Issues with AI Germany is why Germany loses the war in 1941.

[Mindfield]

I dont want to say this, but we face the limitations of the real time constraints for the AI.
The AI can do a lot of amazing things in HoI4, the abilities are here, just lacking enough calculation time.
I dont know how much room there is left for optimisation and bug fixing.
And the load is getting bigger, new features and a revamped air war will cost.

I did a human_ai Germany, and by 1939 they have France, Belgium, Netherlands and Poland occupied.
But completly failing in 1941 against Russia.
Why:
- The AI did not build any synth factories, running into a huge rubber deficit
- Build Battleships
- Build Strategic Bombers
- Unable to build enough Fighters due to rubber deficit: None were build after 1940, and the 600 fighters "airforce" was completly wiped out but a few CAS.
- Unable to defend the coast, huge landing in Belgium
- Big destruction of industry. Resistance? What is that?

Curiosity:
- Exile-Danish troops occupied Rome. Hmm..

Maybe the time has come to rethink the RTS environment with a lot of things the AI has to achieve to fulfill the player expectations.
Reduce load?
Cheating AI?

 

[Gwydion5]

Exactly what i was looking for and then some! Thank you very much for the well thought out reply. I will try to incorporate some of your ideas into my next game as GER. Cheers.

You are welcome. I did beat Russia last night, but it took till October of 1942 for them to capitulate. Even though I had encircled and destroyed about 100 or more divisions from Jul-December of 41, Russia just wouldn't give up and kept putting a wall of divisions for me to slowly grind down. Even when I got Russia down to 10%/10% the bastards annexed Tannu Tuva giving them back 1 more percent towards capitulation while I'm invading Sinkiang.

 

[Meglok]

I dont want to say this, but we face the limitations of the real time constraints for the AI.
The AI can do a lot of amazing things in HoI4, the abilities are here, just lacking enough calculation time.
I dont know how much room there is left for optimisation and bug fixing.
And the load is getting bigger, new features and a revamped air war will cost.

I did a human_ai Germany, and by 1939 they have France, Belgium, Netherlands and Poland occupied.
But completly failing in 1941 against Russia.
Why:
- The AI did not build any synth factories, running into a huge rubber deficit
- Build Battleships
- Build Strategic Bombers
- Unable to build enough Fighters due to rubber deficit: None were build after 1940, and the 600 fighters "airforce" was completly wiped out but a few CAS.
- Unable to defend the coast, huge landing in Belgium
- Big destruction of industry. Resistance? What is that?

Curiosity:
- Exile-Danish troops occupied Rome. Hmm..

Maybe the time has come to rethink the RTS environment with a lot of things the AI has to achieve to fulfill the player expectations.
Reduce load?
Cheating AI?

Your comments regarding the lack of rubber and it's effects on AI industry are right on. It really looks like not enough testing was done on the effects of these industry code changes. Their unintended consequences on AI performance in the air and naval production aspects of the game are just horrendous. Unfortunately nothing will likely be done regarding any hotfixes or repairs until at least next week as PDS is atThe Game Developer's Conference this week.

 

[ltccone]

I would still like to see a startup option which forbids use of volunteers everywhere but the SCW.

I heartily agree with you. In my Japan game before the US finally entered the war I destroyed half their army sent as 'volunteers.' And when I was fighting China I had to face Soviet armor sent to both China and Red China.

 

[Gwydion5]

I did a human_ai Germany, and by 1939 they have France, Belgium, Netherlands and Poland occupied.
But completly failing in 1941 against Russia.
Why:
- The AI did not build any synth factories, running into a huge rubber deficit
- Build Battleships
- Build Strategic Bombers
- Unable to build enough Fighters due to rubber deficit: None were build after 1940, and the 600 fighters "airforce" was completly wiped out but a few CAS.
- Unable to defend the coast, huge landing in Belgium
- Big destruction of industry. Resistance? What is that?

One thing I have noticed in 1.3.3. vs 1.3.2 is that AI Germany is no longer producing enough for a long war. If you watch the AI production screen for Germany you can see it ramp up production lines for a few months and then bring them down to focus on the new thing that has the highest deficit/need. For example :

That's January 10th, 1936. It's putting a high amount of MIC's on light tanks. Which honestly is fine, if that's part of the long term strategy/plan of the AI. Like for example if the AI was doing this because it wanted to produce 6 more divisions of light tanks for a war in late 36, or early 37. But since this is a historical game, does it really need to put so much production on light tanks this early?

February 10th, 1936. The AI has added a second line for light tanks, and reduced Infantry Equipment by a little more than half. Again if there was a reason for this, like AI Germany planning for an early war, then great. But since war is 3 years away, perhaps it should be a bit more reserved and balanced on production priority.

May 10th, 1936. The AI is now spinning down production of light tanks and putting the MIC's in to Artillery. It is pulling MIC's from line 6 which has the most production efficiency for producing light tanks, while leaving Line 1 that has the least production efficiency for light tanks alone. It has also added an MIC back to Support Equipment, but now with a bit of inefficiency.

May 22nd, 1936 : AI continues to pull MIC's from Line 6 Light Tanks to fund this new demand for Artillery in line 5, while leaving Line 1 light tanks alone. This effect where it ramps up and down from 1936-1939, without planning for a long war, and pulling from the most productive lines, as well as pulling from lines only to put them back a month or two later, is what is responsible for the equipment deficits come in 1941.

Your comments regarding the lack of rubber and it's effects on AI industry are right on.

AI Germany does a decent job at using trade to fill in deficits, I would argue it's not worse than 1.3.2. In 1.3.3 AI Germany now builds around 4 refineries by late 1940, around 2 by august 1939, with a third usually in the queue. In 1.3.2 it did not do this, so it's actually generating more resources/rubber than in 1.3.3. I really think this all boils down to the new production priority AI. Looking at the Guatemala game in 1940 I think these screenshots below provide some insight.

So we have a 25 rubber deficit. Motorized is the 1st priority in the build queue consuming 12, then it's aircraft after that. My question is why is the AI Prioritizing motorized?

And the reason why is because it doesn't have a surplus, of anything really. If the AI's production priority logic is not setting production targets higher than existing divisions, and divisions in recruiting, it's doomed to fail. Typically in games where I play Germany, I have plenty of reserves built up for Infantry Equipment, Support Equipment, Artillery and Motorized. I plan for it in 1936 knowing that I'm going to be adding support divisions and that war/attrition is going to consume them. But as I've been seeing the AI's logic is far too sensitive and reactive to immediate needs, without a long term plan / baseline for the equipment it should have by 1940. It's not balancing short term and long term production goals.

Japan on the other hand could be a different matter, I don't play Japan since pacific theater / naval combat is not interesting/appealing enough for me. But with Germany I can confidently say that one of the major issues for them is that AI logic for production is only focused on immediate need/use without balancing/planning for long term production targets. But this issue is multifaceted, where some things are probably amplifying the harm/weakness. For example, I have not taken the time to watch the construction logic. It might be picking low infrastructure provinces over high infrastructure provinces for CIC's, MIC's, refineries, etc... But looking at the save games, it seems to max out the highest the infrastructure areas. So probably not that big of a deal. On the other hand the issue of pulling MIC's from the most productive lines, is amplifying the production deficits. But again how much is that accounting for in terms of total production output, 5%, 10%, 20%? Hard to say without some serious number crunching.

EDIT: In addition I'm going to play a normal/regular difficulty game up to July of 1939 where I play the game like I normally would (probably not change the trade policy), and then hand it off to the AI to see if it does any better with Russia in 1941. If it still fails against Russia, then there is probably another issue present in this version that we may not be talking about.

 

[Mindfield]

Interesting to see my game was different.
- No Vichy France
- Frech Provinces given to Italy
- By that time (more or less) the deficit was -41 rubber

Good finding on how the AI handle production!

 

[golruul]

In my experiences (veteran/historical/all sliders max except me / no mods) in two games so far, Germany took Poland/France/Low Countries within three months. Denmark and Norway fell soon after Germany declared war on them. War against Soviets lasted until 1944, when the Soviets finally won.

Japan defeated China both games rather easily.

I think the max sliders is helping with these "historic" results.

 

[Gwydion5]

So I did a game with regular difficult, no boost, historical focus as Germany today. I managed Germany until June 14th, 1939 and then I handed over control to the AI and it still fails / gets bogged down with Russia. I will leave 3 save games at the bottom of this post for you to test with. "human_ai" is the console command you want to use to turn Germany over to the AI.

So from 1936 - 1939 here was the order of national focuses, political power and refineries.

• NF = Autobahn, Rhineland, Westwall, Army I, Industry 1, Industry 2, Aschluss, Tech Treaty, Air 2, Sudentland, 1st Vienna, Fate of Czech, Rocketry, Extra Research, Army 2, Eastern Claim, Air 1, Ribbentrop Pact, Danzig or War.
• PP = Schacht, Gunderian, Industry Krupp, Funk, Tank Henschel, War Economy, Mauser Infantry, CAS Junkers, Halder, Schorner, and a few others after that.
• Refinery Construction = March 37, May 37, December 37, May 38, 2x June 38, 2x Jan 39

Logistic Screen:

Division Overview:

Positioned pretty well for a good war with Russia. Decent amount of medium armor and self-propelled artillery. Now that I think about it, since the AI doesn't really use cavalry for suppression I probably should've just built regular infantry divisions. Oh well.

• The first zip, is the June 14th 1939 save, before I handed over control to the AI. I highly recommend loading this game up, doing the human_ai console command and watching what the AI does to the production lines over the next 2 - 3 weeks on a day by day basis. Specifically what it does to the medium tank line, the infantry equipment, and artillery. Do it a couple times as the AI does it a little differently, but in the end it will disband the high production efficiency lines. You can also use this to see how AI Russia does against your command of Germany and if that matches the July 1st 1941 save below.
  
  
• The second zip is January 1st, 1941, before war has been declared with Russia. Just incase you want to see what the AI did during my game, or have a game that starts close to the war.
  
  
• The third zip is July 1st, 1941, roughly 6 months of fighting, and clearly Germany is going nowhere. Also take a look at all the division templates the AI created. Kind of crazy / pointless. Is the AI exempt from the army experience costs regarding division design?

I'm not sure what's happening here that is causing such an epic fail against Russia. Is it the large single front? It's never been much of a problem before with the AI. The AI doesn't seem to be using air power like it should or did in the past either. I noticed a lot of wings on standby stationed throughout the theater. I also noticed heavy CAS support in the Northern Balkans, and nothing really in Eastern Poland.

Anyway I'm at a loss as to why the AI is getting bogged down. Economically and logistics wise, Germany should be fine and rather successful against Russia. Yet that is not happening. It can't all just be production / resource issues. There has to be something else going on here. Perhaps continuing the 1939 save as a human until September 1940 before handing control to the AI is worth trying? Maybe I'll try that later tonight and see if it makes a difference.

 

[him_15]

There really has to be a quick hot-fix to improve the Germany AI otherwise the game is just unplayable unless you play Germany herself!

 

[Gwydion5]

I created a new post on the bug report forum and uploaded the June 1939 save, as well as the new August 1940 save. I also uploaded a zip file of screenshots from the 2 AI games I ran from the August 1940 save. The AI does much better against Russia from the August 1940 save than the 1939 save. I think that speaks to the production issue/logic needing some improvement. But I also noticed and believe the AI is worse with air power in 1.3.3 over 1.3.2. Granted these are snapshots at a given point, not tracked and analyzed over time. With that said though the utilization of air power against Russia after being at war for 4 months or more is a bit troubling. Also it must be stated that loading from the August 1940 save, and giving the AI control of Germany, the AI will successfully do Operation Sea Lion on the England and forces the Allies to capitulate by early 1941. So it's only active war is with Russia in both of these air power screenshots.

From the AI_TEST-1 game (First playthrough of loading the August 1940 save game and handing control over to the AI)

Not as bad as the next screenshot, but still a few wings that have no orders as well as some wings stationed in one place trying to operate in an air zone far away. It would be helpful to actually track this over 2-weeks to a month just to make sure though.

From the AI_TEST-2 game (2nd playthrough of loading the August 1940 save game and handing control over to the AI)

And with that I will probably be taking a break for a bit on post spam with screenshots etc... As I said in the bug post, it's a tedious and time consuming process producing all this information and starts to wear thin. Hopefully our feedback will prove useful to PDS on producing a hotfix or two for 1.3.3.

 

[pdstanbridge]

I have played three campaigns with the new patch, one as Germany. In both AI German games it was beaten like a ginger stepchild by 1941. The initial problem is AI naval invasions. They are too frequent and too aggressive, causing Germany major issues.

 

[DukofDeth]

Either they need to program the AI on how to perform resistance suppression, and also coastal defense, and maybe force the AI to assign appropriate forces to a theater set up for these roles [and nothing else], and force the AI to leave said theater alone; or they need to significantly reduce the effects of resistance for AI controlled nations, as well as reduce the instance of invasions vs. AI controlled nations, since the AI just can't handle it; then to prevent players from abusing such a thing, apply the logic that, if you conquered/capitulated it, you garrison it, and you cannot transfer it [unless your AI ally has claims]; they also need to remove rubber from light aircraft for the sake of an AI nation, or reduce the effects for a lack of rubber on AI nations; its one thing to come up with new restrictions against human players - we can adapt. The AI can't, or won't.

I also get the feeling that when they [Paradox] test their changes, they're doing it as Germany, so they aren't seeing just how hobbled the German AI is. Or they aren't playing their test games long enough to figure it out.

EDIT - I just had a thought about how to limit invasions. They need naval superiority to proceed, but they never require air superiority to proceed. One of the real-life requirements for Overlord was air superiority over northern France; why isn't the AI being forced to achieve that, or to protect against it? It would be far easier for AI Germany to maintain air superiority over its own turf early in the war, than to maintain enough divisions to contain/repel any invasions run against it.

Also, the navigable channels between Denmark and Sweden are quite limited, so the mere advent of war between Germany and Britain should be enough to close that part of the Baltic. Isn't the strait between Copenhagen & the Swedish mainland only a few miles wide at some point?

 

[supreme]

All valid complaints in this thread, but the weak German AI has been a problem since release and i doubt its going to be changed any time soon, didnt you all see the post about HOI4 selling 500k copies? Podcat aint got time for fixing that shit now, he is busy snorting coke off a strippers arse in his limo while he comes up with ideas for all the dlc the game needs!

 

[Gwydion5]

I have played three campaigns with the new patch, one as Germany. In both AI German games it was beaten like a ginger stepchild by 1941. The initial problem is AI naval invasions. They are too frequent and too aggressive, causing Germany major issues.

I would like to see AI invasions improved as they are not ideal. But in their current state, I don't think they have too much responsibility in AI Germany not performing well against Russia / 1941. At least I have not observed any issues related to them, like say AI Germany pulling 3/4ths the divisions off the eastern front because the allies are in Western Europe. It might be responsible for AI Germany's high casualty rate with some of the games though.

Either they need to program the AI on how to perform resistance suppression, and also coastal defense,

I agree on this. There needs to be an infantry garrison division design for light/moderate defense of things under the garrison command. Then there needs to be a separate garrison division specifically for resistance suppression. In addition the AI should not (generally) use resistance suppression divisions on nations that have capitulated via a peace treaty. IE Germany annexing Russia for example. If nothing else though and PDS want's a quick fix, it simply needs to take defending coastlines out of the mix. That should reduce the garrison requirements by a lot and free up a few divisions until it can do proper division design and assignment for garrison duty.

I would also add a rule that limits the percentage of divisions the AI can have reassigned off any active combat front for an army and theater. It doesn't have to be an absolute limit either, it could be dynamic based on time and conditions.

All valid complaints in this thread, but the weak German AI has been a problem since release and i doubt its going to be changed any time soon

All we can do is try to provide the best evidence we can on what needs to be fixed. There has always been AI problems, mistakes happen, and I do think HOI4 is making progress. I wouldn't put the time and effort trying to capture them if I didn't believe the game has a solid foundation for greatness that needs a little help and refinement. I guess I still have hope/faith for the game, but am resolved to wait it out.

 

[assassinsat]

One thing worth considering is if Human Germany players are finding Russia too strong in 1.3.3, or significantly boosted with 1.3.3 that it's preventing regular difficulty AI Germany from doing well against them? Just a thought.

Indeed. I started a game as Germany. Russia, GB, Japan and Italy boosted but on regular difficulty. No Ironman. Expert AI 2.0, No Man's Land and No Expeditionary Forces mods enabled.

Even from before the war there are resource shortages leading to production stutters. This compounds once GB enters the war as the convoy route is through the channel into Wilhelmshaven. This does not change to a reasonable port such as Brest even after capturing France. 50 60 convoys die in the channel per month. There is barely enough rubber or oil to produce planes and tanks when it comes to Barbarossa. I'm sure an AI Germany suffers the same and also may be diverting planes to guard the channel in order to let their convoys safely through. Just a thought.

AI Russia is looking very solid though. Gave quite a fight all the way through to Moscow. And even when I was 2 provinces away from there, they launched a massive counterattack that pushed me back some distance away. Thankfully my new panthers led the way right back (after 3-4 months).

 

[him_15]

I am curious that is there going to be any hot-fix patch for this issue, or do we have to wait until patch 1.4 Oak? There seems to be no response from the Dev yet, and I can't imagine to play only Germany/Italy for the next couple of months..

 

[Agiknight]

Indeed. I started a game as Germany. Russia, GB, Japan and Italy boosted but on regular difficulty. No Ironman. Expert AI 2.0, No Man's Land and No Expeditionary Forces mods enabled.

Even from before the war there are resource shortages leading to production stutters. This compounds once GB enters the war as the convoy route is through the channel into Wilhelmshaven. This does not change to a reasonable port such as Brest even after capturing France. 50 60 convoys die in the channel per month. There is barely enough rubber or oil to produce planes and tanks when it comes to Barbarossa. I'm sure an AI Germany suffers the same and also may be diverting planes to guard the channel in order to let their convoys safely through. Just a thought.

AI Russia is looking very solid though. Gave quite a fight all the way through to Moscow. And even when I was 2 provinces away from there, they launched a massive counterattack that pushed me back some distance away. Thankfully my new panthers led the way right back (after 3-4 months).

Had a very similar experience in my Germany games this week (also using the same mods), actually pretty fun to lose for once! Or twice... If AI germany could do as well as AI Russia does now that'd be awesome. I do wish I didn't have to use an AI mod to fix some of the other issues of the vanilla AI, though. Expert AI does noticeably improve their play.

 

[Meglok]

I am curious that is there going to be any hot-fix patch for this issue, or do we have to wait until patch 1.4 Oak? There seems to be no response from the Dev yet, and I can't imagine to play only Germany/Italy for the next couple of months..

They are all at GDC this week. I suspect they are reading, taking notes, and discussing a response plan over adult beverages.

 

[podcat]

I would say Russia is putting up more of a fight than I'm used to

You are welcome. I did beat Russia last night, but it took till October of 1942 for them to capitulate. Even though I had encircled and destroyed about 100 or more divisions from Jul-December of 41, Russia just wouldn't give up and kept putting a wall of divisions for me to slowly grind down. Even when I got Russia down to 10%/10% the bastards annexed Tannu Tuva giving them back 1 more percent towards capitulation while I'm invading Sinkiang.

This also helps establish that AI Russia did not become OP in this patch

Reading your comments I can't really see how you come to that final conclusion. You are clearly saying that soviet is tougher to fight now.

All valid complaints in this thread, but the weak German AI has been a problem since release and i doubt its going to be changed any time soon, didnt you all see the post about HOI4 selling 500k copies? Podcat aint got time for fixing that shit now, he is busy snorting coke off a strippers arse in his limo while he comes up with ideas for all the dlc the game needs!

lol havent people been reading the dev diaries where we have said we are prioritizing bug fixing and a bigger patch over DLC making? Maybe I should have been clearer, but I'm pretty sure I mentioned it in 3 whole diaries. atm we are putting something like 70% of our resources into patch/free update work, which I suspect most other companies would find utterly insane. We are doing it because we care ourselves, more than you guys probably. I heard a great quote from a designer roundtable at GDC the other day where a guy said "the only person more upset about a bug in the game they just ran into and are venting on the forum is the guys who are working their asses off on the game fixing that bug". I'm always chocked when people don't seem to realize this.

Balance in HOI is always going to fluctuate and you will always need to use sliders to boost nations to tailor your experience better. Most people play germany by a vast margin, so we will usually opt for making their experience better when you have a tradeoff. If germany beat soviet union every time, but only because we made soviets weaker thats not very fun when playing germany. Also I am not sure that things have changed that much on the macro level. germany loses slightly earlier now to the stronger soviets but they have been losing to soviets on average in every patch since release basically. To give you an idea on what we aim for "average outcome"-balance wise before releasing patches (this is minimum 20 test games or so):
- Germany should beat france and poland by historical date 100% of the time
- Japan should not lose to china by 1940 in 80% of games (usually ends up being 100% though)
- If germany or its allies are getting invaded sufficiently they should be getting bogged down and start losing in russia in a majority of games. If allies are not pushing them in some way (bombing or whatnot) they should be winning vs soviets in a majority of runs.

I think germany's biggest problem atm vs soviets is front shuffling because its stops their momentum. This however has been a problem since 1.0 basically so its no easy fix. We have tried lots of minor attempts at tweaking, but its become clear that it need some significant rewrites to solve the problems for good. Thats stuff we are working on for 1.4
For 1.3.3 we had to pick between making soviet more of a pushover for german/axis players or have german ai have a tougher time vs soviets and thus making the allied experience potentially worse. I think prioritizing germany was better. especially since sliders will help shift chances about.

 

[grandad1982]

Reading your comments I can't really see how you come to that final conclusion. You are clearly saying that soviet is tougher to fight now.



lol havent people been reading the dev diaries where we have said we are prioritizing bug fixing and a bigger patch over DLC making? Maybe I should have been clearer, but I'm pretty sure I mentioned it in 3 whole diaries. atm we are putting something like 70% of our resources into patch/free update work, which I suspect most other companies would find utterly insane. We are doing it because we care ourselves, more than you guys probably. I heard a great quote from a designer roundtable at GDC the other day where a guy said "the only person more upset about a bug in the game they just ran into and are venting on the forum is the guys who are working their asses off on the game fixing that bug". I'm always chocked when people don't seem to realize this.

Balance in HOI is always going to fluctuate and you will always need to use sliders to boost nations to tailor your experience better. Most people play germany by a vast margin, so we will usually opt for making their experience better when you have a tradeoff. If germany beat soviet union every time, but only because we made soviets weaker thats not very fun when playing germany. Also I am not sure that things have changed that much on the macro level. germany loses slightly earlier now to the stronger soviets but they have been losing to soviets on average in every patch since release basically. To give you an idea on what we aim for "average outcome"-balance wise before releasing patches (this is minimum 20 test games or so):
- Germany should beat france and poland by historical date 100% of the time
- Japan should not lose to china by 1940 in 80% of games (usually ends up being 100% though)
- If germany or its allies are getting invaded sufficiently they should be getting bogged down and start losing in russia in a majority of games. If allies are not pushing them in some way (bombing or whatnot) they should be winning vs soviets in a majority of runs.

I think germany's biggest problem atm vs soviets is front shuffling because its stops their momentum. This however has been a problem since 1.0 basically so its no easy fix. We have tried lots of minor attempts at tweaking, but its become clear that it need some significant rewrites to solve the problems for good. Thats stuff we are working on for 1.4
For 1.3.3 we had to pick between making soviet more of a pushover for german/axis players or have german ai have a tougher time vs soviets and thus making the allied experience potentially worse. I think prioritizing germany was better. especially since sliders will help shift chances about.

An interesting insight into the balance process.

Now get back in that limo!