AI should use sortie

[wielkiciensteam]

EU4 (or PDX games in general) has tons of mechanics that are arbitrarily blocked for AI for two purposes: 1. make game easier for player; 2. so that devs won't worry about programming AI to effectively use mechanics. Let's say you are HRE prince with 9k troops. You declare war on OPM who is allied with two other OPM's - each having 4/5k troops. Normal, early game war. The tactic you have to use is rush first OPM, stackwipe them, leave 0.1 men on their level 3 fort -> rush second OPM, stackwipe them, leave 0.1 men on their level 1 fort -> battle last OPM, siege its fort and then siege other two capitals. Classic strategy, if you're not using it you're doing something wrong. But it's sooo cheesy.
If you leave 0.1k men on their fort with 3000 garrison they are unable to recriut mercs or build armies. But they would easily win if they sortie.
The same tactic is used when battling bigger tags -> carpet siege, leave 0.1k men on forts and try to destroy armies so that later on you can siege without any interference.

AI should also use sortie when attacking army that's besiegieng their fort.
I have no hope that devs will implement it, because of one (or both?) reasons I've mentioned above but maybe it's worthwile to start a little discussion about it.
I know that devs need to find the right balance so that game wouldn't be either too hard or too easy but maybe redoing difficulty settings and instead of just bonuses to various modifiers AI would be allowed to use more effectivelly mechanics against players? Of course - improving AI has little no to reward when it comes to game development - bad AI = better games for most players (both casual and more hardcore), who just want to relax and do map painting. I don't want EU4 to have Open AI level of AI but making AI use all possible mechanics in hard and very hard mode seems like a reasonable request.

 

[RichardOlcese]

I agree sometimes if you have a big army that's on step say 9 of besieging a fort, you move most of it to help in a battle, but leave one unit not to lose progress, well, then yes, sorties might be a good option if the rest of the army is compromised somewhere else. But I imagine it's hard for the AI to understand when it's convenient, and it could allow the player to trick the AI into sortie, then moving a big army into the province and boom! you get the fort. So, while sortie is a good decision sometimes (very few times), it's not doable by the AI.

 

[Canute VII]

I agree sometimes if you have a big army that's on step say 9 of besieging a fort, you move most of it to help in a battle, but leave one unit not to lose progress, well, then yes, sorties might be a good option if the rest of the army is compromised somewhere else. But I imagine it's hard for the AI to understand when it's convenient, and it could allow the player to trick the AI into sortie, then moving a big army into the province and boom! you get the fort. So, while sortie is a good decision sometimes (very few times), it's not doable by the AI.

If it's 3000 garrisson vs. 0.1 then it will be a very very quick battle, though. So maybe AI could sortie if the besiegers' size is <15% of garrisson size?

 

[TheCrimsonMajor]

EU4 (or PDX games in general) has tons of mechanics that are arbitrarily blocked for AI for two purposes: 1. make game easier for player; 2. so that devs won't worry about programming AI to effectively use mechanics.

You hit the nail on the head right here. I feel like Paradox has just been adding new mechanics over and over and only teaching the AI to use them half the time. Makes sense, but once you realize this, it does feel a bit disappointing.

 

[Kimbole]

EU4 (or PDX games in general) has tons of mechanics that are arbitrarily blocked for AI for two purposes: 1. make game easier for player; 2. so that devs won't worry about programming AI to effectively use mechanics. Let's say you are HRE prince with 9k troops. You declare war on OPM who is allied with two other OPM's - each having 4/5k troops. Normal, early game war. The tactic you have to use is rush first OPM, stackwipe them, leave 0.1 men on their level 3 fort -> rush second OPM, stackwipe them, leave 0.1 men on their level 1 fort -> battle last OPM, siege its fort and then siege other two capitals. Classic strategy, if you're not using it you're doing something wrong. But it's sooo cheesy.
If you leave 0.1k men on their fort with 3000 garrison they are unable to recriut mercs or build armies. But they would easily win if they sortie.
The same tactic is used when battling bigger tags -> carpet siege, leave 0.1k men on forts and try to destroy armies so that later on you can siege without any interference.

AI should also use sortie when attacking army that's besiegieng their fort.
I have no hope that devs will implement it, because of one (or both?) reasons I've mentioned above but maybe it's worthwile to start a little discussion about it.
I know that devs need to find the right balance so that game wouldn't be either too hard or too easy but maybe redoing difficulty settings and instead of just bonuses to various modifiers AI would be allowed to use more effectivelly mechanics against players? Of course - improving AI has little no to reward when it comes to game development - bad AI = better games for most players (both casual and more hardcore), who just want to relax and do map painting. I don't want EU4 to have Open AI level of AI but making AI use all possible mechanics in hard and very hard mode seems like a reasonable request.

Reason 2 a lot more than reason one I suspect. But anything that gets the AI functioning better gets support from me. Maybe teaching the AI to use the sortie button would be fairly low down my personal list of priority AI ineptitudes that need fixing but I agree with the general thrust.

 

[wielkiciensteam]

Reason 2 a lot more than reason one I suspect. But anything that gets the AI functioning better gets support from me. Maybe teaching the AI to use the sortie button would be fairly low down my personal list of priority AI ineptitudes that need fixing but I agree with the general thrust.

Of course, AI using sortie is very minor thing that has little to no impact in most games. But I believe that allowing AI to use it under some restrictions wouldn't be too hard to implement - AI already knows how much it should risk when engaging battle, AI sees surrounding armies that can reinforce battle so it can be easily, I guess, programmed to do sortie if there are no big enemy armies and besieging army is something like 50% smaller than garrison.
Current AI behaviour regarding sortie is one of many things that just give huge, unrealistic leverage to player and it results, at least for me, in huge immersion loss.

 

[MatthewP]

Of course, AI using sortie is very minor thing that has little to no impact in most games. But I believe that allowing AI to use it under some restrictions wouldn't be too hard to implement - AI already knows how much it should risk when engaging battle, AI sees surrounding armies that can reinforce battle so it can be easily, I guess, programmed to do sortie if there are no big enemy armies and besieging army is something like 50% smaller than garrison.
Current AI behaviour regarding sortie is one of many things that just give huge, unrealistic leverage to player and it results, at least for me, in huge immersion loss.

It makes sense in theory, but I suspect without very careful programming it would be exploited.

For example, player moves their army off the fort, leaves army 40% of the garrison. AI sorties, player comes back before the battle is over and takes the fort. Ok, you say, only sortie when there’s no army in a neighboring province. But then the player has an army in the last day of moving back to the neighboring one. For that matter, in many places you can be two provinces away and still get back, especially with a little army quality. Ok, so only sortie if the army is < 20% of the garrison and no army next door. There are still probably some situations it can fail, but you’re mostly safe. But now all you’ve done for most of the game is forced the player to remember to leave a mostly full regiment instead of a damaged one.

Tl:dr: I think the point about not wanting to code the ai to use these corner case strategies is correct, and the reason is that it’s very hard to code an ai to beat a competent human when the criteria are so finicky. Trying to do it is very likely to replace one minor exploit with a much worse one, and I doubt paradox is willing to invest the effort (and probably iteration over multiple releases to fix exploits discovered by players) it would take to do it right.

 

[wielkiciensteam]

It makes sense in theory, but I suspect without very careful programming it would be exploited.

For example, player moves their army off the fort, leaves army 40% of the garrison. AI sorties, player comes back before the battle is over and takes the fort. Ok, you say, only sortie when there’s no army in a neighboring province. But then the player has an army in the last day of moving back to the neighboring one. For that matter, in many places you can be two provinces away and still get back, especially with a little army quality. Ok, so only sortie if the army is < 20% of the garrison and no army next door. There are still probably some situations it can fail, but you’re mostly safe. But now all you’ve done for most of the game is forced the player to remember to leave a mostly full regiment instead of a damaged one.

Tl:dr: I think the point about not wanting to code the ai to use these corner case strategies is correct, and the reason is that it’s very hard to code an ai to beat a competent human when the criteria are so finicky. Trying to do it is very likely to replace one minor exploit with a much worse one, and I doubt paradox is willing to invest the effort (and probably iteration over multiple releases to fix exploits discovered by players) it would take to do it right.

You are right, this could be exploited in players favor. But if you run some tests then defeating 0.1-1k army vs comparable quality 3k army takes literally days, not weeks or months. The code that allows AI to calculate whether it will win against one army+potential reinforcements from several provinces away is already in the game and can be used to allow AI to sortie without much new programming.
Overall, I think that PDX policy regarding war AI is just - leave it as it is, let 100k stacks sit on level 1 fort, let AI siege Siberia, let AI prioritize abandoning its provinces for the sake of meaningless sieges inside enemy territory. As I've said earlier - making AI better in PDX game almost never has any impact on player experiance, unless it's critical bug that concerns player (hello there bug fixing patch 1.6 for Stellaris that made AI unable to declare any war). Who cares if AI dies of attrition, allows to be sieged down or abandons 49% siege without any reason just to go fight rebels 30 provinces away. In fact most players enjoyment of the game will rise if AI plays bad, it's map painting game after all....

 

[MatthewP]

You are right, this could be exploited in players favor. But if you run some tests then defeating 0.1-1k army vs comparable quality 3k army takes literally days, not weeks or months. The code that allows AI to calculate whether it will win against one army+potential reinforcements from several provinces away is already in the game and can be used to allow AI to sortie without much new programming.
Overall, I think that PDX policy regarding war AI is just - leave it as it is, let 100k stacks sit on level 1 fort, let AI siege Siberia, let AI prioritize abandoning its provinces for the sake of meaningless sieges inside enemy territory. As I've said earlier - making AI better in PDX game almost never has any impact on player experiance, unless it's critical bug that concerns player (hello there bug fixing patch 1.6 for Stellaris that made AI unable to declare any war). Who cares if AI dies of attrition, allows to be sieged down or abandons 49% siege without any reason just to go fight rebels 30 provinces away. In fact most players enjoyment of the game will rise if AI plays bad, it's map painting game after all....

There is certainly some truth to that. But I think also these are harder problems than you're giving them credit for. The logic the AI uses to decide if it will win a battle is hardly bulletproof. Players have certainly found exploits, and even without anything that really counts as an exploit you can trick the AI. Normally If it fails and loses a battle, ok it lost a battle. But if it uses the same logic to sortie from its capitol or a key fort, now it potentially lost the whole war. I'm not convinced this would actually make the game harder if implemented.

I strongly suspect a lot of the other issues you mentioned came out of well-intentioned efforts to fix other problems that seemed obvious. AI sits on forts because before the player would snipe siege stacks. AI lets itself get sieged down because people complained it was stupidly suiciding into stronger armies. And so on. Paradox may not prioritize the AI, but it is not actively trying to make it bad. It's very hard to write a good AI for a game this complex, especially when you constrain yourself to also try to have the AI act "believably," which Paradox does.