The great official AI thread

[Careful Plum]

An observation about how the AI is managing its puppets.

What the AI is doing wrong: It imports only as much ressources from its puppets as it needs in steps of 8. I. e., ENG can import 80 rubber for 1 CIC, but it will only import 16 if it doesn't need more.
What the AI should be doing instead: It should import as much as it can get for the same price.

Not importing this reduces the potential of re-exportation and maybe makes this ressources available for potential enemies.

 

[bERt0r]

An observation about how the AI is managing its puppets.

What the AI is doing wrong: It imports only as much ressources from its puppets as it needs in steps of 8. I. e., ENG can import 80 rubber for 1 CIC, but it will only import 16 if it doesn't need more.
What the AI should be doing instead: It should import as much as it can get for the same price.

Not importing this reduces the potential of re-exportation and maybe makes this ressources available for potential enemies.

Thank you for pointing out this flaw in the trade system while at the same time complaining that the AI does not exploit it.

 

[browd]

Also, if you take 80 rubber for that one factory, your puppet's autonomy skyrockets quickly, requiring you either to build in your puppet more aggressively (diverting valuable construction turns from your own states; lend lease simply can't keep pace) or risk losing your puppet to colony and dominion status and ultimately independence.

 

[Careful Plum]

Also, if you take 80 rubber for that one factory, your puppet's autonomy skyrockets quickly, requiring you either to build in your puppet more aggressively (diverting valuable construction turns from your own states; lend lease simply can't keep pace) or risk losing your puppet to colony and dominion status and ultimately independence.

Oh, I didn't realize that the autonomy scales with how much you take! (As in, I thought it would only be by how much you GIVE - and that's actually bad, since as a player, you can only buy 80 or 0!)

Thank you for pointing out this flaw in the trade system while at the same time complaining that the AI does not exploit it.

What? How is this a flaw? It's meant to represent the overlord exploiting the puppet's resources without having to give compensation. And it then makes sense that the master can re-export the resources, too.

 

[bERt0r]

I meant the whole "buy ressourced you dont need" thing. It is especially nasty with the USA as they are able to give away free CIV factories they can't use themselves to other nations (especially france). However, this has nothing to do with AI so its off topic.

 

[Bambilambi]

So... Playing as Soviet
Patch 1.3.3
No AI mods, the only non-ironman mod is a impassable terrain mod for Africa
+ 1 strength on all AI mayors.

Bakground:
Its september 1941. The axis are doing well, Spain joined the allies and even Australia have fallen... by Germany. good work there SteeVolt.

No attack is forthcoming against the red army however.
So I Justify a war goal....and attack late oktober.

What the AI is doing wrong:
There are no german army stopping me, now on november 8th I am at Berlin.I have basically walked there with no stops. When I tag over to investigate they have 27!! divisions only defending with maybe 100 in Africa. In total Germany only have 142 divisions. One clue might be that they are transported to Africa via the English channel...


What the AI should be doing:
Keeping a Army vs Soviet.
Do not transport trough the English channel!
Do not move to many units to Africa!

...feels like we have regressed to a earlier patch.
Save game included (autosave from november 1)

 

[MadDjinn]

Playing 1.3.3/veteran far from Europe and Italy is the leading cause of Axis failure by 1940 (purely AI, no human interference).

In multiple games I've seen AI Italy dive right into the 'War with Greece' option as quickly as they can. It's never fast enough to prevent the Allies from jumping right into the war.

UK Balkan focus can usually be done about the same time, so those two collide and Italy starts WWII with France/UK shortly after Germany manages to get Austria. This ends a year or so later as Germany is still doing it's own focus trees and has to fight wars when it doesn't have the advantage. Germany usually now gets stuffed at the belgian/dutch borders rather than running through them. Italy makes it into France just to get landed on in Rome/etc so it then capitulates. The soviets wipe the germans when they get the chance.

What would be nice, if a Major civ is going to join a faction and not be faction leader, is the AI looks to see if WWII is going to break out for utilizing a CB (preferably doesn't waste the time to research the Focus either). Once WWII is going, then the AI should just do what it wants.

 

[Axe99]

Here's something I've noticed a couple of times now. I can't be sure that what I think's going on is going on, but this post is structured around the suspicion. Steelvolt - if you know it doesn't work this way, please ignore and roll on . The context is a player-driven GER invasion of ENG in early 1945.

• • What the AI is doing wrong

What seems to happen is the AI strongly (very strongly) prioritises the first port invaded to defend, and rushes troops there. In this particular instance, it was Dover, and it turned out that my secondary invasion (of Southampton, which took a few more days to arrive) meant the port was quickly surrounded and cut off. Even after being cut off, the AI continued to ship troops to Dover, such that it had 60-odd divisions there at one point (mostly ENG, some USA). However, my secondary invasion found Southampton completely undefended.​

• • What you think it should be doing instead.

Even if half the country has been overrun, I'd think it would be in the AI's interest to continue to at least lightly garrison all of its ports, to give time to reinforcements to assist in the defence of secondary and tertiary invasion sites.
​

It's worth noting that if AI ENG/USA had successfully garrisoned its ports, it's likely that my invasion of ENG would have failed (I'd underestimated their defending force substantially) and 40-odd marine divisions would have been cut off and destroyed (which is probably how it should be during an invasion of ENG at that point in time, given how strong the defenders were ).

 

[Meglok]

So... Playing as Soviet
Patch 1.3.3
No AI mods, the only non-ironman mod is a impassable terrain mod for Africa
+ 1 strength on all AI mayors.

Bakground:
Its september 1941. The axis are doing well, Spain joined the allies and even Australia have fallen... by Germany. good work there SteeVolt.

No attack is forthcoming against the red army however.
So I Justify a war goal....and attack late oktober.

What the AI is doing wrong:
There are no german army stopping me, now on november 8th I am at Berlin.I have basically walked there with no stops. When I tag over to investigate they have 27!! divisions only defending with maybe 100 in Africa. In total Germany only have 142 divisions. One clue might be that they are transported to Africa via the English channel...


What the AI should be doing:
Keeping a Army vs Soviet.
Do not transport trough the English channel!
Do not move to many units to Africa!

...feels like we have regressed to a earlier patch.
Save game included (autosave from november 1)

please send this as a bug report. That way it wil not be missed. The Africa uber Alles mentality of the ai has got to be addressed by a hot patch, amongst several other glaring issue of 1.3.3.

 

[hillcf]

please send this as a bug report. That way it wil not be missed. The Africa uber Alles mentality of the ai has got to be addressed by a hot patch, amongst several other glaring issue of 1.3.3.

Even worse is watching both the ENG and GER divisions being shipped around Africa, alongside each other, to India, where Germany has managed to occupy Calcutta and is moving towards Delhi. And Germany now owns Malaya thanks to Siam and Japan.

 

[Dan1109]

• • What the AI is doing wrong

in regards to Axe99's comment above concerning invasions, I'd like to point out that that AI in the pacific atleast, has horrible invasion attempts. "Successful" invasions by the AI are actually occurring, but if you look at the details of why, because of troop shuffle, ports are left undefended and typically that's the only time the AI succeeds. This is the norm for historical invasions of Norway, Singapore, Indonesia, and elsewhere in the Pacific which are LIGHTLY defended.

• • What you think it should be doing instead.

Troops assigned for an invasion should be LOCKED DOWN, unless there is a catastrophic failure elsewhere. Tired of seeing an army group of 8 divisions sitting at a port and when the invasion finally occurs, only 1-2 divisions go on the trip. Further, I have seen the AI send CALVARY on an invasion, and a Marine division sitting at the same port stays at home. Unacceptable. 1 defending division can easily repel any invasion the AI sends out. Naval forces should always be assigned to the enemy seazone to assist with shorebombardment bonuses. If the AI nation has carriers, yes, carriers should be there too to a) protect incoming troops b) assign airpower to the invasion zone.

The goal I'd like to see, when giving Japan buffs (to get them strong to fight the USA), is to have them conquer Malaysia, Borneo/Indonesia by the time the US comes into the war. This is a major requirement for a good Island campaign to eventually come to the game. There are SO MANY vital resources Japan can take from the allies in Southeast Asia, who have only 1-3 divisions per state...but because of this horrendous invasion implementation, never happens, The new IC requirements of 1.3.3 make resources even more vital, and taking Allied Rubber away from them with put a serious dent in their airpower.

 

[Saarb97]

I was playing in the recent 1.33 patch with a friend with historical focus on (UK and JAP players and no mods, Japan wasn't in the Axis)
What the AI is doing wrong:
The German AI declared war on Belgium, the Netherlands, Luxembourg and France, didn't, got quite stuck on the border just so it can declare war on russia just a few days later, losing its eastern border quickly because he hadn't anywhere near as much forces on the border to even hold the soviets. needless to say germany got capitulated after a couple of months.
What I think the it should be doing instead:
Not trying to invade Russia when it is not even winning on the western front, maybe conditioning German declaring war on USSR only after the capitulation of France, in case germany is in war with France and yet conquered it. Or making the AI understand that it is not wise to start war on two fronts versus strong nations or starting a war against a strong nation when it is is sort of active combat versus another strong nation

 

[billcorr]

What I think the it should be doing instead:
Not trying to invade Russia when it is not even winning on the western front, maybe conditioning German declaring war on USSR only after the capitulation of France, in case germany is in war with France and yet conquered it. Or making the AI understand that it is not wise to start war on two fronts versus strong nations or starting a war against a strong nation when it is is sort of active combat versus another strong nation

Hmm...the experience of AI Germany fighting a losing two-front war seems to go against the "win a war before starting another" philosophy of patch 1.3.3.

Or as the patch notes for 1.3.3 maternally states:

"- Eat your vegetables before you get dessert: AI will no longer be declaring wars or justifying claims while in a losing war. "

 

[Meglok]

Also, if you take 80 rubber for that one factory, your puppet's autonomy skyrockets quickly, requiring you either to build in your puppet more aggressively (diverting valuable construction turns from your own states; lend lease simply can't keep pace) or risk losing your puppet to colony and dominion status and ultimately independence.

A player needs to be given a choice on how much to import due to the autonomy hit, another hot fix issue.

Hmm...the experience of AI Germany fighting a losing two-front war seems to go against the "win a war before starting another" philosophy of patch 1.3.3.

Or as the patch notes for 1.3.3 maternally states:

"- Eat your vegetables before you get dessert: AI will no longer be declaring wars or justifying claims while in a losing war. "

Yep, this is a definite "back to the hot fix drawing board" on this issue. Focus choices are completely overriding this change and any diplomacy code. The ai goes straight to DoW as soon as a focus is chosen.

 

[Gort11]

Hmm...the experience of AI Germany fighting a losing two-front war seems to go against the "win a war before starting another" philosophy of patch 1.3.3.

Or as the patch notes for 1.3.3 maternally states:

"- Eat your vegetables before you get dessert: AI will no longer be declaring wars or justifying claims while in a losing war. "

It's tricky with Germany, because you have to balance the logic of "Take down one opponent before you start on another" with the historical declarations of war against the Netherlands, Belgium, Denmark, Norway, Yugoslavia, the USSR and the USA all the while being at war with at least one other major power.

That said, there should almost certainly be "gateways" for the German AI (at least in historical mode) where they don't start the next war until they've dealt with the previous one, like this:

• Don't declare war on Belgium and the Netherlands until Poland and Denmark have fallen (Norway should probably be kind of independent from this process since I could see the war there going the Allies way, and the attack on France shouldn't be stymied indefinitely because of that)
• Don't declare war on Yugoslavia or Greece until France has fallen
• Don't attack the USSR until Yugoslavia and Greece have fallen

Doing it like this would minimise the number of fronts Germany would have to deal with, which should make things easier on the AI, which is often the best way to "improve" AI.

 

[Meglok]

That said, there should almost certainly be "gateways" for the German AI (at least in historical mode) where they don't start the next war until they've dealt with the previous one, like this:

Don't declare war on Belgium and the Netherlands until Poland and Denmark have fallen (Norway should probably be kind of independent from this process since I could see the war there going the Allies way, and the attack on France shouldn't be stymied indefinitely because of that)

The ai now will declare war on LUX 28 days, HOL, 29 days, and BEL 30 days after it fires Danzig or War, ie the time it takes to finish Around Maginot focus.

Don't declare war on Yugoslavia or Greece until France has fallen

Have not seen this in 1.3.3, usually because France is gone by then due to it's now complete inability to leave forces in the Maginot Line forts. The unit shuffle is horrendous.

Don't attack the USSR until Yugoslavia and Greece have fallen

Have often seen Italy Dow Greece with no troops set up because focus is completed. Germany will completely ignore Greece as will the Axis minors and Italy loses Albania. I have even seen a Greece invade Suez to capture it back from Italy who captured it and then left it completely ungarrisoned while it sent 30+ divisions to fight in the Congo.

All of your suggestions stem from focus completion overriding all other considerations, and until this is fixed the ai will continue to function in an incompetent and irrational manner.

 

[FrancescoT]

@instago post removed as for https://forum.paradoxplaza.com/foru...es-read-before-you-post.875352/#post-22481869

 

[Gort11]

All of your suggestions stem from focus completion overriding all other considerations, and until this is fixed the ai will continue to function in an incompetent and irrational manner.

Yeah, so I'm suggesting an additional check of "Have I defeated my previous set of opponents yet?" that takes place before the "Have I got a justification for war?" check.

Naturally this would just be for the AI, human players can make their own choices.

 

[Premu]

I was playing with the 1.3.3 patch as Germany and started an invasion of Canada. I created a small, well defendable bridgehead which soon was surrounded by troops of the USA which had a large amount of divisions ready.

What was the AI doing wrong?

It deployed a huge amount of divisions right at the front and started constant attacks crossing a river, leading to some nasty penalties and a very low attack efficency. They had up to 60 divisions in reserve - so they soon got serious supply issues, which were even increased by the occasional nuke I deployed on them. The rear on the other hand was hardly guarded at all. with another naval invasion in Boston I could encircle and destroy a lot of divisions.

What I think the AI should do instead?

The AI should not commit more troops to a front than it can practically use. Should the AI have some spare troops ready, it should use them for alternative purposes, in this priority.
1. Create a second defense line to prevent breakthroughs and encirclements.
2. Keep enough reserves to potentially counter invasions in other areas.
3. If possible start an invasion in the rear of an invading bridgehead, lengthening the front and by this being able to actually exploit the numerical superiority.

 

[Rider_of_Doom]

What the AI is doing wrong:

In my last game something unspeakable happened: Italy conquered the British Isles... per naval invasion. The British Empire had nearly zero troops defending their home, because they were busy conquering Africa. I never saw something like this before. Also the British Navy wasn't defending the sea tiles around their home.

http://images.akamai.steamuserconte...028/2ED6752419DD975738622D31E581F60760334CE0/

The moment, when I saw this headline... unbelieveable.

What you think it should be doing instead:

GB shouldn't throw everything over to Africa (or somewhere else) including fleets. The main focus for the British AI should be the defense of their home, if the Lowlands and/or France are defeated. Normally, the AI can handle this at least with the navy, but sometimes... it seems still off.