AI: The "scareed chickens heap"-Defense

[demon72]

With a certain degree of sarcasm, one could claim that the AI is acting historically, but in my current game I have noticed a problem that undermines any reasonable defence of the AI.

To make it understandable and reproducible for everyone, I would like to describe the problem with the smallest possible example:

• take 4 identical units of Paras and drop them in 2 adjacent regions (A and B) so that 2 land in each region.
• since this phase is controlled by the Battleplan AI, the subsequent behaviour is extremely telling.
• one unit stops in each region, but one unit runs from A to B, while the other unit runs from B to A.
• these movement commands make absolutely no sense, but explain the subsequent problem very well.

In the current game, as DR, I attacked the US via Mexico from the south, with troops facing each other along the border. Due to the significantly worse infrastructure and the resulting lack of supplies, my troops were significantly outnumbered (about 1:3) and I started to build up defence positions to wait for the infra to build up.

For no apparent reason, the AI suddenly withdraws all units on the east coast (Texas) in two border regions, although some of my armoured divisions were located there. Guderian could hardly believe his luck and advanced north without hesitation.

What happened next was almost more unbelievable: the USA moved about 90% of its divisions by rail, the well-fortified front in the direction of Mexico collapsed completely within days, and from then on only divisions with minimal ORG wandered pointlessly across the map.


After the inevitable surrender that followed, I then tried to protect the US coasts via the new garrisons/defence orders. In order not to overtax the AI, I placed identical w20-INF divisions in each port, grouped them under a general and then marked all coastal regions only with the order to defend the ports - what happened?

...suddenly two identical divisions start swapping places. Why? This question could already be asked in the example with the PARAs. To make matters worse, the divisions were not in two adjacent regions but one on the east coast and the other on the west coast. Somehow this movement triggered a chain reaction and suddenly almost all the divisions assigned to the port defence were on the move - on foot, by rail and sometimes even by ship. After each division had taken its personal favourite place, I actually thought everyone was happy now - or so I thought!


To solve the supply problem, I had started to expand a Mexican port on L10 before the US troops self-dissolved. Of course, this takes time, but by expanding the port, it suddenly became so important for the AI that it would be better to guard it with 2 units (but there was just one Unit for each piort available).

Not surprisingly, this now triggers the next chain reaction, almost all divisions are on the move and in the end the only port that really needed some defence was unguarded.


Even though I didn't realise the connection at this point, during the subsequent attack on the SU, something similar happened after my tanks broke through.

Only suddenly a large part of the Russian divisions were marked with a small shield! Exactly this small shield, which also indicated with my divisions that these units were assigned to a garrison/defence command. Why I could see this as an opponent - I don't know, but if the AI uses this completely bugged command for defence, it simply cannot work!

The result of Barbarossa in terms of losses was then logical: 60k (DR) vs. 10.87m (SU) and this does not even include the 200+ encircled divisions that vanished into thin air at the surrender!

It may be historically true that every army has struggled against aggressive and highly mobile attacks in WWII, but if the defenders can no longer get out of the status of "scareed chickens heap" and do not manage to build up a new defensive line even after a breakthrough, then it becomes highly pointless.


The 1st step should be to analyse why the AI orders the PARAs to swap places for no reason!

 

[GrandVezir]

The 1st step should be to analyse why the AI orders the PARAs to swap places for no reason!

I can tell you the in-game logic: a landing para force creates its own front. Since the landings at A and B are adjacent, each front widens to cover both tiles, and each pair of paras (heh) splits to cover both tiles of its own front.

 

[Harin]

I know it makes for a good laugh when a player with hundreds of playing hours calls the game "unplayable". Unfortunately, the game has so much going on, much of it not obviously wrong, that it can take a long time to realize that some of the previous fun was due to not knowing about these type flaws. A new player spends hours on his game, he is having a tough, but fun time of it, then suddenly, he gets that break through and wins. Hoorah, New Guy! It might be hundreds of game hours later before he realizes that he may not have been winning as much as the AI lost its silicone mind.

I do not care if people want alt-history, crazy-history, plausible history, or railroaded history. I just want the developers to fix it

 

[Roland Traveler]

Quick question, were the Soviets garrisoning a state that you partially control? If so, it’s well known that garrisoned divisions will charge any enemy that steps into their state, even if it’s suicide. It’s for that reason (and the AI deciding to nonsensically shuffle troops like you described) that I prefer to delete the order once everybody’s in position.

 

[demon72]

I can tell you the in-game logic: a landing para force creates its own front. Since the landings at A and B are adjacent, each front widens to cover both tiles, and each pair of paras (heh) splits to cover both tiles of its own front.

but just swapping places doesn't make any sense, it just keeps Org lower as it could be.

Also that two (or even more) identical units are swapping places in the port-defence example doesn't make any sense.

This gets even worse, when the AI checks every game-hour if there is a need to "optimize" defense positions, as this creates chain-reactions!


That the AI tends to move to much is a known problem since HOI I IIRC, but as the AI seems to use the new guard-orders, things get a little stupid!

 

[demon72]

I know it makes for a good laugh when a player with hundreds of playing hours calls the game "unplayable". Unfortunately, the game has so much going on, much of it not obviously wrong, that it can take a long time to realize that some of the previous fun was due to not knowing about these type flaws. A new player spends hours on his game, he is having a tough, but fun time of it, then suddenly, he gets that break through and wins. Hoorah, New Guy! It might be hundreds of game hours later before he realizes that he may not have been winning as much as the AI lost its silicone mind.

I do not care if people want alt-history, crazy-history, plausible history, or railroaded history. I just want the developers to fix it

I was close to not using the gap the US-Forces offered me, but I was to curios to see what's happened.

But it's exactly as you say, at first you may be briefly happy about your supposedly "good strategy", but in the end the victory just has a bad taste, because it was given to you by the AI.

 

[demon72]

Quick question, were the Soviets garrisoning a state that you partially control? If so, it’s well known that garrisoned divisions will charge any enemy that steps into their state, even if it’s suicide. It’s for that reason (and the AI deciding to nonsensically shuffle troops like you described) that I prefer to delete the order once everybody’s in position.

How can I know this?

There were about 500+ divisions in movement - almost all Divisions my tanks met had very low ORG. Many of these Divisions were marked by that little shield - I not even understand why I got this information.

 

[demon72]

From my point of view the AI needs to learn:

1. not to weaken the main frontlines!
2. reduce move orders (especially to Divisions with a guard order) to a minimum
3. keep a sufficient and highly mobile reserve in a second line to contain breakthroughs or counter PARA-Drops or unexpected Naval-Invasions

 

[Ololorium]

The AI is way too eager to move (or act in general). It seems to be intentional: I remember one of the devs replying to the question about AI suicide attacks with something like "well, it's impossible to create a perfect AI, so we had a realistic choice between making it overly cautious and passive and making it overly aggressive and suicidal, and we went with the second one because the first type of AI would be less fun to play against". It kind of makes sense, but only in the context of aggressive actions, while this philosophy "it's better if the AI does something rather then nothing" seems to be applied everywhere, and when unit movement is concerned, it really makes the game less fun and more frustrating. The AI doesn't seem to realize that movement has costs: fuel, entrenchment, leaving provinces vulnerable for a time. Too often I observe the situation where divisions from two opposite edges of my frontline swap places for no reason. It should NEVER happen: why would a division left its place and go far away just to take place of similar division? I wish we had frontline settings similar to the attack order settings (which determine how aggressively the divisions should push), to determine how actively divisions would move along the frontline, with the lowest setting potentially making frontline less optimal, but removing chaotic shuffling we have now.

 

[Roland Traveler]

How can I know this?

There were about 500+ divisions in movement - almost all Divisions my tanks met had very low ORG. Many of these Divisions were marked by that little shield - I not even understand why I got this information.

If a small number of divisions keep on attacking only one or two provinces in your front line and they’re in a state the AI controls but doesn’t have full control of, I’d say it’s a safe bet that the divisions are on garrison order.

 

[demon72]

The AI is way too eager to move (or act in general). It seems to be intentional: I remember one of the devs replying to the question about AI suicide attacks with something like "well, it's impossible to create a perfect AI, so we had a realistic choice between making it overly cautious and passive and making it overly aggressive and suicidal, and we went with the second one because the first type of AI would be less fun to play against". It kind of makes sense, but only in the context of aggressive actions, while this philosophy "it's better if the AI does something rather then nothing" seems to be applied everywhere, and when unit movement is concerned, it really makes the game less fun and more frustrating. The AI doesn't seem to realize that movement has costs: fuel, entrenchment, leaving provinces vulnerable for a time. Too often I observe the situation where divisions from two opposite edges of my frontline swap places for no reason. It should NEVER happen: why would a division left its place and go far away just to take place of similar division? I wish we had frontline settings similar to the attack order settings (which determine how aggressively the divisions should push), to determine how actively divisions would move along the frontline, with the lowest setting potentially making frontline less optimal, but removing chaotic shuffling we have now.

For most smaller nations the "AS-IS"-situation works "somehow" - but definitly not for nations with very large territories like US and SU!


As a rough guideline the AI needs to learn to build 2/3 INF (or other cheap low-mobility Divisions) and 1/3 (Tanks and/or mobile INF)

As long as a nation is in any defensive or passive situation, the AI needs to learn, that if just deploys the 2/3 low mobility Divisions on the frontlines with a very passive-setting concerning "moves"!

Next it has to learn to use the 1/3 divisions as "mobile reserve", for these units the setting should be more active (but still avoiding transports by train or ship besides peace-times)

EDIT: maybe a movement-limit should be also implemented for both unit types at war-times


Shouldn't be that hard to fix!

 

[Sbrubbles]

I wonder if at least part of this problem would be solved if the AI (both battle planner and garisson AI) used rail-move more often. I get that it often uses regular move so that it doesn't arrive without org at the front, but the ammount of time wasted is gigantic and as player I'm having to manually use rail-move way more than the AI.

 

[elektrizikekswerk]

but just swapping places doesn't make any sense, it just keeps Org lower as it could be.

That's true in core - and in regards to your port guarding example accurate. But in this para case you are expecting too much, imho.

First rule: Frontlines expand to neighboring provinces. This is so to prevent gaps. It sometimes makes you (as a player) furious, especially when the neighboring province is already secured by another frontline. To remove this problem we need the ability to define strict divisional/army borders of operation that a unit may not cross under any circumstances. But this may result in other problems and overstretching frontlines which is - to a degree - prevented by the current system.
Second rule: Each frontline tries to cover its assigned area as good as possible.

The combination of these two rules results in the behaviour of the paras.
Each para order created an individual frontline for province A and B, respectively.
The landing of the other paras resulted in the expansion of both frontlines.
Both frontlines' assigned units try now to cover their front as good as possible. As both frontlines cover both provinces both orders' units will try to cover both provinces.
Without having the involved frontlines in mind thsi indeed makes no sense at all. With the frotlines it does.
To prevent this we needed either the possibility for several front line orders to "talk" to each other (which, I bet, would go wrong in so many other situations), or we need the tool of declaring strict borders of operation as stated above - which also somes with its own problems.

 

[demon72]

That's true in core - and in regards to your port guarding example accurate. But in this para case you are expecting too much, imho.

First rule: Frontlines expand to neighboring provinces. This is so to prevent gaps. It sometimes makes you (as a player) furious, especially when the neighboring province is already secured by another frontline. To remove this problem we need the ability to define strict divisional/army borders of operation that a unit may not cross under any circumstances. But this may result in other problems and overstretching frontlines which is - to a degree - prevented by the current system.
Second rule: Each frontline tries to cover its assigned area as good as possible.

The combination of these two rules results in the behaviour of the paras.
Each para order created an individual frontline for province A and B, respectively.
The landing of the other paras resulted in the expansion of both frontlines.
Both frontlines' assigned units try now to cover their front as good as possible. As both frontlines cover both provinces both orders' units will try to cover both provinces.
Without having the involved frontlines in mind thsi indeed makes no sense at all. With the frotlines it does.
To prevent this we needed either the possibility for several front line orders to "talk" to each other (which, I bet, would go wrong in so many other situations), or we need the tool of declaring strict borders of operation as stated above - which also somes with its own problems.

wouldn't be a small check every game-hour if units are just "swapping places" be enough to solve this?

or something like "don't expand frontlines in provinces with friendly units"?

 

[elektrizikekswerk]

wouldn't be a small check every game-hour if units are just "swapping places" be enough to solve this?

Only if you allowed/enabled different front lines to talk to each other. Which I think would go wrong in many other situations. I think in this regard that the current situation is the lesser of two evil.

or something like "don't expand frontlines in provinces with friendly units"?

Only if it was optional - sometimes (most of the times, in fact) you want frontlines to overlap.

 

[Sbrubbles]

Only if it was optional - sometimes (most of the times, in fact) you want frontlines to overlap.

I've found it a lot easier to organize front lines with field marshalls and not overlapping (single-general) fronts. The UI is fiddly, but it works and it's also a way to get planning bonus with unassigned units, which is great for microing tanks.

Generally speaking, I only get overlapping is with naval invasions and paradrops

 

[elektrizikekswerk]

I've found it a lot easier to organize front lines with field marshalls and not overlapping (single-general) fronts.

I see your point but I like army-specific front lines more. In fact I often even have several frontlines per army. It makes it easier for me to tell my units exactly where I want them (short of using direct orders which then are getting overwritten by frontlines).

My personal experience with fieldmarshal frontlines is that for some reason 20th INF Division wants to swap places with 34th INF Division.
Sure, both divisions are part of an siege. Only one is in Leningrad while the other is in Sewastopol...

 

[GrandVezir]

but just swapping places doesn't make any sense

Computer logic doesn't always make sense to people.

Force A has 2 divisions, and 2 tiles to cover. The only logical way to cover 2 tiles with 2 divisions is to split them. Force B has 2 divisions, and 2 tiles to cover. The only logical way to cover 2 tiles with 2 divisions is to split them. The AI doesn't care that A and B are covering the same two tiles. It only recognizes that each force has a front, and must cover that front with the troops it has.

The only way* to fix that problem is for Paradox to develop AI that can perform pattern recognition at near human levels, so it can recognize the front overlap and adjust accordingly, either by shortening or merging the fronts. I wouldn't hold my breath on that one.

*I suppose PDX could also remove the frontline system entirely, but that would basically be rewriting the entire land combat system from scratch.

 

[elektrizikekswerk]

Computer logic doesn't always make sense to people.

Force A has 2 divisions, and 2 tiles to cover. The only logical way to cover 2 tiles with 2 divisions is to split them. Force B has 2 divisions, and 2 tiles to cover. The only logical way to cover 2 tiles with 2 divisions is to split them. The AI doesn't care that A and B are covering the same two tiles. It only recognizes that each force has a front, and must cover that front with the troops it has.

The only way* to fix that problem is for Paradox to develop AI that can perform pattern recognition at near human levels, so it can recognize the front overlap and adjust accordingly, either by shortening or merging the fronts. I wouldn't hold my breath on that one.

*I suppose PDX could also remove the frontline system entirely, but that would basically be rewriting the entire land combat system from scratch.

That's an even better way to explain it than I did.

 

[HugsAndSnuggles]

I know it makes for a good laugh when a player with hundreds of playing hours calls the game "unplayable".

Is it, really? When said hours are required to understand the very basics of the game?