AI observations, normal difficulty, November 1940

[Gort11]

I've recently been playing a game as historical France and figured I'd tag around the other major nations and see how the AI has been handling them. This is in November 1940, and I've held off Italy and Germany, and kicked them both out of Africa. Here's what struck me when I tagged around the various majors:

Everybody:

There are lots of units using old, decommissioned templates. The AI armies are very messy in this regard.

Everyone has researched 1940-era armoured cars, but only Germany is building any, and even then not many.

Very little focus on tanks in favour of lots of infantry and artillery.

-----

USA:

They have TONS of dockyards - they're up to 67 from their starting 22, and are building more! However, they're building lots of "early" cruisers and destroyers, which is pretty wasteful given that they have 1936-era hulls researched.

They're running a deficit of 11 rubber and 26 steel despite having huge numbers of civilian factories and being at peace.

They have a weird mixed light/medium tank template, but no medium tank production, so all they actually have is infantry divisions.

-----

UK:

They've garrisoned every little rock in the British Empire. They have 76 divisions spread across 23 different armies, guarding worthless things like Ascension Island which the Royal Navy can guard just fine.

Again, TONS of dockyards. They're up to 49 from their starting 19, and are building more. Again, they're building early destroyers and submarines despite having 1936 destroyer tech from day 1.

Tiny focus on tanks, with only two factories assigned to inter-war mediums.

Some focus on strategic bombers, but all 150 are in storage instead of bombing Germany.

-----

Germany:

For some reason they're building lots of civilian factories here in the middle of the war.

They've built a small number of dockyards. They started with 10, captured 14, and built 6 themselves.

All ships under construction are 1936 models, weird that the least naval major nation seems to be the smartest about building ships.

They LOVE infantry gear. They have 36 factories making it, and a surplus of 136,000 units. They have deficits in AA guns, artillery, trucks and medium tanks that they could be building instead.

They've ignored the synthetic rubber techs, despite having synthetic refineries and importing rubber at great risk from Brazil and Siam.

-----

USSR:

Some focus on dockyards, starting with 6, getting 2 from focuses, and building 8 more manually.

The same problem with infantry gear as Germany - 32 factories making it, a surplus of 138,000 units, despite deficits in AA, AT and medium tanks.

Some tank production - 11 on light tanks, 3 on mediums.

Their tank template is weird, with light and medium tanks in the same division.

-----

Italy:

Again, too much infantry gear, surplus 94,000 units with 16 factories on it.

Too many trucks, surplus of 6,000 units with 9 factories on it.

Very little focus on dockyards, having only built one extra.

Has every synthetic oil and rubber tech, and is researching synthetic oil ahead of time, but has no refineries! They have ONE scheduled to be built, but it's way down the list of priorities.

-----

Japan:

A fair focus on naval dockyards, up to 31 from their starting 14. They're still building early subs, cruisers and destroyers, and running steel and rubber deficits for no reason.

-----

Too long, didn't read:

1. The UK, USA, Japan and to a lesser extent Germany and the USSR build too many dockyards when they should be building military factories.
2. The AI builds "early" ships when it has better ships available.
3. The AI can be bad at trade, running resource deficits when it doesn't have to.
4. The AI focuses researching armoured cars and then doesn't build many, or in many cases any at all.
5. Germany and Italy overbuild infantry equipment.
6. Few tanks get built, and they often end up in weird mixed light/medium tank divisions.

 

[Beagá]

Good lord. This version still is a beta then...

 

[billcorr]

Too long, didn't read:

1. The UK, USA, Japan and to a lesser extent Germany and the USSR build too many dockyards when they should be building military factories.
2. The AI builds "early" ships when it has better ships available.
3. The AI can be bad at trade, running resource deficits when it doesn't have to.
4. The AI focuses researching armoured cars and then doesn't build many, or in many cases any at all.
5. Germany and Italy overbuild infantry equipment.
6. Few tanks get built, and they often end up in weird mixed light/medium tank divisions.

Good analysis.

Thank you for taking the time to tag-switch, analyze, and publish your findings.

You've identified some things that could make the game more fun.

(and perhaps help increase $ale$!)

 

[nuarbnellaffej]

I haven’t tagged switched, but in my own German campaigns I have found the Soviets to be very tank lite. Armored divisions are far and few between, and when I do encounter them, their hard attack/soft attack, defense, and breakthrough values in battle are utterly pathetic compared to my own panzer divisions.

Even my motorized divisions with some light TD/SPG can melt through them pretty reliably.

I might try the expert AI mod to have at least some challenge/resistance.

 

[SeekTruthFromFx]

@Gort11, were you playing with NSB enabled? That might have a big effect on tank usage.

But whatever constellation of DLC you had, that's really disappointing reading. Fixing these issues doesn't require programmers to rewrite code; it doesn't even require particularly good maths skills These are things that might be improved in script by Content Designers (with QA if available) iterating through different values in the strategy files.

 

[Northernwwater]

Anyone know how the Expert AI mod handles this?

 

[Koba133]

it's about time paradox does just a litte bit of effort to improve Hoi4's AI, like they are doing now for Stellaris and EU4

 

[Jays298]

Hey I'm using weird mixed light medium divisions! I'm not sure if they perform worse but from a logistics perspective of having to produce all that, I like 2 light, 1 medium, 1 medium destroyer. Gets tanks in the field faster.

USA , devs need to add 1936 naval designs for stuff they have researched in 36. Not hard for the player to spend 30xp and mess around. Much harder for the AI.

 

[Gort11]

@Gort11, were you playing with NSB enabled? That might have a big effect on tank usage.

Yeah.

 

[SeekTruthFromFx]

Hey I'm using weird mixed light medium divisions! I'm not sure if they perform worse but from a logistics perspective of having to produce all that, I like 2 light, 1 medium, 1 medium destroyer. Gets tanks in the field faster.

I am not much of a minmaxer and I do this too for a transitional period. But normally I want to keep my light tanks separate for their speed bonus. They are not just poor-quality medium tanks; they have a role of their own. And I thought the division template scripting was designed around allocating appropriate equipment to divisions with different roles.

 

[Arheo]

There are lots of units using old, decommissioned templates. The AI armies are very messy in this regard.

The AI does not act like you might expect a human player to, as regards templates. We have a backlog task to improve it for observer use, but don't read too much (ie, anything) into what it decommissions.

USA:

They have TONS of dockyards - they're up to 67 from their starting 22, and are building more! However, they're building lots of "early" cruisers and destroyers, which is pretty wasteful given that they have 1936-era hulls researched.

The naval AI does some pretty deep consideration of what it thinks is a 'good' ship - this is heavily economy driven. The recently rolled back issue causing the AI to cancel historical ships was caused by the AI deciding that these historical ships were not a good use of time and money.

They have a weird mixed light/medium tank template, but no medium tank production, so all they actually have is infantry divisions.

This sometimes happens when they're 'in between' upgrading to armor templates. it tends to resolve itself in the early war years, but can sometimes stick if conditions are not good.

Again, TONS of dockyards. They're up to 49 from their starting 19, and are building more. Again, they're building early destroyers and submarines despite having 1936 destroyer tech from day 1.

I'll look into the dockyards, though they do try and balance things out to a degree. The issue might be more that the order in which the AI builds factory types is at fault.

Germany:

They LOVE infantry gear. They have 36 factories making it, and a surplus of 136,000 units. They have deficits in AA guns, artillery, trucks and medium tanks that they could be building instead.

This one is not so odd. As you've pointed out elsewhere, the AI loves infantry - they also make heavy use of battleplan pushes (the merits of this vs 'micro', and the ability of the AI to perform manual encirclements is valid, but to say this would be a big deal to change would be an understatement). If things play out historically, you'll see Germany/Italy eat a lot of that infantry equipment holding the line and getting involved on the Eastern front.

USSR:

Some tank production - 11 on light tanks, 3 on mediums.

Their tank template is weird, with light and medium tanks in the same division.

As above, the AI upgrades their templates sequentially, so they often end up between templates. This usually resolves itself when they feel they have enough XP to burn.

-----

Italy:

Has every synthetic oil and rubber tech, and is researching synthetic oil ahead of time, but has no refineries! They have ONE scheduled to be built, but it's way down the list of priorities.

That sounds odd.

Overall some good observations, and the points I haven't mentioned are probably logged for us somewhere or in need of some attention. As you might imagine, working on the AI for a game like HoI is an ongoing challenge, and even small changes tend to have far-reaching and hidden effects on how the AI plays.

 

[Xinamon]

The AI does not act like you might expect a human player to, as regards templates. We have a backlog task to improve it for observer use, but don't read too much (ie, anything) into what it decommissions.



The naval AI does some pretty deep consideration of what it thinks is a 'good' ship - this is heavily economy driven. The recently rolled back issue causing the AI to cancel historical ships was caused by the AI deciding that these historical ships were not a good use of time and money.



This sometimes happens when they're 'in between' upgrading to armor templates. it tends to resolve itself in the early war years, but can sometimes stick if conditions are not good.



I'll look into the dockyards, though they do try and balance things out to a degree. The issue might be more that the order in which the AI builds factory types is at fault.



This one is not so odd. As you've pointed out elsewhere, the AI loves infantry - they also make heavy use of battleplan pushes (the merits of this vs 'micro', and the ability of the AI to perform manual encirclements is valid, but to say this would be a big deal to change would be an understatement). If things play out historically, you'll see Germany/Italy eat a lot of that infantry equipment holding the line and getting involved on the Eastern front.



As above, the AI upgrades their templates sequentially, so they often end up between templates. This usually resolves itself when they feel they have enough XP to burn.



That sounds odd.

Overall some good observations, and the points I haven't mentioned are probably logged for us somewhere or in need of some attention. As you might imagine, working on the AI for a game like HoI is an ongoing challenge, and even small changes tend to have far-reaching and hidden effects on how the AI plays.

It would be great to see some work being done on the ai in later patches

 

[Jafkka]

Overall some good observations, and the points I haven't mentioned are probably logged for us somewhere or in need of some attention. As you might imagine, working on the AI for a game like HoI is an ongoing challenge, and even small changes tend to have far-reaching and hidden effects on how the AI plays.

It’s really great that the game’s director regularly reads comments and feedback (whether useful or otherwise) from fans.

No doubt it’s tough logging in and finding a kernel of useful feedback sandwiched between two posts titled “Nationalists always win SCW - FIX THIS!” and “Why do Republicans always win SCW?!”

Though I may not personally like specific approaches here or there on DLCs, the collaborative relationship between the fans and designers is what really make PDX stand out as a company.

 

[Lamartine]

2. The AI builds "early" ships when it has better ships available.

If you hang out on Reddit you'll hear from seasoned players that roach destroyers are the meta. You need LOTS of cheap screens to protect the capital ships. Early Destroyer hull costs 90 compared to 120 for 1936 hulls.

If the AI was building ASW destroyers or spotter cruisers with those hulls then an argument could be made against that, but later ships =/= necessarily better ships in terms of the role they play in the navy.

 

[sekelsenmat]

There are lots of units using old, decommissioned templates. The AI armies are very messy in this regard.

Everyone has researched 1940-era armoured cars, but only Germany is building any, and even then not many.
.....
Japan:
A fair focus on naval dockyards, up to 31 from their starting 14. They're still building early subs, cruisers and destroyers, and running steel and rubber deficits for no reason.

The AI template mess does not indicate anything wrong, it might be an efficient way for an AI to deal with templates in fact. Even those with mixed tank types, since the AI doesn't "micro" battles, it is of little relevance that mediums are mixed with lights.

Early destroyers aren't useless, they are more IC efficient at escorting against subs

But early subs I wouldn't build ever, they die like flies. Even 36 subs are really bad. They are maybe even ahistorically weak, I'm not sure aside from snorkel/radar that newer subs really had a huge advantage in hiding themselves like in the game. And of course you could build big subs in ww1 and small ones in ww2, I dislike the idea that newer=bigger.

I also wouldn't ever build early cruisers due to being too slow, but here I'm not sure the difference in performance is so big.

Hard to believe that Germany won't research rubber, I have a hard time covering my rubber needs despite a huge focus on that when I play it.

 

[jpd]

Germany has an NF that adds a +2 rubber to all your synth plants. That's the equivalent of two rubber techs. Removing the need to go after these techs very aggressively.

 

[Gort11]

I don't think this is the naval AI following a strategy of cheap destroyers to act as screens.

The UK has spent a fair bit of naval XP creating variants of the 1936 and early destroyer hull. You can see in the screenshot that it has a close-to-optimal anti-submarine destroyer designed. However, it's devoting most of its effort to building V/W Class destroyers, which are early hulls of a design that has been available since day 1. (level 1 everything, plus an extra level 1 depth charge and level 1 AA) If it was following the "cheap screen destroyer" strategy it wouldn't have the depth charge, and probably not the AA either.

The US is spending a lot of effort making these... things... which aren't really much good as screens or as ASW.

This behaviour doesn't seem to be AI-wide. As I noted in the original post, every ship Germany is building is a 1936 hull, so the naval AI doesn't always build out-of-date ships. You'd have to presume that the design intention of the game designers is that the game loop goes "invest in tech -> get better things to build -> build those things" instead of "invest in tech -> get less efficient things to build -> build the things you've had access to since day 1".

It's also a bigger problem for the AI than just ending up with bad ships. The US and UK are putting enormous amounts of production into dockyards which could have gone to military factories, and the US has a steel shortage due in part to the large amounts being used building ships. If the returns the US and UK get from all this investment are bad, the AIs for those countries will also be bad.

 

[Gort11]

Figure I might as well upload the save game so others can dig into it and draw their own conclusions.

 

[Gort11]

Germany has an NF that adds a +2 rubber to all your synth plants. That's the equivalent of two rubber techs. Removing the need to go after these techs very aggressively.

The Germany in this save is in a rubber deficit and importing from Siam and Brazil by sea, though. They could certainly benefit from the techs.

 

[Gort11]

The AI template mess does not indicate anything wrong, it might be an efficient way for an AI to deal with templates in fact. Even those with mixed tank types, since the AI doesn't "micro" battles, it is of little relevance that mediums are mixed with lights.

The US can't deploy a tank division since their template has a mix of light and medium tank battalions, and they have no medium tanks or medium tank production. If it was a straight light tank template, it would be deployed already.