AI won't fight their own war

[PDXOxycoon]

Ok, so. This particular problem stems from the fact that the fact that the player is an ally in the war. I would like to emphasize that this is the current implementation.

So, the issue arises from how the War Coordinator decides which War Stance it is in. In most circumstances it does exactly what it needs to do, but the player throws a bit of a complication into this. The War Coordinator chooses the Offensive stance if they outnumber the opposing side, and the Defensive stance if they are being outnumbered. The player(s) are, from the war coordinator's perspective, extremely unreliable allies. The player(s) does what the player(s) wants, and the war coordinator has no way of knowing what your intentions are. Because of this the player's strength is not counted for the army strength on the side they are on, it is extremely pessimistic here. On the other hand, the other side is counting on the player's strength when calculating the strength of the opposing side, because it is pessimistic on its odds and never knows what the other side is up to.

So what we then have is:

	Attacking War Coordinator Sees	Defending War Coordinator Sees
AI Attacker	2610	2610
AI Defender	2941	2941
Player Attacker		1167

This pessimistic look means: the attacker sees 2610 strength on their side and 2941 strength on the defender side; the defender sees 3777 attackers and 2941 defenders.

The attacking war coordinator only see the units it is in control of, so it thinks it is outnumbered and is thus in a defensive stance. The defending war coordinator calculates the full strength of the opposing side, regardless of which opposing war coordinator is in charge of it. So we are now in a situation where both sides are in a defensive war stance, which is why the attacker is holding back in this war.

This behaviour is something we want look into at some point in the future, but I cannot give a timeline to give as to when.

I hope this has clarified it.


Now to address some of the other points in the thread.

There's been a thread on this already but from experience:
1. If the player is the primary attacker/defender, then the AI will follow the player.
2. If the AI is the primary attacker/defender (and can't win on its own) then the AI expects the player to follow it.

1 is correct, the units assigned to the war coordinator on the player's side will follow the player to a self-detrimental level.
2 is by the logic given above not accurate, the AI really does not expect anything from the player allies, for better or worse.

does AI really understand: (at the moment i would answer No to this points)

• where the war-goals are? (siege land of a enemy-ally that give 0 war-score instead of target, even if border war-target)
• what war-goals are? (stand on enemy land therefore gain no supply, but not move army to a land where it can siege)
• what war-goals are? (walk away from siege, even if only a few days left. when player walk to next siege for example)
• distances, and know the situation of an ally? (call to attacking war when ally is at war // or need months to get there)

The AI understands what war goals are and what gives war score. The AI will however make a decision for whether they can feasibly (it will try to avoid attrition) reach the provinces that give war score or not and is supposed to begin prioritizing provinces in the direction of the war goal. There is also some prioritization on what's close to the armies, where it will prioritize closer objectives higher than far away provinces. There might be some numbers to tweak in the war stance priorities that can help here, but anything touching AI needs thorough testing before we commit it for an update. We want to avoid making a change and have the result be worse than current behaviour, and this is something we only can test by having the change live for a long time on our development branches.

 

[helveticadomes]

thx for this insights

(it will try to avoid attrition)

i think AI my wight this too much.
that would explain the unwilling to siege and also the besiege of non war-target if war-goal is not much deployed, but ally is

it will prioritize closer objectives higher than far away provinces

that is ok, as long there is no direct way to war-goal.
and AI should only besiege allies on the way to target if it can not walk around without losing many mans to attrition or it has to embark army.

 

[prismaticmarcus]

Ok, so. This particular problem stems from the fact that the fact that the player is an ally in the war. I would like to emphasize that this is the current implementation.

So, the issue arises from how the War Coordinator decides which War Stance it is in. In most circumstances it does exactly what it needs to do, but the player throws a bit of a complication into this. The War Coordinator chooses the Offensive stance if they outnumber the opposing side, and the Defensive stance if they are being outnumbered. The player(s) are, from the war coordinator's perspective, extremely unreliable allies. The player(s) does what the player(s) wants, and the war coordinator has no way of knowing what your intentions are. Because of this the player's strength is not counted for the army strength on the side they are on, it is extremely pessimistic here. On the other hand, the other side is counting on the player's strength when calculating the strength of the opposing side, because it is pessimistic on its odds and never knows what the other side is up to.

So what we then have is:

	Attacking War Coordinator Sees	Defending War Coordinator Sees
AI Attacker	2610	2610
AI Defender	2941	2941
Player Attacker		1167

This pessimistic look means: the attacker sees 2610 strength on their side and 2941 strength on the defender side; the defender sees 3777 attackers and 2941 defenders.

The attacking war coordinator only see the units it is in control of, so it thinks it is outnumbered and is thus in a defensive stance. The defending war coordinator calculates the full strength of the opposing side, regardless of which opposing war coordinator is in charge of it. So we are now in a situation where both sides are in a defensive war stance, which is why the attacker is holding back in this war.

This behaviour is something we want look into at some point in the future, but I cannot give a timeline to give as to when.

I hope this has clarified it.


Now to address some of the other points in the thread.

1 is correct, the units assigned to the war coordinator on the player's side will follow the player to a self-detrimental level.
2 is by the logic given above not accurate, the AI really does not expect anything from the player allies, for better or worse.

The AI understands what war goals are and what gives war score. The AI will however make a decision for whether they can feasibly (it will try to avoid attrition) reach the provinces that give war score or not and is supposed to begin prioritizing provinces in the direction of the war goal. There is also some prioritization on what's close to the armies, where it will prioritize closer objectives higher than far away provinces. There might be some numbers to tweak in the war stance priorities that can help here, but anything touching AI needs thorough testing before we commit it for an update. We want to avoid making a change and have the result be worse than current behaviour, and this is something we only can test by having the change live for a long time on our development branches.

my biggest complaint about AI behaviour is in the scenario when a number of AI allies and the player are all converging on an AI army from different directions.

say the enemy army is 5000 and the total allied side is 8000. we're all converging. it looks good. but i find if the player army of 500 happens to reach the AI army first and gets stackwiped, then the rest of the allies will often immediately stop advancing and turn and go the other way, when they could still probably win the battle.

are you able to shed any light on why this happens?

 

[PDXOxycoon]

say the enemy army is 5000 and the total allied side is 8000. we're all converging. it looks good. but i find if the player army of 500 happens to reach the AI army first and gets stackwiped, then the rest of the allies will often immediately stop advancing and turn and go the other way, when they could still probably win the battle.

are you able to shed any light on why this happens?

I can only speculate on this one, I'm afraid. Despite having looked at the war coordinator over the last 6 months for personal development time, there are still small things here and there that I haven't looked too closely.

I think the problem is that the unit stack gets wiped out completely, and because of this there are no provinces left for the AI to send its armies to support, so its current support command is cancelled, or it looks for another stack to support if you have multiple armies. Are you talking about a case where the player has just one army of 500, or one where they have several split up army stacks of varying strength?

Best I can give you is to investigate it and have that be part of the dev diary on the war coordinator that'll be coming at some time between now and the heat death of the universe.

 

[prismaticmarcus]

Are you talking about a case where the player has just one army of 500, or one where they have several split up army stacks of varying strength?

the two or three times i've seen it i've been in multiple wars and my army's therefore been reduced to a smaller number.

it really feels to me like the AI armies are set to 'follow the player' rather than 'head for the army the player's heading for'

 

[x4077]

So what we then have is:

	Attacking War Coordinator Sees	Defending War Coordinator Sees
AI Attacker	2610	2610
AI Defender	2941	2941
Player Attacker		1167

Why does the portion of the AI that decides to start an offensive war treat the player ally's force as reliable and think it could win an offensive war when the portion of the AI that actually carries out the war does not?

 

[CollectorOfBolts]

So what we then have is:

Attacking War Coordinator Sees	Defending War Coordinator Sees
AI Attacker	2610	2610
AI Defender	2941	2941
Player Attacker	1167

This pessimistic look means: the attacker sees 2610 strength on their side and 2941 strength on the defender side; the defender sees 3777 attackers and 2941 defenders.

Would the outlook here change at all if the player set their army to follow one of the AI armies?

 

[Coyote_X_Starrk]

I don't need anything to be perfect. The main thing I would just like to see is the AI be more assertive when it has a clear advantage. I wish there was a way to make the AI be able to take the player numbers into account. Too many times I have lost a war where my allies and I vastly outnumber the enemy initially, but end up losing because the AI backs out of a fight at the last second and leaves me to die instead of joining the battle and turning the tide.


Nothing sucks more than when you see the AI clearly marching into battle so you go to join them and then they randomly just stop or turnaround, but you are now too close to the enemy to fall back or change direction. I have actually lost my characters that way more than once. It's very frustrating.

 

[guiskj]

It's funny how this game is all about characters and it is in a time period where wars were a very personal affair for nobles, yet once wars start they become completely mechanical affairs.

Why can't the player coordinate with other nobles. Plan offenses, get betrayed, loose because of craven nobles or foolhardy ones...

Anyway, at the very least the war coordinator SHOULD take player force into account IF said forces are attached to the AI. As that is a clear intent signal from the player.

 

[OnceSetCannotBeChanged]

I feel like this problem has become a lot worse with Fate of Iberia. I've had many wars where an AI ally just sits in a random province and doesn't go anywhere. Incredibly frustrating.

 

[junassa]

@PDXOxycoon Hi. thanks for bringing clarity to this issue.

I think this is very problematic.

1. So far as I know, the AI takes the strength of its human allies into account when deciding to declare war
2. As you've stated, the AI doesn't take the strength of its human allies into account when actually prosecuting said war.

While it's all well and good that PDX is working on a solution to this problem I think you've left the players in an intolerable position. In the interim my recommendation would be to change (1) so that the AI only sees its non human allies.

 

[PDXOxycoon]

It's something I'm looking at, but sadly cannot make any commitments as to when a solution will be made. I'll set you know when I know more.

 

[Dominico]

Couple of things you can do in this situation.

Click on the enemy in their territory to make your army attack. Watch as your army moves one or two counties towards them. The enemy should follow you 1 county behind. This means they are going to help and will follow you into battle.

Second thing you can do is go to the enemy and attach your army to his, this gives him control of your army and he will now have 3k or so troops so may feel more confident about attacking. He will be in full control of both armies.

The other point is they are in the forest so he probably realises he is going to lose that attack and is just hoping he can sit and wait for the enemy to run out of money.

Finally WHAT THE HECK IS A SCOTTISH COUNT DOING IN TRANSILVANIA!!!

 

[Liggi]

Does the AI consider naval paths with a war goal?

In a recent game of mine, I called Wessex into a war in Ireland against a rival Duke. The Duke was outnumbered, so he took his entire army and started seigeing down Wales (which was part of England). It seemed like an odd choice when they could have just attacked Wareham (the capital) on the coast if they wanted to draw the Wessex army away?

I also think the AI is a bit keen to lead their own armies in insiginifcant wars they are called into. The King of Wessex personally sailed all the way to Ireland to fight in a war between two rival tribal Dukes? Seems odd. I've also had the Byzantine Emperor show up in Yorkshire personally in the past.