Devs, for 1.31, please sit down and fix the AI

[starwarsbv]

Once the summer break is over, please consider doing a big patch to work on the AI. The AI is unanimously agreed to be inadequate. We really don't need more provinces, tags, missions, or events when the AI performs as poorly as it does.

Some of the biggest glaring AI issues off of the top of my head:

-debt: AI countries (especially majors) try to be at 100+% forcelimit all the time (sometimes with expensive mercenaries) which puts them in massive debt. It's not uncommon to see Ottomans, England, Russia, France, Spain/Castile being in 3k ducats in debt before 1500. It's the worst with AI Russia, which spams streltsy every chance it gets.

-army composition: AI doesn't know how to design armies. Russia still has only infantry armies. Ottoman armies are mostly artillery. AI either under utilizes or over utilizes cavalry. AI doesn't know how big or small to make its armies. I've seen AI nations create doomstacks of 50k soldiers, which it then marches through the sands of Arabia.

-ai behavior: one of the biggest issues is that the AI doesn't know how fight wars. when the AI is in a defensive war, I often see their armies go on a hike, getting as much mil access as possible, to go around the enemy armies and carpet siege their enemy. In doing this, the soldiers of course abandon their land, families, and homes to occupy some backwater in the Balkans. Here is an example. Qara Qoyunlu is getting invaded by Mamluks, Ottomans, Aq Qoyunlu, and Great Horde. Instead of defending its homeland, the black sheep turkomans go and carpet siege the deserts of kazakhstan. This is especially painful because the AI could have easily stackwiped my armies if they were actually trying to put up a fight. But for some reason, the AI thinks annoying the enemy is a good strategy.

There are of course many more issues with the AI that need to be addressed.

Devs, when you're back from your well-earned summer vacation, consider fixing some of the AI problems in 1.31.

 

[Canute VII]

1.30.4 please

 

[starwarsbv]

1.30.4 please

That would be ideal but the AI issues are likely too big for a 1.30.4 patch fix

 

[Orioniys]

As the guy who mods the AI, I can say this update actually fixed some major AI deficiencies, bugs & and added new settings & new AI logic. It's in bit of wip state right now, but it's major step forward.

I think I solved most of the issues you mentioned in my mod.

Debt spiral, I think it's combination of new mercs and new AI loan settings. Mercs need urgent fixing.
Army composition, is strongly influenced by AI's income.
AI behavior, can be partially achieved with existing settings.

 

[Froonk]

As the guy who mods the AI, I can say this update actually fixed some major AI deficiencies, bugs & and added new settings & new AI logic. It's in bit of wip state right now, but it's major step forward.

I think I solved most of the issues you mentioned in my mod.

Debt spiral, I think it's combination of new mercs and new AI loan settings. Mercs need urgent fixing.
Army composition, is strongly influenced by AI's income.
AI behavior, can be partially achieved with existing settings.

If possible, could you check AI behaviour for building non-income buildings, especially coastal defences and manpower/sailor manufactories? I feel like there is definitely something off about how hard AI prioritises these.

 

[Orioniys]

If possible, could you check AI behaviour for building non-income buildings, especially coastal defences and manpower/sailor manufactories? I feel like there is definitely something off about how hard AI prioritises these.

There is hardcoded layer and general Ai will do values that can be modded. You can generally test hardcoded values in the game by simply giving AI cash and tech. Just from experience I can say that force limit modifiers have very high hardcoded value.
In addition to that building.txt have left over code that multiples(or adds to) this very high hard coded value.
Example from shipyard. Regiment camp lineup have the the same problem.

ai_will_do = { factor = 1 modifier = { factor = 105 # Little above build cost FROM = { navy_size_percentage = 0.95 } }
When AI is above 95% of his force limit it starts working.
Force limit building are generally very good and often underrated buildings, but they are not that good.
I also think that new manpower manufacury is incredibly good, but vanilla AI prob builds it everywhere, possibly ignoring trade goods bonus.

 

[Kimbole]

As the guy who mods the AI, I can say this update actually fixed some major AI deficiencies, bugs & and added new settings & new AI logic. It's in bit of wip state right now, but it's major step forward.

Maybe so but there’s still some interesting stuff going on. In a war with a huge France on my side vs GB, Britain managed to peace out France because France kept walking small cannon stacks into a moderately sized stationary British army. With about 50 k French troops standing somewhere nearby doing nothing.

I also have four big subjects for a change and watching how they move their troops around when at war has been revealing. The AI never seems to group up to attack a smaller enemy? Even if it would relieve a key siege. Ideally if there are three or four armies around that collectively outnumber the enemy (comfortably), they would be a bit better at forming up and fighting that battle. Often what happens in my experience is the individual countries will wander about until they get picked off. They will all pile into a battle if a small stack gets caught, which is nice, but the AI does like to stray too far from allies to be helped in time.

Not saying it is easy of course. But I guess some sort of desire to concentrate forces when the war is evenly balanced numberswise might help.

 

[Orioniys]

Maybe so but there’s still some interesting stuff going on. In a war with a huge France on my side vs GB, Britain managed to peace out France because France kept walking small cannon stacks into a moderately sized stationary British army. With about 50 k French troops standing somewhere nearby doing nothing.

I also have four big subjects for a change and watching how they move their troops around when at war has been revealing. The AI never seems to group up to attack a smaller enemy? Even if it would relieve a key siege. Ideally if there are three or four armies around that collectively outnumber the enemy (comfortably), they would be a bit better at forming up and fighting that battle. Often what happens in my experience is the individual countries will wander about until they get picked off. They will all pile into a battle if a small stack gets caught, which is nice, but the AI does like to stray too far from allies to be helped in time.

Not saying it is easy of course. But I guess some sort of desire to concentrate forces when the war is evenly balanced numberswise might help.

50k stack standing somewhere nearby doing nothing. Sounds like current bug with mercs.

AI can see approximate strength of ur armies. You can stack combat bonuses, AI just picks this stuff somewhat randomly.
AI can concentrate his forces, for many reasons this rarely happens in vanilla.

The game changes quite significantly when AI is generally efficient at everything. When AI is going for every combat bonus, u kinda forced to mirror that. It brings larger questions about the state of ingame balance, the state of late game economy and performance.
On other hand there is the stuff like debase currency, base game AI will always debase at 0 corruption. 1.30 added a setting that can be used to turn it off, but it's on by default.
I honestly don't understand why it's like that, purely moneywise it's like loan but several times worse. If AI conquered a bunch of land, he will spend even more on rooting out сorruption.

 

[JaxElite]

Once the summer break is over, please consider doing a big patch to work on the AI. The AI is unanimously agreed to be inadequate. We really don't need more provinces, tags, missions, or events when the AI performs as poorly as it does.

Some of the biggest glaring AI issues off of the top of my head:

-debt: AI countries (especially majors) try to be at 100+% forcelimit all the time (sometimes with expensive mercenaries) which puts them in massive debt. It's not uncommon to see Ottomans, England, Russia, France, Spain/Castile being in 3k ducats in debt before 1500. It's the worst with AI Russia, which spams streltsy every chance it gets.

-army composition: AI doesn't know how to design armies. Russia still has only infantry armies. Ottoman armies are mostly artillery. AI either under utilizes or over utilizes cavalry. AI doesn't know how big or small to make its armies. I've seen AI nations create doomstacks of 50k soldiers, which it then marches through the sands of Arabia.

-ai behavior: one of the biggest issues is that the AI doesn't know how fight wars. when the AI is in a defensive war, I often see their armies go on a hike, getting as much mil access as possible, to go around the enemy armies and carpet siege their enemy. In doing this, the soldiers of course abandon their land, families, and homes to occupy some backwater in the Balkans. Here is an example. Qara Qoyunlu is getting invaded by Mamluks, Ottomans, Aq Qoyunlu, and Great Horde. Instead of defending its homeland, the black sheep turkomans go and carpet siege the deserts of kazakhstan. This is especially painful because the AI could have easily stackwiped my armies if they were actually trying to put up a fight. But for some reason, the AI thinks annoying the enemy is a good strategy.

There are of course many more issues with the AI that need to be addressed.

Devs, when you're back from your well-earned summer vacation, consider fixing some of the AI problems in 1.31.

There are multiple issues with updating the AI. I will lay out a few different issues and which you can expect to be improved of fixed and which will likely stay bad.
In general AI works like this:
If (conditions) do x.
To change the AI there are in general 3 ways to edit behaviour. You can adjust conditions (do the same stuff in different situations), adjust x (do something else than before in an old situation) or add more if statements (add new "tools' to the AI).
In case of changing the conditions there can be added new ones or the old ones could be changed.

So these are the tools the developers generally have. Now what happens in each case:
Changing old conditions: This is the easiest, most efficient and often one of the most effective methods. There are no additional computations needed, so there is no effect on runtime. It is basically tweaking numbers and can tackle many problems quite well, if not fix them. Expect for example the coastal defense spam and the debt spiral improved via this method (althgogh the debt spiral will likely need other changes as well)

Adding conditions: adding conditions makes the AI feel smarter in their decision. It is basically the difference in how kids and adults make life decisions and the AI feels more competent the more checks there are on a decision. But the cost of that is higher runtime. Each check means one more calculation per AI instance per tick (per day on most checks). This quickly adds up to a lot and is the main cause of the game slowing done in recent patches. It is a great act of balance to only use this when necessary without compromising AI too much. This will probably still be one of the major tools for the debt issue.

Change x:
This happens rarely and usually correlates with new gameplay (think that the new loan privilige from the burghers can be used instead of usual loans). It is unlikely that this will help with the current big AI issues. The good thing about it is that there are no additional calculations needed.

Add more if statements:
This is in most cases the worst choice. Since many effects check the same conditions it means a nonlinear increase in runtime. This hurts performance even more than just adding more conditions. The issue here is that it is easy. The programmer doesn't need to check the code and can just add his own stuff.
This is the reason why AI behavior in wars likely won't get fixed in the near future, because it woul mostly rely on more conditions and more ifs. Adding hundreds of calculations per day and severely hurting performance.

So in total we csn expext the debt spiral and building weighting to be fixed in the near future. While we will likely not see better army composition (lots of conditions) or much better "AI behaviour" (lots of ifs and conditions).

 

[Orioniys]

So in total we csn expext the debt spiral and building weighting to be fixed in the near future. While we will likely not see better army composition (lots of conditions) or much better "AI behaviour" (lots of ifs and conditions).

As I stated earlier debt spiral problem consist of two things Mercs and new AI settings for handling loans. Repaying loans, is not a absolute priority as it was in 1.29. In fact almost everything else have higher priority, buildings, upgrading trade centers etc. That's the core of the Ai's debt spiral, he accumulates interest from loans and fails to adjust his balance properly.
I've tried increasing Ai's scoring for repaying loans, and it works. Whatever the reason, the math is wrong or just the values are to low. So I think it mostly existing code that needs to be fixed.
As for Army composition, I don't see this problem in my mod, and I think there is a strong correlation with AI's income.

 

[JaxElite]

As I stated earlier debt spiral problem consist of two things Mercs and new AI settings for handling loans. Repaying loans, is not a absolute priority as it was in 1.29. In fact almost everything else have higher priority, buildings, upgrading trade centers etc. That's the core of the Ai's debt spiral, he accumulates interest from loans and fails to adjust his balance properly.
I've tried increasing Ai's scoring for repaying loans, and it works. Whatever the reason, the math is wrong or just the values are to low. So I think it mostly existing code that needs to be fixed.
As for Army composition, I don't see this problem in my mod, and I think there is a strong correlation with AI's income.

There certeinly are issues with army composition especially with a good economy. They AI starts building way too many cannons way too early.

Other than that what you describe as likely fixes is the same i mentioned, tweaking a condition

 

[Blk82]

Because of how battles work, looping around and attacking the enemy from behind is probably the best strategy the AI can pull of. Waiting around on your homeland because "Muh Last Stand" when you are outnumbered is basically giving your opponent warscore. The only reason QQ can potentially stack wipe your troops is because you know what the AI is doing, so you are not massing your troops in their territory. Also, QQ is not just fighting you, but also the Mamluks and the Ottomans.

Actually, the troop placement of QQ is not that bad. In an deluge situation, the best course of action is to focus on the easiest war. They are in a good position to drive out the Great Horde's troops than carpet siege it. (Obviously, since they are fighting a human, the AI might not be smart enough to execute such a strategy, but the groundwork is there.)

 

[Orioniys]

There certeinly are issues with army composition especially with a good economy. They AI starts building way too many cannons way too early.

Other than that what you describe as likely fixes is the same i mentioned, tweaking a condition

Army composition is a setting that already exist in the game as well.

ARTILLERY_FRACTION = 0.35, -- Ratio of artillery to infantry AI will build FORCE_COMPOSITION_CHANGE_TECH_LEVEL = 11, -- Tech level at which AI will double its artillery fraction

Vanilla value is just to high. 0.35 dubled at tech 11. Having full back row of cannons is not that crucial at tech 11. Tech 16 on other hand is a major cannon upgrade.
Almost all of the problems with AI are in existing code and can be solved. Performance is a big issue thought.

I believe Devs are pretty open about what is the major performance strain, it's trade system. You can experience it firsthand in my mod in a late game, because AI builds manufactory in every province.