The great official AI thread

[janat08]

AI won't use continues national focuses in favour of not doing anything, with USA not being in war and not wanting to join and germany being strategic bombed by me with a very sizable que for repair. I will also happily bash against maginot line while my defences on italy and spain where inadequate. I only won because units on maginot wound up with something like 5 percent equipment. While spanish and italian border units were up to strength. Also in case you programmed for AI to get lax, or I just haven't played for too long, for when computer is slow, I do have slow computer.

PS also they never ever shifted air force around even when they got through on french border all of the axis continued to funnel airforce onto maginot.

 

[billcorr]

Reminder...

Please don't post feedback collected in 1.2.1 in this thread! We'd love to hear 1.3 feedback though.

And make sure you post detailed explanations on what aren't performing well.

To quote SteelVolt:
"Post your constructive AI criticism and feedback in this thread. Discuss what frustrates you about it in a civil manner, please

So, if you have something you want to lift in this thread, I would like you to include the following information:

• 
  
  • 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."

Non-constructive or outdated feedback will be removed to make this thread useful to us.

• 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.

 

[janat08]

They don't support the breakthroughs with air or even contest superiority, nor weigh other options as in if entrenched enemy would simply stand in way of germany at maginot with it not perusing any other fronts?

 

[joreri508]

While I agree to everything you said this is actually one of the problems. The AI don't allow their units to stay still long enough to build up good planning bonuses for them. They attack and keep attacking despite just atrociously destroying their equipment and manpower when they need to rest and build up equipment for an offense.

I know the AI will NEVER be as smart as a player, but I do think there is hope for improvement and I just point out what I think these improvements might make for good use given how the game works.

To introduce hierarchy, with armies commanded by general-level leaders and treating each army as a front which uses the battlefield planner, and having the armies organised into theatres, with a field marshall commanding each theatre, and having supply, unit allocation (depending on known threat level and VC importance) to the armies from the theatre level, and then having a super-level with overall allocation of units, replacements and supplies depending on strategic importance of theatre at the time, this could solve the problem. No matter how the code is written, hierarchy should be a plausible solution. @SteelVolt, are you listening? I would be happy to expand on this, if you like. I have time this week, back to work from 2/1, and then no time.

Regarding the AI smart or not, I does not need to be machine learning. It just needs to have the appropriate approach and structure for each problem at hand.

 

[Schatten51]

that actually happens.
not so many planes but at my invasion of USA i went up from Florida....i build the Airport here to lv 10 and suddenly there were like 30k Planes on it as all my allies, and i mean really all have moved their planes to this one airfield too, ignoring the Airports in Caribic that were close enough for Florida too.

One more bug in a endless list of bugs and wrong code....

 

[Dalwin]

1.3.1, NO TFV, only mod is No Mans Land.

AI sends troops far away from assigned front line, has them take a truly silly boat ride, then eventually sends them back toward the line.

The Russian troops taking the train to Odessa are assigned to a front line along the Polish border. There are no breaks or pockets in that line. They embark at Odessa and then after a short boat ride land at the port one province west of Odessa. They eventually make their way back north to their assigned positions. Save file attached.

Obviously the expected behavior is for troops to stay relatively near the line to which they are assigned, even if they do shuffle too much.

My best theory is that since the line extends all the way from the South of Poland to the Baltic. These troops are trying to shift from the southern part of the front to the northern. Instead of simply marching or taking a train to do so the try to take a boat. Only after their boat embarks in the Black Sea do they realize that the Turks will not let them pass the Bosporus so that admittedly foolish route is blocked, as it should be. They then abort the cruise at the nearest port and march back more or less to where they started.

 

[joreri508]

1.3.1, NO TFV, only mod is No Mans Land.

AI sends troops far away from assigned front line, has them take a truly silly boat ride, then eventually sends them back toward the line.

Yet another issue which would be solved by introducing resource allocation through hierarchy.

 

[Xuanzue]

played till 1941.

No real improvement in the AI. it stills stamps non-stop itself against a couple of guys in every port.

what must be doing? after the first 20 failed attempts of disembark it must try to use marines and attack with air superiority.

 

[Bambilambi]

Ai is still killing of Italian Army by transporting trough english channel....

Game mode:
Normal, historical game as UK with one extra strength point for Germany and Italy. The only mods active are graphical and music.

Issue:
(I tagged over to Italy)
And here we see a very determined Italian, trying to reinforce Ethiopia from Wilhemshaven, I am guessing non will even get past the channel... please fix. I call it a bug since I will kill to much of the Italian army this early. Because of this reason I also posted this one in the bug forum since stupid moves likes this breaks the vs AI game mode.

Swastikas removed - Secret Master

 

[FrancescoT]

@Bambilambi remove the swastika from the screenshot or remove the screenshot asap or I'll remove the whole post.

 

[Dracolithfiend]

Alright so I am not sure if this belongs here but in game right now I am playing France. I went Fascist (which was too easy IMO) and now everytime I start fabricating a claim on anything the British immediately guarantee it. Canceling does not give back the political power. This is ridiculous.

Edit: I just got Britain to guarantee Finland. . . .

 

[Xaelyn]

Alright so I am not sure if this belongs here but in game right now I am playing France. I went Fascist (which was too easy IMO) and now everytime I start fabricating a claim on anything the British immediately guarantee it. Canceling does not give back the political power. This is ridiculous.

Edit: I just got Britain to guarantee Finland. . . .

That's WAD. Why wouldn't Britain attempt to check the aggression of an expansionist France? That's been Britains national pastime for damn near a millennia. It costs them more political power to do it each time though (caps out at ~100 after a few), so you can bait them into guaranteeing countries you don't want to invade (or better yet, countries that a rival country will want to invade) and then fabricate and declare on your real target before they can get enough PP back. Super gamey, but it's an option. The downside is that if world tension is high enough (~50%?) the country might join the allies immediately even if not guaranteed.

EDIT: Also, they wont guarantee fascist or communist nations.

 

[Dracolithfiend]

It just feels incredibly cheap for them to be able to check every country on earth every single day to see if I have started fabricating a claim. Regardless things that game the AI and point out its flaws is pretty much exactly the purpose of this thread no?

 

[Xaelyn]

It just feels incredibly cheap for them to be able to check every country on earth every single day to see if I have started fabricating a claim. Regardless things that game the AI and point out its flaws is pretty much exactly the purpose of this thread no?

The difference is you are complaining that the AI is too good at something. I'd rather they didn't start sabotaging the AI in areas where it's better than a player (daily checks of many variables) until it's able to compete with the player in all other areas. Which won't be for decades at least.

 

[Dalwin]

It just feels incredibly cheap for them to be able to check every country on earth every single day to see if I have started fabricating a claim. Regardless things that game the AI and point out its flaws is pretty much exactly the purpose of this thread no?

If you are ever in the position of wanting to do the same yourself, don't check every country. In fact, don't check the targets at all. Check the aggressor nations only. A quick glance at German, Italy, or in your case France, will show who they are targeting.

 

[joreri508]

Alright so I am not sure if this belongs here but in game right now I am playing France. I went Fascist (which was too easy IMO) and now everytime I start fabricating a claim on anything the British immediately guarantee it. Canceling does not give back the political power. This is ridiculous.

Edit: I just got Britain to guarantee Finland. . . .

Why ridiculous? Britains behaviour is the tit-for-tat game which went on between great powers for centuries. As Clausewitz said, "War is merely the continuation of policy by other means". Before war there are foreign policy games, guarantees, sanctions...

Look at the world right now. Earlier this December, the Obama-administration eased sanctions to provide Syrian rebels with arms. Today in the morning, the Russian foreign office makes a strong remark about this. Majors expending political points, in real life.

 

[Hammer54]

Nice thread!


I’ve had some experiences with a recent game as Germany. Overall I think the game is doing quite well as long as you stay on a fairly historical path. I like the state of navies in HOI4. To the topic. I think the AI should be a little more stressed about strategic encirclement. On the Russian front, where there is a lot of room to manoeuvre, you can easily run panzers through the front and behind. This was done in real life, but considering the available forces in the area, they should make harder attempts to break out. If encircled, or they are about to be encircled, they should probably stop whatever they are doing, to try and stop the encirclement (As Japan, I’ve seen india doing some pretty amazing escaping from strategic encirclement in Bengal). This problem doesn’t show in smaller theatres, as troops seem to want to escape when its just one or a few provinces involved.

The AI doesn’t seem to use fallback lines (rivers are great places to fall back to).

Russians seem to have little pazers in the mix.

There is also an issue with easy pickings. Frontline troops should be more inclined to capture empty or lightly defended areas when they are standing right next door (could maybe be a switch for the armygroup, to prevent troops from wandering).

I’ve seen puppets and allies doing some great naval invasions, but they aren’t very good at following through with more troops when they capture a harbour. They don’t take airpower seriously either.

Germany never conquers Norway as AI (problems doesn’t seem to be with the invasion AI. It just doesn’t happen.)

AI doesn’t seem to be using panzers properly. In my experience, the AI place troops/panzers all along the front, and tries to win the grind. Separating panzers and mobile units in fast armygroups to pierce the front doesn’t seem to be in their toolkit. (To this effect, I wouldn’t mind some breakthrough tools for less micro as a player. If breaktrough was introduced as a concept, where you could assign an armygroup to “fill the void” behind the breakthrough. And the armygroup which is making the breakthrough have an option to set the planner to a single province path.)

After loosing lots of units (for example in strategic encirclements) AI is slow to respond with underequipped and n00b armies (better than nothing.)

Invading England, AI doesn’t seem to panic by sending troops home. A small armygroup can take its time taking the isle. (Fleets doesn’t seem able to prevent supplies from coming through). AI allies didn’t seem to be hindered by enemy fleets.

The thing about the historical war, was that there were reasons for going where they did. Winning the war in North Africa, while not taking Irak for oil, seems pointless.

I hate it when Poland yields Danzig. There should be an option to attack anyway. I seem to remember Hitler dint believe in DOW.



Off topic


I which my airpower could bring back some intelligence from behind the front (general suggestion)

 

[Axe99]

After playing with some of the smaller CW countries with TfV, I've spent a bit more time watching ENG's behaviour, and what comes to mind is that they've a bit too easy to start new fronts (usually through invasion) when the current fronts aren't stabilised - or, even worse, take a stable front and remove large numbers of divisions to support an invasion (often of dubious merit), causing issues for both fronts.

• What the AI is doing wrong
  • Starting new fronts when there aren't sufficient troops to support (current fronts + new fronts).
• What you think it should be doing instead.
  • Before planning an invasion or offensive, the AI should assess whether it has sufficient units to achieve that goal without weakening existing fronts. If the AI can make progress with an existing front, it's probably better for them to concentrate their force their (up to supply limits) in terms of them getting results, than starting new ones (unless presented by large natural barrier/line of fortresses).

Some thoughts on how to go about this, in case useful (noting they might be useless, depends a lot on how things are sorted in the engine):

- Once the AI has sufficiently stabilised its defences and is ready to go on the offensive (or has sufficient troops to support its current offensives), have the AI pick a goal (invade Sicily/Madagascar/Norway) and then build up troops for this, not actually undertaking the offensive/invasion until sufficient forces are ready. If this can be linked to DoWs, it could do things like making sure Barbarossa isn't launched until the AI assesses it has sufficient troops.

- Defensive needs should generally trump offensive operations. The AI shouldn't invade Sicily if Japan is pushing down into Malaya.

- If there are existing land fronts without substantial natural barriers, it's likely these are better bets in terms of offensive choices than starting new invasions.

 

[Dalwin]

After playing with some of the smaller CW countries with TfV, I've spent a bit more time watching ENG's behaviour, and what comes to mind is that they've a bit too easy to start new fronts (usually through invasion) when the current fronts aren't stabilised - or, even worse, take a stable front and remove large numbers of divisions to support an invasion (often of dubious merit), causing issues for both fronts.

• What the AI is doing wrong
  • Starting new fronts when there aren't sufficient troops to support (current fronts + new fronts).
• What you think it should be doing instead.
  • Before planning an invasion or offensive, the AI should assess whether it has sufficient units to achieve that goal without weakening existing fronts. If the AI can make progress with an existing front, it's probably better for them to concentrate their force their (up to supply limits) in terms of them getting results, than starting new ones (unless presented by large natural barrier/line of fortresses).

Some thoughts on how to go about this, in case useful (noting they might be useless, depends a lot on how things are sorted in the engine):

- Once the AI has sufficiently stabilised its defences and is ready to go on the offensive (or has sufficient troops to support its current offensives), have the AI pick a goal (invade Sicily/Madagascar/Norway) and then build up troops for this, not actually undertaking the offensive/invasion until sufficient forces are ready. If this can be linked to DoWs, it could do things like making sure Barbarossa isn't launched until the AI assesses it has sufficient troops.

- Defensive needs should generally trump offensive operations. The AI shouldn't invade Sicily if Japan is pushing down into Malaya.

- If there are existing land fronts without substantial natural barriers, it's likely these are better bets in terms of offensive choices than starting new invasions.

I am certain that those same points would be valid for Italy and a few others and not simply the UK. You are absolutely right, however.

 

[Axe99]

I am certain that those same points would be valid for Italy and a few others and not simply the UK. You are absolutely right, however.

Aye, I think the general principles could probably be applied to any nation (but particularly those that do a lot of invasions), but the particular observations were in relation to ENG.