The great official AI thread

[ChaChaLoco]

They do this even from the 1936 start, at least if you play Germany historically. AI France and UK will send the large bulk of their navy into the Baltic (even if you later invade Denmark and trap them - at least I think that traps them?), and then land troops all over the north German coast to get quickly encircled and captured. They even land in provinces without ports, as I always garrison my ports with 2 Garrison divisions (2x3Inf).

 

[Aodhan_]

I just move the invasion of Denmark up a few months and close it of to enemy ships, plus it helps with the invasion of Sweden now that you need their resources.

 

[Sanny]

I can confirm this

 

[Dalwin]

The same happened in my one time (recently) playing the '39 scenario. It was made both worse and better by the fact that I reinforced the Poland offensive before actually declaring war thereby leaving Rostock without a garrison when the Allies landed.

The reason this ended up being better for Germany was that with a captured port, the Allies reinforced their invasion. By the time I deployed reserves to the region and contained and eliminated it, they had sent 40 divisions to their deaths. These were units they would soon badly need for the defense of France.

 

[KiwiNoob]

The AI still cant draw up a front line that makes any sense or advance that line without all hell breaking loose.

You continue to focus on a bunch of fringe case issues when the basics are broken. You're cheering because you've fixed the squeaky hinge on the passenger side door when your car's tyres are all flat.

Step 1. Make sure the AI can assign troops evenly along a front line with terrain and opposition strength taken into account.

Having 8 divisions on one province and then one division on the other 8 = broken.

Step 2. Make sure that when attacking the AI keeps the line together and dispenses with all the bullcrab strategic re-deployments.

When a normal person advances a front they need the occasional re-deploy. Mostly it can be achieved quicker and without the org hit by shuffling a series of divisions one province over.

The AI having 50% of the divisions strat re-deploying at any one point = broken

Step 3. Fix whatever you want to fix - at least the game will be playable.


The inability to handle even simple front lines properly means that people HAVE to micro manage their troops if they want them to be even remotely efficient - even in places like Central Africa (snore). And even with a small nation it's way too easy to beat large AI nations as their inability to maintain a line makes encirclements pointlessly easy.

You may point to the number of hours I've played HOI4 as proof the game is fun. But you should know that the sequence of events goes like this:

Start HOI -> Choose small nation -> Buildup small nation into mighty warhorse -> WW2 starts -> See that AI still sucks -> Stop playing

If the game was 'Build up to war 4' then gg - the buildup is a lot of fun.


P.S. I read that you had some hiccups with 'if' statements being the wrong way round - another place to check would be where the AI decides which province to attack. At the moment it seems to favour attacking into mountains over rivers which seems like something there might be backwards too.

Works well for Ethiopia at least since Italy goes out of it's way to scoot it's divisions west to the river. Nice of them to give your 10% strength divisions a fair chance though.

Basics please.

 

[yess]

If you don't exaggerate this much you have a valid point, the battle planner is not perfect, but it is far from broken. I use the battle planner, and almost never micro my units. If I where to play competitive multiplayer I would micro more, but battle planner works well for casual multiplayer and single player.
Game is more relaxing and focused on running your country with battle planner.
Game is more challenging because you can not snake 1 tank to encircle anymore.
People will disagree on this, but I find the game more enjoyable when playing it as intended instead of exploiting every single ai issue.

 

[rcbricker33]

once again a thread that has valid points that are reduced in value based on crappy attitude. try a to present your case in a more articulate and logical manner without all the foot stomping.

I agree that the BP needs work with exactly the points you made. It would just be better presented in a more civil manner which invites conversation. your way puts people instantly in a defensive position due to its combative styling.

 

[bowman2011]

It's a good post because it starts off as a question, and not a statement.

Let's not socially engineer over the situation, which as OP lays out is kinda sad.

It is not OP's fault that the situation is kinda sad.


------

 

[Uniform764]

Step 1. Make sure the AI can assign troops evenly along a front line with terrain and opposition strength taken into account.

Having 8 divisions on one province and then one division on the other 8 = broken.

Agreed. No argument here.

Make sure that when attacking the AI keeps the line together and dispenses with all the bullcrab strategic re-deployments.

When a normal person advances a front they need the occasional re-deploy. Mostly it can be achieved quicker and without the org hit by shuffling a series of divisions one province over.

The AI having 50% of the divisions strat re-deploying at any one point = broken

Also agreed. Anytime you make any sort of breakthrough or salient, dozens of divisions on both sides will imeddiately reduce their org to zero and start manically redeploying. It's bloody obtuse.

 

[Gyrvendal]

The main difference between AI and player when it comes to front shuffling is that players often prefer to "attack shuffle". Say I have a hole in my line, instead of trying to fill this hole by moving units laterally, I will instead attack the province the enemy is coming from to block him from taking that province and/or try to advance and plug the holes this way. Sometimes when you are overwhelmed you might try to shuffle you units back instead of laterally. Instead the Ai knows only lateral movement and this leads to all kinds of nonsense.

Typical case is the AI attacks one province with divisions from three different provinces. This results in 8+ AI divisions in that province after they win the fight, and empty provinces all around. The Ai then shuffles laterally to fill the gaps. In this case, the efficient action would have been to attack with 8 divisions, then cancel the movement of most of them after the fight is done so the front stays manned; but no, the AI will just send all of its divisions to that province then realise it has created gaps in the frontline and try to fix them. Obviously this behaviour can be exploited very easily by a player.

 

[KiwiNoob]

If you don't exaggerate this much you have a valid point, the battle planner is not perfect, but it is far from broken. I use the battle planner, and almost never micro my units. If I where to play competitive multiplayer I would micro more, but battle planner works well for casual multiplayer and single player.

If you play as China and manpower shortages are a story made up to scare children then sure - go with the battle planner. But whatever you do don't watch it because even if you don't care about the wasted human lives it's painful to watch. It's like a teenager - it seems to identify the right choice then deliberately do the opposite.

...I agree that the BP needs work with exactly the points you made...

Glad to have your support

 

[bERt0r]

Step 1. Make sure the AI can assign troops evenly along a front line with terrain and opposition strength taken into account.

First, I can only agree with half of this. Imho the AI does an OK job of distributing your troops along a frontline. It could be improved if it took terrain into account.
Second, I think a lot of the problems the AI has with this is exactly because the Ai tries to gauge the strength of the enemy and then tries to redeploy accordingly. It would be a lot better if the AI just tried to calculate how much opposition a frontline might field, depending on IC, army size, strategic imporatance of a frontline and just stick to that number. It should then send an appropriate force there and distribute it evenly. If it can account for terrain, even better.
All of the AI's attempts to react to enemy force strengths dynamically usually turn into the "8 divisions in one province, 1 in next" disaster and the redeployment mess.

 

[Bambilambi]

Naval AI is malfunctioning
I am playing as USA Torch 1.3.3 BETA patch 52b0 i think...
basically only cosmetic mods active, no ai mod. all mayors except USA on +1 strength.

What it is doing:
Making bad fleets and loose the naval war within a few week (japan).

I made two 4 cv 2 BB fleets with CL +DD escort. They went patrolling around the philipines.
And basically the whole japanse fleet is dead within a week. My 9 BB fleet bases from Hawai barely had time to launch. Save game included if needed. For me the issue happens every game.

edit: a week is ofc a obvious exaggeration, but it was no more than about 1-2 months
What it should be doing.
It should be using good fleet "templates".
It should be building enough fleet.
It should be able to estimate enemy fleet based on encounters (like the land army can do on a front).

Otherwise, Good work @SteelVolt. The AI and thus the game is getting better and better!
/Bambi

 

[RELee]

AI thread merged with the Great Official AI Thread per the dev's wishes.

 

[Syntalynder]

Playing a game with Germany. Just conquered France in July 1940. There was one coastal area unguarded. The UK AI invaded that area despite there being over 100 divisions close by and me having naval and air superiority. Something is wrong.

 

[MadDjinn]

I'd agree with this.

Even if you have a fleet and a lot of naval bombers/fighters in the area, the AI just plows in with naval invasions. 300+ Naval Bombers combined with convoy raider fleets is not apparently enough to kill incoming transports. Worse... the Naval bombers aren't always part of the naval battles either.

granted, a player must get above 50% naval power in a region before they can cross it with invasions. Is the AI ignoring that? Or is the combo of France+UK fleet just an 'all in' push for them rather than keeping the fleets spread out?

 

[RELee]

Ai threads will be merged with The great official AI thread per the wishes of the developers.

 

[KiwiNoob]

Second, I think a lot of the problems the AI has with this is exactly because the Ai tries to gauge the strength of the enemy and then tries to redeploy accordingly. It would be a lot better if the AI just tried to calculate how much opposition a frontline might field, depending on IC, army size, strategic imporatance of a frontline and just stick to that number. It should then send an appropriate force there and distribute it evenly. If it can account for terrain, even better.
All of the AI's attempts to react to enemy force strengths dynamically usually turn into the "8 divisions in one province, 1 in next" disaster and the redeployment mess.

The AI ending up with 8 in 1 & 1 in 8 doesn't mean it shouldn't worry about opposition strength - it just needs to do it properly instead of whatever it's doing now.

All the things you've listed that should be taken into account "IC, army size, strategic importance of a frontline" are all the things the battle planner AI SHOULD NOT take into account and if they are trying to do that then that is part of the problem. The strategic AI should be worrying about that stuff and choosing how many divisions to assign to a front line. The battle planer AI should only worry about the divisions it has and the provinces it needs to cover, then just deploy as best it can with that information.

 

[bERt0r]

That was what I meant. The Strategic AI should determine how many troops per front and the battleplan AI how to deal with its given units.
Right now, it seems as if the shuffling of troops causes the Battleplan AI to constantly call for more troops. The expanding and overlapping frontlines (especially in AI controlled countries) dont help either. Also take note that AI controlled Germany immediatly makes about 8 Army groups out of its starting 30 or so divisions.

I think the 2 Battleplan AI's on one frontline are escalating each others redeployment madness by reacting to troop strength/positioning of the other. If you plan out a frontline, do you take into account how many divisions your enemy may have stacked in one province? No, you distribute your units, make use of the terrain and try to fill each province up to at least 80 combat width.

Simplest example is the Siegfried Line and the Maginot Line. The AI somehow loves to put more troops into the mountain provinces behind the river than the plain provinces every single time.

 

[Hanti]

1.3 feedback

• 
  
  • What the AI is doing wrong
  • What you think it should be doing instead.
  • Any potentially relevant context, as many things when are crazy in one situation is a perfectly reasonable thing to do in another. For example, if Japan does not invade USA it may simply be because they are getting their ass handed to them by Soviet."

What the AI is doing wrong?

The most critical AI wrong-doing is wasting it's resources. There seems to be no AI friendly template of what to do with all that factories, manpower, natural resources, equipment.

What you think it should be doing instead?

AI should make calculations like this:
- should I start training 25 motorized divisions without equipment in pools for them and manpower in reserve?
- should I keep training 25 divisions and zero pool in manpower while my ships can't be uploaded because lack of manpower?
- should I keep producing 100 trucks per day, while having shortage of 50k infantry equipment?
- should I keep my aircraft producing lines at the bottom while there is no natural resources for them to be built, and while having literally no air cover on map?

There could be improvement in AI behavior if it could re-consider it's actions at least one in a while during campaign.
It seems that when AI starts some action (eg. train that division) it never reconsiders it and is not able to cancel its actions.

Playing Torch 1.3.2