Victoria 3 - Dev Diary #59 - AI

[Wizzington]

That's simply the best diary I have seen so far - beautiful immersive world!

My question would be, can we have some clarification on how AI calculates the cost of war?
- on the streams we have seen Great Britain lose over 300k Pop's in war with China; can we hope for a mechanism that AI actually realises how much it is going to lose so much people that the wargoal is not really worth it?
- another thing: when AI is passing a law that can potentially trigger a civil war, is it going to calculate that it night be too risky with reward being just passing non hereditary beaurocracy?
- is boldness described in the diary as property attached to Great Powers a factor that is going to push Great Powers to engage in more bloodbath wars for signuficantly smaller benefit? Are there aby mechanisms to mitigate this behaviour

Again I would like to say: many thanks for this juicy, transparent and informative diary!

1) The AI does have some consideration here for how dangerous a war will be for it but I will say it probably needs a bit more work on actually comparing gains vs potential losses. May have time to do it for release, otherwise for sure an area where it'll be polished more post-released.
2) The AI does consider how much it *likes* the law it's passing compared to reluctance for civil war, but I don't think it should make further considerations than that - that would be too much meta-gaming for me.
3) The boldness factor from rank is relatively weak and is more meant to be something like 'a great power is more likely to throw the dice on something like WW1 because they *could* come out on top while a small power knows it's going to suffer horribly' rather than 'great powers always escalate to war'

 

[Wizzington]

In a diplomatic play, is the amount of sympathy a country receives dependent on their war goal? For example, would a country that demands the return of a part of its homeland be afforded more sympathy by outside actors than one that demands land purely on the basis of might-makes-right naked aggression? In a similar vain, are countries with claims to the lands they demand afforded more sympathy? Or are countries more sympathetic towards diplomatic actors trying to free a people from a discriminatory overlord, especially those countries whose primary cultures share a heritage or two with the people of that potential country?

Indirectly yes, in the sense that the amount of Infamy they took from the wargoal plays a role and 'rightful' wargoals like returning lost land results in much less infamy.

 

[Wizzington]

Im suprised that there is not a single word about ideological alingement affecting AI sympathy and decisions. Do I understand correcly that absolute monarchy has no higher chance of supporting fellow conservative regime for example against liberal revolution?


Of course it wasn't uncommon, especially with years passing, that monarchies were cooperating with revolutionary forces. In the end Russian absolute monarchy ended in alliance with French Republic and liberal Britain after all.

But I think that one of possible AI strategies should definitely be fight against revolution or the opposite - supporting liberal changes abroad.

At the beginning of Victoria The Holy Alliance was still sort of alive. Well at least Russia treated it seriously and was called "the gendarme of Europe". Russia wanted to march and crush French and Belgian revolutions but was stopped by November Uprising in Poland. Later Russians intervened and helped Habsburgs fight Hungarian revolution.

After Russia was betrayed in Crimean war by ungrateful Austria, their ideological zeal and readiness to support conservative regimes abroad was weakened. That could be in game terms - strategy change from crushing the revolutions to more pragmatic stance.

Austria also was focused on fighting revolution until Metternich fell in 1848. After that Austria changed their strategy.

This is definitely a thing but for some reason I left it out of the diary! The AI has something called 'Ideological Opinion' which compares the ideology of its government to the ideology of the other country and which plays into diplomacy, strategy, who it's likely to sympathize with in diplomatic plays (especially revolutions) and so on. So it's a very important thing, I just forgot to write about it!

 

[Wizzington]

Any chance you could explain why this is? I certainly believe you - I’m not at all a programmer - this just seems like an interesting limitation I wouldn’t have expected. Does it just have to be formatted in an inefficient way to be accessible to modification?

In a word: Yes. To expose the core loop we would need to make the core loop run through a script language which would be insanely performance-inefficient. HOI3 tried this and it resulted in an AI that had to be 100% railroaded for the game to work at all.

 

[Wizzington]

There are reasons why games give bonuses to IA to increase difficulty.
An experienced player will always outplay an IA in a fair competition.
It doesn't matter how aggressive or how biased against the player you program your IA, a human will always perform better.
Refusing to give bonus to the IA can only result in an easy and unchallenging game.
Very sad that Victoria 3 is going to that direction.

It's very easy to add an AI bonuses game rule as a mod if that's something you want, but we really don't think it fits our vision for the base game.

 

[Wizzington]

Given price fluctuations, what is really preventing AI from entering the unbroken cycle of building / degrading buildings?

For one, the AI doesn't downsize production buildings - there's really no reason for it to do so and if it needs to prop up a building it can use subsidies. When it comes to government buildings, they're not going to downsize a gov. admin just because paper got more expensive - the AI will only downsize government buildings if it believes that there is no other sensible way to balance its budget because it doesn't want to throw away these invested resources.

 

[Wizzington]

Shouldn’t that be 50 + 30 = 80 as the threshold for Prussia?

Nope. It doesn't need to beat the combined scores, just whichever is greater.

 

[Wizzington]

How impactful will this opinion be? Will it be anything close to the Reaction triggered modifier in EU4?

It can range from just a little bit impactful to very impactful, it depends on how large the ideological gulf is and also on the AI's strategy - an AI that's preserving the status quo cares more about it while one that's playing in a more 'machiavellan' way cares less. It also varies with what kind of action it's influencing - for example, the AI cares more about ideology when signing alliances than when signing trade agreements.

 

[Wizzington]

If it beats the score diff but not neutrality threshold: "Yeah, I do want to help you.... but not enough to actually do it"
If it beats the neutrality threshold and not the score diff: "Yeah, I do care... but not in your favor enough sadly"
two different criteria, no need to add them up

Exactly this

 

[Wizzington]

I only have one question really, I know we the player can ask "will you help for an obligation?" in our plays, but is there a way for us to say, "I'll help you for an obligation." In their plays?

From what we've seen all you can do is pledge to one side seemingly for free. Is there a way to ask, or are we at the mercy of the AI overcoming it's social anxiety to ask us.

Unfortunately this will not be in for release, but 'reverse-swaying' is absolutely something I want to get into the game sooner rather than later, as well as just adding more things to offer and be offered in Diplomatic Plays in general.

 

[Wizzington]

When a new building is modded in, will the Spending Variables be updated to keep the AI decisionmaking sane?

Like if I added a new more efficient version of a maize farm and called it a potato farm, would the AI ever build potato farms?

Do Spending Variables need to be manually modded in for each new building created?

Spending Variables are dynamically calculated from building properties, not separately scripted - they should work seamlessly with new buildings, production methods and goods without extra effort.

 

[Wizzington]

All in all this is one of the most interesting reads I’ve had for a long time.

My only real concern on AI
( granted I’m sure there will be bits to adjust on release) is the extent with which a state may commit to foreign wars.
is it something that might be in the pipeline that a European state shouldn’t offer to bring its entire military to bare IF in far off parts of the world it has been persuaded to join a play, or is this already in there somehow?

The AI does have a target mobilization level for each play/war depending on their level of involvement, expected opposition, enemy mobilization and so on. The AI doesn’t just 100% mobilize France's army to help Sardinia against Tuscany or conquer a weak country in Indonesia - they will try to raise a proportional force to avoid unnecessary costs and attrition.

 

[Wizzington]

> Well, just as in our other Grand Strategy Games, the AI plays the same game you do.

This is a flat out lie. The AI in EU4 cheats. It does. It can see through Fog Of War.

The "no cheats" part specifically refers to V3 but the AI in EU4 does play the same game you do - it doesn’t just fake an economy or ignore restrictions that bind the player. Last time I worked with EU4 it also respected fog of war, at least when it came to enemy armies.

 

[Wizzington]

How much interaction is there between the AI Strategies for a country? Will the Administrative AI understand it shouldn't prioritize advanced productions if the Political AI has decided to suppress the relevant IGs? If the Political AI is barely keeping a lid on French separatism will the Diplomatic AI avoid conquering more French territory?

They don’t really coordinate much on purpose - the AI deciding to industrialize while cracking down on the liberals should create some interesting (and historical) conflicts for example.

 

[Wizzington]

Would a historical only game rule be possible? Similar to hoi4?

HoI4 covers a few years of a single (if massively important) war. V3 covers one of the most transformational centuries in history. It’s not really feasible, no.

 

[Wizzington]

Does this account for Pops?
If AI conquers a state with only unqualified Pops, will the building still keep "Must-have" status?

It's not a property on the building, the new owner would calculate its own priority for it.

 

[Wizzington]

UI question: Is there a way to see in-game what you explained in the caption, namely that expanding the farm will reduce profits because of the price impact from increased supply? In the screencap, I can see the first half, that profits will go down if the farm is expanded, but nothing indicating why.

It's showing why right there in the screenshot - the added wages cost more than the added sale of goods bring in under the new market conditions.

 

[Wizzington]

How would this revision and the said principles address the problem of UK taking an inland isolated province of Qing and Prussia annex Denmark mainland instead of Holstein?

That was just a bug with a script weight being inverted, not really relevant to the principles here asides from the fact that yeah, doing so isn't behaving very plausibly and thus is clearly a bug.

 

[Wizzington]

I am guessing, from the picture of Confidence of Austria against Prussia, that "Level of mobilisation" includes number of fleets. (I might be completely wrong here, is always hard to understand everything from a static image)

My understanding is that fleets are always mobilised, so this gives an (unfair?) advantage during diplomatic plays to nations with a strong navy even when the war won't really involve naval warfare. Am I reading this correctly? Is it WAI?

No, it only counts mobilized armies - in this case it's only comparing potential strength since nothing is mobilized, but the moment mobilization starts the side that mobilizes faster is going to appear stronger than they previously did since then both potential and mobilized strength is taken into account.