 to whichever tag (of the pair) you felt like, without counterbalancing in seperate controls (which would have doubled the number of conditions you would have to run).

 

[unmerged(2456)]

Another thing to keep in mind is that it could be if the AI has too many targets, it can't chose which is best, especially if they're all similar and just sits their in awe of the possibilities (like a kid with a little money in a candystore).

I know for a fact that some engines based of the EU2 engine have this problem.

 

[unmerged(17856)]

Another thing to keep in mind is that it could be if the AI has too many targets, it can't chose which is best, especially if they're all similar and just sits their in awe of the possibilities (like a kid with a little money in a candystore).

That's what I figured too, but if I understood dharper's experiment 2 correctly, he used only two tags for targets and eliminated the rest from his experimental setup.

 

[dharper]

That's what I figured too, but if I understood dharper's experiment 2 correctly, he used only two tags for targets and eliminated the rest from his experimental setup.

That's exactly right - sorry, I wasn't clear and should add that.

Even with only two targets, the computer still delays. The occasional game has a DOW in months, but more often than that 30 years pass without one and I have to restart (by then relations have changed too randomly, the smaller countries start to discover France and the Holy Roman Emperor gets chosen). The average seems to be 10-20 years.

Luckily, I can just run it in the background and come back to it every 20 minutes or so - but it IS time-consuming. If only I could get it working again so the AI declared war immediately, I could make this go much faster.

 

[dharper]

What about the tag variable? Your results are potentially confounded with the tag.

A possibility I hadn't thought of! Definately worth pursuing - the AI doesn't have any countries in its hit list and has only default ai, but it's possible that something else could be at work...even the geographic location, come to think of it. All the targets neighbour France, but it's possible the game sets different standards on 'France' vs 'Netherlands' - however, Havard's list of regions suggests that all of France and the Low Countries are in the same region ("Western Europe").

I'll definately look into this with experiment 3, although I want to finish #2 first and see how the non-core combinations compare.

Curiously, there may be some room for an interaction(s) here; when you combined religion and culture there was at least 1 DoW, whereas culture and religion, on their own, resulted in no DoWs.

I noticed that and desperately wanted to increase the number of trials to at least 10, preferably more, just to see how common this behavior was - but given how slowly I'm getting DOWs, I had to make the decision to move on to test other combinations.

 

[unmerged(31425)]

But what about the relations? Is the country with the core consistently having worse relations? W/O knowing this, we don't really know if the core is instigating the DOW or not as it could just be because of worse relations. Core claims cause relations to deteriorate, but I don't know if it makes them deteriorate faster than a temporary CB or the same.

And also, I still don't see even a scrap of evidence that the AI is paying any attention whatsoever to culture or religion.

 

[DSMyers1]

I thought of a few more ways to start up the wars faster: perhaps set the AI on Very Hard/Furious and give the targets plenty of BB. Hopefully that will get a response.

If you use the windowed patch, with the run-in-background patch, you can run multiple instances of the game simultaneously...though I am not sure if there are interactions. I ran as many as 4 tests at a time....

 

[unmerged(2456)]

Yea its best if your testing to use one of the user-defined tags. Those don't have any impact for certain.

It might also be best to change all religion variable to the same thing, reguardless of religious type. Then change commodities to the same (if you haven't), manpower, etc. n provinces so they are all the same. Then set the dp sliders to the middle.

 

[DSMyers1]

Okay, I have created a perfectly balanced scenario for these tests. To expedite things, use very hard/furious. You will get a DoW within the first few days...because the two minors each have 100 badboy. Everything between the two minors is completely balanced....completely. Each has -200 relations with the major, which is U01. Also use the attack.ai:
Completely balanced AI test.eug
attack.ai

And here are my initial results, only dealing with religions: Number of DoWs by the Major:

Religion Test 1: Provinces Identical (catholic),

Test Major Catholic (BB 100 both)
Britanny Catholic 0
Huguenots Reformed 20



Religion Test 2: Provinces Identical (reformed), -200 all

Test Major Catholic (BB 100 both)
Britanny Catholic 30
Huguenots Reformed 14



Religion Test 3: Provinces the religion of the nation, -200 all

Test Major Catholic (BB 100 both)
Britanny Catholic 11
Huguenots Reformed 33



Religion Test 4 (Control): Everything catholic, -200 all

Test Major Catholic (BB 100 both)
Britanny Catholic 10
Huguenots Catholic 10

Very interesting results...the AI targets countries that have provinces of different religions from the country's religion. And it prefers to DoW countries of different religions, but not to the same extent.

 

[Toio]

A possibility I hadn't thought of! Definately worth pursuing - the AI doesn't have any countries in its hit list and has only default ai, but it's possible that something else could be at work...even the geographic location, come to think of it. All the targets neighbour France, but it's possible the game sets different standards on 'France' vs 'Netherlands' - however, Havard's list of regions suggests that all of France and the Low Countries are in the same region ("Western Europe").

After testing these "areas of expansion" (AOE) with POR for over 60 games. I have concluded that they do affect DOW to a degree.

Firstly let me say that the only viable option you have in AOE is the France and low countries bracket (havard brackets), any thing higher ie "western europe" and anything lower ie individual provinces within these bracket does not work very or as well.

for reference. my tests have concluded that (in order of influence the DOW) ,
Nations within the Combat list is the main thing to have for DOW, then
Core, AOE, culture.
Religion, strength of neighbour, value of province, COT's etc do play a part but it is the AI files for these countries which will influence the AI to go in a certain direction.
Note: manufactory in provinces swings the percentages of the Value of province to sit under Core.

Since the introduction of the BPAI (I am not saying they are bad) , AGCEEP game has not achieved its previous standard in gameplay. I think the AI is confused with the multiple files it has. To resolve these problems and get a true reflection on what the AI is doing we need to consider either amending the BPAI files or removing them and then introducing them individually as they are rewritten.

THE testing you have done to this point is really not a true picture on how the AI thinks.

suggestion for testing.
Remove all BPAI files and run the game with the same parameters as previous posts. I assume the old files are still there.
Then let us know on the results.

Note: the current POR -BPAI files are so bad that in 10 games of the current version I have not have any African Provinces for POR being settle in the first 100years. I will reintroduce my files and then post for all to use.