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

[DSMyers1]

How many troops did the DOWing AI have? Others have already run troop tests and it was found out that they do effect DOWs. However, I did all along argue that even though they may effect DOWs, they do infact not effect it as much as economic strength, hitlists, and probably relations as well. It is also possible that the AI is weighing troop discrepency as opposed to it simply giving permanent values to a given a mount of troops.

The Attacking country has 50,000 troops...hmmm, that's still more than the target. I'll give the target a huge preponderance of force to get some conclusive results.

All this is pretty solid evidence that the AI is weighing mutliple economic factors cumulatively. Most likely it is literally weighing monthly income. You should try making an event for one of the AIs that gives it a big cash bonus day one and see if the DOWing AI will then pick the non-event recieving AI instead.

Do cash bonuses count as monthly income? I don't think they do...I don't know any good way to change strictly monthly income, but I agree with you--I am fairly sure monthly income is what the AI is weighing.

Yes it is. And this is solid evidence that an AI can be screwed over by hitlists as well. Like I said in my guide, they can be very useful but very crippling as well and really need playtesting to get right, as a historic target may ne too powerful or often in powerful alliances, and thus essentially cause the AI with him on his hitlist to commit suicide.

Yep. In the AI's I have written, hitlists are avoided unless they are for a major, and the targets are minors.

I'd be interested to know how cores fair vs hitlists. I put my money on the hitlist over the core. Also cores vs temporary CBs as well as core vs no CB at all and no CB or core vs only temporary CB.

I'll do that test soon.

It would also be nice to weigh BB. It would need to be below 25, as WISK has shown that that is a threshold for AI BB. You could try various levels of BB vs no BB at all and then use the Hitlist to speed up the DOWing too.

Yeah. I wonder if the threshold changes when I set it on very hard/furious. I wonder if there are simple threshold values, or if the AI looks at the actual value, and would DoW more a 100 BB than a 70 BB.

I'd also like to see if navies have any effect. I'd guess you would need a very large navy vs no navy. And don't give up on the troops either. Make larger discrepencies and see how it effects DOWing.

Right.

Great test results BTW. Thanks for sharing them. I think I'll update my AI guide with the new info when this experiment is all said and done.

Thanks for your encouragement! It's always nice to get feedback. It's funny--I run experiments at a research lab for work...I essentially do the same type of thing...with concrete instead of AI.

 

[DSMyers1]

Here is the Core vs. Hitlist:

Test 14: Core vs. Hitlist
Brittany (Core) .......... 0
Huguenots (Hitlist) .......... 10

Wow! The attacker didn't even have a casus beli on the Huguenots. Remember both have BB 100, so the AI really wants to attack both of them. I am very surprised that the AI favored the hitlist so heavily over cores.

And conclusive results for troops: the AI does not take them into account:

Test 16: Troops vs. Nothing
Attacker has 50,000 troops
Brittany (0) .......... 17
Huguenots (120,000) .......... 20

 

[Norrefeldt]

Oh, I had alwys thought that the cores was the single most important thing, and a lot of energies have gone into modding it. Very interesting mr Myers. I look forward to the tests with temporary CB.

 

[unmerged(31425)]

Wow! The attacker didn't even have a casus beli on the Huguenots. Remember both have BB 100, so the AI really wants to attack both of them. I am very surprised that the AI favored the hitlist so heavily over cores.

I'm not. Cores are very useful to give to AIs to get historic results though, as they can then demand them in wars, even if they don't occupy them. They can also be offered, even if not occupied too. The only really bad part is that cores reduce relations.

And conclusive results for troops: the AI does not take them into account:

Test 16: Troops vs. Nothing
Attacker has 50,000 troops
Brittany (0) .......... 17
Huguenots (120,000) .......... 20

Well, I guess that settles that.

 

[unmerged(17856)]

DSMyers1, in your religion test 1, when you wrote 'Provinces Identicle (catholic)', you are referring to the province culture of all provinces?

Then when you wrote, for example, 'Test Major Catholic' etc. you're referring to the state culture of the countries?

There seems to be a slight tendency for the AI to attack countries with tags close to its own, but given that it only did it 4% more of the times, and given that it tends to repeat itself in behavior it can not be concluded, that this is indeed the case with any kind of certainty.

According to your experimental results, the difference between the tags was 12% not, 4%, if I understood your experiment correctly. Your experiment essentially describes an appreciable effect of tag on DoW behaviour. I will run stats on your religion tests when I understand the methodology and conditions correctly

Very nice work to everyone who has contributed so far. Extremely useful and insightful!

 

[unmerged(16460)]

According to your experimental results, the difference between the tags was 12% not, 4%, if I understood your experiment correctly. Your experiment essentially describes an appreciable effect of tag on DoW behaviour. I will run stats on your religion tests when I understand the methodology and conditions correctly

No, not exactly, it does two things. The main object was to see if the AI took any account of culture. One of victims controlled a province of the same culture as the agressor, while the other countries had a province diferent to the agressor. I made the two victims exactly the same in everything except culture and tags (for obvious reasons).
When I had run 25 tests I swithced tags to se if the had anything to do with it. I did this because I had noticed a tendency for the randomizer to chose low province id numbers when randomly relocating a capital, and wanted to eliminate the posibility of this occouring in regards to choosing tags when declaring war.
As my results showed, in 25 out of 50 cases the ai went for provinces with its own province culture, when confronted with two options. This would suggest to me that it does not take culture into account when declaring war.
I wrote down the two sets of tests, and noticed that out of the 50 tests the AI chose to declare war on the lowest id 28 times. But this was not the original object of the test, as much as it was a bi-product.
As to my calculation, you are right that the diference in what tag was chosen was not 4%, but 12%. Still this is within the error-margin one could expect, and though it might suggest that tags does play a part, it would require more test, with tags wider apart!

 

[unmerged(17856)]

I ran a within-'test' t-test and the difference fails to reach significance with alpha set at 0.05, but it's statistically different with an alpha set at 0.1

I'll try to explore better statistical options tomorrow while at work.

 

[unmerged(16460)]

I ran a within-'test' t-test and the difference fails to reach significance with alpha set at 0.05, but it's statistically different with an alpha set at 0.1

I'll try to explore better statistical options tomorrow while at work.

What is the within-'test' t-test?

 

[Norrefeldt]

Ribbon22, since you name drop some statistical terms: what is needed in tests like this to be able to say that something is statistically true with a certainty of 90-95%?

 

[DSMyers1]

DSMyers1, in your religion test 1, when you wrote 'Provinces Identicle (catholic)', you are referring to the province culture of all provinces?

Then when you wrote, for example, 'Test Major Catholic' etc. you're referring to the state culture of the countries?

Very nice work to everyone who has contributed so far. Extremely useful and insightful!

Province religions were all Catholic in the first test. Test Major refers to the state religion of the large country that is declaring war--in this case it is catholic. It is essentially France, with a different tag. Then I listed the State Religion of each of the two minors that are targets. I am never refering to culture in tests 1-4, I am referring to religion. The culture tests are tests 8 and 9. Does this clear everything up?

 

[unmerged(17856)]

What is the within-'test' t-test?

what you were labelling as 'tests' (i.e. 'test 1' and 'test 2') I'd have called groups, but it's merely a matter of convention. A t-test basically tests for differences between the means of two data sets (for a defined significance value, alpha, which is usually set to 0.05 and essentially forms a 95% confidence criteria) and will depend upon the variance within each data set. You can run between groups and within group t-tests, for example.

what is needed in tests like this to be able to say that something is statistically true with a certainty of 90-95%?

always an equation Sometimes a table like the t-value distributions. The equation will depend on what kind of data you have, the nature of the measurements, and what you want to do with the data itself! For the t-test, you need to have calculated the mean, variance, standard deviation, degrees of freedom, and your confidence criterion.

After further thought, I think I should have run a chi-squared test...and it fails to reach significance (alpha < 0.05).

 

[unmerged(17856)]

Province religions were all Catholic in the first test. Test Major refers to the state religion of the large country that is declaring war--in this case it is catholic.

If I understood correctly, in test 1, the provinces are all catholic and the only non-catholic state religion is reformed which was assigned to the Huguenots?

If this is so, how did you make sure that Huguenots did not switch to catholic state religion before DoW occurred, or province conversion to reformed? This general issue potentially arises in test 2 as well. Did the speed at which DoWs occurred prevent such possibilities, or at least minimize this adequately, in your opinion?

I am never refering to culture in tests 1-4, I am referring to religion. The culture tests are tests 8 and 9. Does this clear everything up?

I don't know why I wrote 'province culture', I meant province religion. o I wanted to clarify the distinctions between province religion and state religion assignments within the tests.

 

[DSMyers1]

If I understood correctly, in test 1, the provinces are all catholic and the only non-catholic state religion is reformed which was assigned to the Huguenots?

If this is so, how did you make sure that Huguenots did not switch to catholic state religion before DoW occurred, or province conversion to reformed? This general issue potentially arises in test 2 as well. Did the speed at which DoWs occurred prevent such possibilities, or at least minimize this adequately, in your opinion?

I don't know why I wrote 'province culture', I meant province religion. o I wanted to clarify the distinctions between province religion and state religion assignments within the tests.

Due to the fact that there is 100 BB for each minor, the Huguenots and Britanny, the DoW occurs within 7 days after starting. Always. Hence the number of tests that I have been able to do.

 

[unmerged(2456)]

I'm not. Cores are very useful to give to AIs to get historic results though, as they can then demand them in wars, even if they don't occupy them. They can also be offered, even if not occupied too. The only really bad part is that cores reduce relations.

Yea, but they are given out via the same events to humans as freebies basically when all they are meant to do is guide the ai historically. I'm not talking here about historically justifiable ones, i'm talking about adding cores to say the OE to make it expand in a given way by the ai.

 

[unmerged(31425)]

Yea, but they are given out via the same events to humans as freebies basically when all they are meant to do is guide the ai historically. I'm not talking here about historically justifiable ones, i'm talking about adding cores to say the OE to make it expand in a given way by the ai.

Well doing that definitely helps. Daywalker does it in his mod too. Is has a sizeable effect. But cores for just AIs would be AI cheating, so not too sure you can capitalize on it in the AGCEEP.

 

[unmerged(2456)]

Well doing that definitely helps. Daywalker does it in his mod too. Is has a sizeable effect. But cores for just AIs would be AI cheating, so not too sure you can capitalize on it in the AGCEEP.

Well the irony is imo that because of the way they are handed out now, they are more of human cheats than ai cheats in AGCEEP for places like Russia and OE in many cases.