The great official AI thread

[dm85]

The post-war peace process still has significant issues. Major countries still take provinces they shouldn't, whether historically or ahistorically. For example, UK took core provinces from France even though they should have been liberated back to France during the war. I've also had instances where US and UK took coastal provinces on the Chinese coast even though these should go to China. I've had Soviet Union not create East Germany even though they honored the script that's supposed to do so. These issues make it nearly impossible to try and replicate a historical peace process that at least mostly looks like it should. I've even had Nazi Germany survive through the peace process and keep a random province.

 

[Headhunter_]

I'd like to share my observations during a game as Germany on patch level 1.3.1 (with TfV DLC, no mods).
- regular difficulty / historical mode

I've been playing HoI since HoI1 and have all incarnations.

Air combat
AI crams their aircraft way above the limits of an airport.

• I was wondering why the sky above the English channel was so quiet. It's because the AI has stacked all their aircraft first in Malta, then in Algiers and finally in England. Always being significantly above what the airport can hold.

Ground war
AI does not focus enough on their mainland.

• Britain leaves their mainland mostly undefended to support overseas operations and naval invasions.
• France pushed ~20 divisions into Africa to support operations there. I would have secured the borders of mainland France. I guess they would have agreed two weeks later
• Italy, once allied, tend to be all over the place. Invading Norway, border patrol in France, fighting in Africa, but do not defend their core sufficiently. It would be nice to tell them to sit still and prepare for an offensive in the East or focus on one thing like naval superiority in the Mediterranian or Egypt.

AI forethought would be nice. AI Soviet DOWs Finnland and immediately looses Leningrad. Some border security would be nice before you DOW someone.

Sahara. In my current game there are 20+ allied divisons roaming through the Sahara.

Naval warfare
Naval invasions performed by the AI are suicidal.

• England continues to invade Belgium with 1-6 divions within a half a year after the fall of France. I've got 40+ divisions in the region, so I let them invade and then push them into the sea.
• As I conquer Spain, spanish troops assist a naval invasion in Belgium. As I'm not looking they get 25 divions on ground, but I can push them into the sea with my 40+ divisions, plus they would have needed their troops in the spanish mainland.

It would be nice if the AI could calculate if they have a chance of a sustainable invasion. E.g. if all my troops are facing the Soviets, then it's a good time to look at Belgium / France. If I just got done conquering it and have 40 divisions sitting around, then it's not a good time to perform a naval invasion.

~~~~~~~
As a consequence of the allied countries being all over the place and neglecting their core, I do not seek allies but conquer them. This way I do not have to deal with these issues.

What I really liked

• The AI does a good job of pulling troops out of areas that could become encirclements.
• I thought I was prepared for Barbarossa until the Soviets showed me otherwise. You really have to start thinking Barbarossa in 36 and not get sidetracked.

 

[Hammer54]

With regards to Headhunters observations. I agree With ground war observations and defending homeland, also With the doomed naval invasions. With regards to naval warfare, one thing bothers me. The AI doesnt seem to consider changing fleets if I change mine. If i go from small fleets to large fleets, AI doesnt respond in kind, and stays in the fight over areas With small fleets. They should withdraw og atleast match local fleet Power. Now they just loose their fleets.

And With regards to Barbarossa. In my latest game, i was busy in England when russia decided it was time for war. 41 was a bad year for me until the empire surrendered. Small scale encirlements are hard to pull off right now, and in general it will be hard to win the grind against russia. In 42 i decimated russia With a series of large scale pincers, which the AI doesnt seem to cope With.

Also, played Poland last night. Made as much infantry as I could, and started to delay german advances. I was all but broken, and they were just outside warzaw when they decided to open the west front. With a human player in poland that is a very bad move. It makes poland a bit more interessting to play, but i understand why german AI has problems on its own. Poland should fall atleast by the end of september, and the AI doesnt need to hurry west. Move west doesnt need to start before spring 1940, and I think the AI, in a historical game, should finish poland take its time making preperations, and get a solid foothold in norway before they attack france.

The thing about the focus tree, I thought that it sort of directed the AI, or the other way around. That the AI chose a path, for example east then North then west, THEN used the focus tree to make that possible, and started planning actions on the ground accordingly (like setting up naval invasion of Norway before actual DOW). Now it just seems to start to grind in every direction, where the AI doesnt prioritize tasks in any way. (I've seen russia pull troops away from japan, so something is happening, but maybe not so much With regards to advancing Strategic goals.)

 

[Dracolithfiend]

Alright so if you check the US divisions that are on the Solomon islands they have been sitting in that spot with "no viable path to target" for about 2 years. It would be great if the AI took some initiative and did something with them.

 

[VanMavus]

Just started a game as Spain to play as Fascist Spain and I noticed this oddity.

The two Italian volunteer divisions have been sitting there the entire civil war. I'm not sure if this has something to do with the tag switch, as I apparently have no capital either.

 

[PsychoLold]

@SteelVolt

I haven't seen this AI behaviour before, so I think it's worth mentioning here.

For some time now in my current campaign I've noticed that the UK has some troops on the USSR-Sinkiang border. After tag switching to UK, I've noticed that they have a battle plan to attack Sinkiang activated They have no war goal against Sinkiang, and we're all fighting a common enemy. Yet the UK wants to conquer them.

 

[Soranya]

I suggest you watch Quill18's latest Lets Play on HoI with canada - the last 5 or so Episodes should be enough to get a good impression of some mistakes the AI still makes. (Basicaly from the episode Quill18 does invade Spain onwards.

 

[Moser]

In other news, Germany has decided to make Marsa Matruh its new nuclear testbed.

I'm not gonna post the save game because I've modded quite a few AI templates and focuses, and I'm using No Man's Land, but I thought this was pretty funny and probably unintended, so I'd thought I share it.

 

[Kai Lae]

You still see this as much in 1.3?

Not as much, much improved. However, instead, the english AI tends to ship 15 divisions to dunkirk where they sit around and do nothing. No idea why.

 

[Kai Lae]

Problem: AI does not understand attrition warfare

Expected behavior: AI should understand attrition warfare.

What's happening is that as france, my general strategy is to build an army that upon the start of war I can push into benelux to stop the Germans. The way that I do this is by defending key points with cities, forests, and especially river crossings to give me a large defensive advantage that negates their numerical advantage. I then begin a period of attrition warfare where I invite them to attack where I will kill more of them than they will of me, until a period occurs that they have manpower problems.

The issue is I can understand what I'm doing, any player could as well, but the AI can't. Using this strategy in my last game I had 300k in French casualties, while the Germans had over 8.5 million - roughly a 26-1 advantage. In effect, I killed a platoon for every French soldier lost. The results of this is without fail, the german line collapses in 1942 because they simply run out of men. At this point, a counterattack that causes German collapse is child's play. Against any human player, they would see this coming a long time before this and would change their strategy to prevent it from happening.

In effect, I create a mini maginot line, and the AI is more than willing to completely bleed itself dry because it lacks the ability to recognize what is happening. This is....not ideal and should be looked into.

 

[Dalwin]

As Australia, I took Rhodes after North Africa fell. The problem shown here is the Dutch on Rhodes planning an amphibious invasion of Rumania. This is an invalid route since the Turks will not allow them passage through the straits. The planning AI should realize this and spend its limited resources planning other invasions. Those might also be futile but at least they should be valid.

At the same moment British forces in southern Norway have plans to invade a small island in the Baltic even though the Germans have already seized Denmark thereby sealing the straights.

 

[Moser]

Problem: AI does not understand attrition warfare

Expected behavior: AI should understand attrition warfare.

What's happening is that as france, my general strategy is to build an army that upon the start of war I can push into benelux to stop the Germans. The way that I do this is by defending key points with cities, forests, and especially river crossings to give me a large defensive advantage that negates their numerical advantage. I then begin a period of attrition warfare where I invite them to attack where I will kill more of them than they will of me, until a period occurs that they have manpower problems.

The issue is I can understand what I'm doing, any player could as well, but the AI can't. Using this strategy in my last game I had 300k in French casualties, while the Germans had over 8.5 million - roughly a 26-1 advantage. In effect, I killed a platoon for every French soldier lost. The results of this is without fail, the german line collapses in 1942 because they simply run out of men. At this point, a counterattack that causes German collapse is child's play. Against any human player, they would see this coming a long time before this and would change their strategy to prevent it from happening.

In effect, I create a mini maginot line, and the AI is more than willing to completely bleed itself dry because it lacks the ability to recognize what is happening. This is....not ideal and should be looked into.

I believe the AI is set up in a way to reinforce a front if the plan progress is less than some percentage (50%?). Don't they stop execution of the plan at some point? And should they not execute the plan if the defensive side is too strong? Perhaps they don't calculate the plan value correctly?

I've seen at times the enemy strength indicator completely flip-flop at times. It will go from a overwhelming enemy strength, to overwhelming friendly strength, then back to overwhelming enemy strength in a couple of hours. The AI then executes its plan when it briefly thinks its much stronger.

 

[Dracolithfiend]

Well I have seen the AI repeatedly make the same plan and cancel it. Maybe it is a similar issue?

 

[Spruce]

I'm playing as Japan. Germany totally screwed up and got steamrolled by the Soviets in 6 months time - I reloaded an older save file

1. Von Rundsted commanding 435 divisions to attack Moscow. The game tells me that Von Rundsted can command 24 divisions without efficiency loss.
2. The AI is not using support attack in a consistent way.
3. Rommel is Guarding the German heart land (on defense plan), but nobody is guarding Western Europe (Vichy is not present -> I presume this might the reason that the AI is not defending Western Europe). So the area for Rommel is limited to the German old borders.
4. Von Manstein is sitting with quite a strong army at the border of Nationalist Spain - why ? Nationalist Spain is not a imminent threat, ... and in France, Belgium, Netherlands there's not one German division.

The UK AI is planning an invasion of the Westwall - nobody is guarding the westwall ... all divisions are in Germany, Spain border doing nothing to prevent the German downfall. And Von Rundsteds 435 divisions doesn't seem right.

hope this helps

 

[Toledo37]

AI (battleplan) doesn't seem to know how to do replacements in combat, maintaining lots of disorganized divisions as reserve in the battle, thus not allowing them to recover...
It would be nice if AI would care about letting divisions recover and don't allow too much divisions together in one province, specially if they need to recover, they should stay in the back.

Or let divisions in reserve recover, if AI just can't make replacements

 

[Dalwin]

I'm playing as Japan. Germany totally screwed up and got steamrolled by the Soviets in 6 months time - I reloaded an older save file

1. Von Rundsted commanding 435 divisions to attack Moscow. The game tells me that Von Rundsted can command 24 divisions without efficiency loss.
2. The AI is not using support attack in a consistent way.
3. Rommel is Guarding the German heart land (on defense plan), but nobody is guarding Western Europe (Vichy is not present -> I presume this might the reason that the AI is not defending Western Europe). So the area for Rommel is limited to the German old borders.
4. Von Manstein is sitting with quite a strong army at the border of Nationalist Spain - why ? Nationalist Spain is not a imminent threat, ... and in France, Belgium, Netherlands there's not one German division.

The UK AI is planning an invasion of the Westwall - nobody is guarding the westwall ... all divisions are in Germany, Spain border doing nothing to prevent the German downfall. And Von Rundsteds 435 divisions doesn't seem right.

hope this helps

Not sure where you got 24 divisions for von Rundstedt. He starts as a field marshal, in fact the only good one the Germans get. He definitely has unlimited command capacity. This does not excuse that the AI plan of putting so many in one army and giving it one giant arrow is terrible tactics.

 

[fabius]

AI (battleplan) doesn't seem to know how to do replacements in combat, mantaining lots of disorganized divisions as reserve in the battle, thus not allowing them to recover...
It would be nice if AI would care about letting divisions recover and don't allow too much divisions together in one province, specially if they need to recover, they should stay in the back.

Or let divisions in reserve recover, if AI just can't make replacements

This ! I see this really weaken the AI time and time again.

Fix AI template, research and production choices. Also fix this, abandoning fronts, and over-stacking for mass supply attrition and we''ll have much more challenging game.

 

[Spruce]

Not sure where you got 24 divisions for von Rundstedt. He starts as a field marshal, in fact the only good one the Germans get. He definitely has unlimited command capacity. This does not excuse that the AI plan of putting so many in one army and giving it one giant arrow is terrible tactics.

Yes you are right, Von Rundstedt is Field Marschal, sorry for my mistake. But still it feels very weird to have over 430 divisions under one leader.

Now that you mention that, the Soviets also have one Field Marschal "Tuckachevsky" overseeing a large part of their front (over 100 divisions). I reloaded when Barbarossa was in the begin fase. This Tuckachevsky has 6 traits accumulated, has skill level 7 and as such this gives a huge amount of bonuses to the Soviets while they have far less experience in warfare at the start of Barbarossa. As a reference, Von Rundstedt has 3 traits and skill level 7, but has done a lot more fighting before Barbarossa.

When I did a reload, some months before Barbarossa was started - I found that Tuckachevsky only had 1 trait and skill level 4. Tuckachevsky "over boost" above Von Rundstedt is not feeling good. This guy gives huge bonuses (over 100 divisions).

 

[Moser]

Yes you are right, Von Rundstedt is Field Marschal, sorry for my mistake. But still it feels very weird to have over 430 divisions under one leader.

Now that you mention that, the Soviets also have one Field Marschal "Tuckachevsky" overseeing a large part of their front (over 100 divisions). I reloaded when Barbarossa was in the begin fase. This Tuckachevsky has 6 traits accumulated, has skill level 7 and as such this gives a huge amount of bonuses to the Soviets while they have far less experience in warfare at the start of Barbarossa. Even more, the Soviets had a white peace with Finland during their winter war.

I think this Tuckachevsky "over boost" doesn't feel right.

There are couple of parameters in the defines.lua that specify the maximum and minimum length of a front that a single commander can hold, but that feature doesn't seem to work, and may have been scrapped altogether before release.

 

[Dalwin]

The AI really likes to put most of its units under one field marshal and then give a number of other generals very small groups to command. Personally I think this is a major contributor to how badly it performs. The battle plan AI does not do well with large groups and long frontages. In essence our automated opponent has a preference for doing something that has a large negative impact on its efficiency.

You can try this yourself sometime. Make a really large army as either Germany or Russia and fight the war, but do it with as little manual micro of units as possible. You will get a feel for just how much of a handicap the AI has given itself by taking this approach.