If anything, I've found the AI much improved this patch - they are absolutely merciless about carpet sieging, and they're more likely to run away if they see you coming to break a siege with superior numbers.
- 12
Let's talk AI
- Thread starter klingonadmiral
- Start date
-
We will be taking the forums down for scheduled maintenance on Tuesday, May 22nd 2023 at around 8:00 CDT / 13:00 UTC for up to an hour hour.
-
We have updated our Community Code of Conduct. Please read through the new rules for the forum that are an integral part of Paradox Interactive’s User Agreement.
Coallition war against me. BOH and HUN under PU with me and..... They just stay at home.
- 5
Have you tried changing their war stance through subject interaction? With subjects its usually very easy, if all else fails change their focus to attaching your armies and move 2 regiments next to them and allow attaching. Have not seen that fail. But if they are really inactive otherwise then consider uploading the savegame to bug forums. It helps devs to find possible bugs and improve AI behaviour.
- 1
Yes, with changing this, they improve a bit.
If I tell them to be aggressive, they attack everyone, and when I say everyone is everyone, even if they are 50k against 20k.
If I tell them to stay with me, they just stay with me, doing nothing, only be my shadow.
I don't know if is a bug, because I'm not the only one with this issue. Allied AI it's simillar but good, vassals and PU'd are just useless.
If you tell your vassals to be supportive they just follow your armies around join every battle. That is what you want no?
As for allies, if we wanted a good AI that played like a human they should sometimes just refuse to actually fight because they don't want to lose troops or go off and siege their own claims because they want that in the peace deal, that's what a human does. But I doubt people would want that. What people really want is not an AI that plays well, but an AI that plays how the player wants it to; which is an entirely different thing.
As for allies, if we wanted a good AI that played like a human they should sometimes just refuse to actually fight because they don't want to lose troops or go off and siege their own claims because they want that in the peace deal, that's what a human does. But I doubt people would want that. What people really want is not an AI that plays well, but an AI that plays how the player wants it to; which is an entirely different thing.
Be supportive was just a test to see what they do. I just want an AI simillar to the previous patch.
Yesterday I attacked Denmark, at some point when we have inland provinces captured, HUN just went home, then they come back, then home again.....
Or they started to siege a province and run, second after siege again and run and siege and run.....
I found that problems mostly occure with the defnd and siege stance. There they go yolo 5k vs 30k. I'm now using only passive, attach, no focus or if the enemy is broken and it's just carpet siegeing anymore siege. (Disclaimer I have never tried offensive, that sounds like they go kamikaze, but intentionally.)
I would agree that 1.18 would be a massive improvement if every AI worked together as it used to do but they don´t; this means that if you are stronger than your strongest AI opponent sometimes they don´t even try to invade let alone win the war. The AI still does not know what it is doing at a strategic level, it ignores forts (no forts means no territorial gains, etc.), it ignores much weaker armies and its own territory in favor of carpet sieging therefore giving all the time in the world to the player (...) I did a stream yesterday where I fought enemies that heavily outnumbered me (iirc up to 4 times) but at a challenge level it was a very big yawn only comparable to when the enemy armies are asleep.
I am sure PDS will figure it out but right now I would not call 1.18 an improvement, quite the opposite actually. For me the last AI setting that made sense was back in 1.14 or 1.13 where they prioritized defending their territory no matter what (...)
Last edited:
Army AI is solving multi objective optimization problems, meaning that as you improve one aspect of it, you're likely to degrade something else. That's exactly what happened in case of carefulness, and one of the main tasks for Gnivom in 1.19 will be to improve the AI's ability to realize when its alliance is stronger without inducing YOLO behavior.
Not sure what you mean by "ignores forts". Yes, there's a bug where it sieges fortless provinces that get flipped. A lot of the behavior lies in the realm of good sometimes, bad sometimes however. Deciding whether to continue sieging a hostile fort is an extremely tough problem as of 1.18 because you're going to be slapped by terrain penalties. Players get annoyed by AI giving up too easily, yet if it didn't it would be slaughtered far more often.
Just to recap the reason for us messing around with the AI, the 1.17 AI:
- Slept a lot of the time
- Often did not join allies in battles
- Created massive attrition taking stacks (still happens, but much better)
- Did not understand terrain much; this was terrible after the fort change
- Was oblivious to larger enemy stacks
- Couldn't fight multi front wars other than by accident, typically clumped up
A lot was done in short time by necessity, and 1.18 is not the pinnacle of AI, I'm hoping Gnivom can make significant further improvements. Anyway, had we shipped with 1.17 AI, I can assure you there would be no challenge whatsoever after fort change, as you would always be catching the AI sieging stuff oblivious to terrain penalties.
Not sure what you mean by "ignores forts". Yes, there's a bug where it sieges fortless provinces that get flipped. A lot of the behavior lies in the realm of good sometimes, bad sometimes however. Deciding whether to continue sieging a hostile fort is an extremely tough problem as of 1.18 because you're going to be slapped by terrain penalties. Players get annoyed by AI giving up too easily, yet if it didn't it would be slaughtered far more often.
Just to recap the reason for us messing around with the AI, the 1.17 AI:
- Slept a lot of the time
- Often did not join allies in battles
- Created massive attrition taking stacks (still happens, but much better)
- Did not understand terrain much; this was terrible after the fort change
- Was oblivious to larger enemy stacks
- Couldn't fight multi front wars other than by accident, typically clumped up
A lot was done in short time by necessity, and 1.18 is not the pinnacle of AI, I'm hoping Gnivom can make significant further improvements. Anyway, had we shipped with 1.17 AI, I can assure you there would be no challenge whatsoever after fort change, as you would always be catching the AI sieging stuff oblivious to terrain penalties.
Last edited:
- 4
- 1
Well, I think my discovery matches this thread.
Situation is, my ally Mong Yang just stands around in the most far province they can reach, a province of my ally Hsenwi, not involved in war, but I have military access there. Ofc., I cant remove it as long as my allies have troops there. The Mong Yang stack came to the province hunting down an Ava stack. So far, so good, but now they didn't come back for years.
Events leading to situation are just, when the war was declared, we had ~double their troops, but I had 2 uprisings at 90% in my territory. After first battle and siege starting of the capital (Avas forces couldnt be destroyed since their last province couldnt be reached) I led return my troops home to fight uprisings, Ava won vs the sieging Mong Yang army, after several movings forward and back one of my uprisings was killed by me and Mong Yang, but I lost also nearly all my manpower. Now I can't fight the 2. uprising or siege the fort without Mong Yang, if I don't accept to go deep into debt for hiring mercs.
Situation is, my ally Mong Yang just stands around in the most far province they can reach, a province of my ally Hsenwi, not involved in war, but I have military access there. Ofc., I cant remove it as long as my allies have troops there. The Mong Yang stack came to the province hunting down an Ava stack. So far, so good, but now they didn't come back for years.
Events leading to situation are just, when the war was declared, we had ~double their troops, but I had 2 uprisings at 90% in my territory. After first battle and siege starting of the capital (Avas forces couldnt be destroyed since their last province couldnt be reached) I led return my troops home to fight uprisings, Ava won vs the sieging Mong Yang army, after several movings forward and back one of my uprisings was killed by me and Mong Yang, but I lost also nearly all my manpower. Now I can't fight the 2. uprising or siege the fort without Mong Yang, if I don't accept to go deep into debt for hiring mercs.
The AI gives higher priority to carpet sieging unfortified provinces than going after forts, un-siege home territory or even engage "beatable" enemy armies; without a fort it can´t make territorial demands.
Wrt the AI strategic priorities we´ve discussed it months ago when I tried to catch your attention to what I thought was an AI downgrade; I showed a video of the French AI sending its entire army to invade a strategically worthless India (since IIRC my capital was in the new world) by land while I was sieging their country unopposed. I don´t recall exactly in which patch this happened but I think it was in 1.15 or 1.16. In the previous patch (maybe 1.14, I don´t recall) the AI always sent its army back to home territory to try to un-siege its provinces. In 1.18 most of the times the AI ignores its home provinces in favor of, again, carpet sieging an enemy.
Don´t get me wrong I appreciate your work and I even praised it in one of my streams before 1.18 but despite all the improvements the lack of coordination between the AI armies of different countries (amongst other things that I explained in my Friday stream) makes this game much less challenging. Furthermore the AI behaves in a way that is highly exploitable to the point of the player having the ability to literally freeze its armies.
If you can watch this vod, especially when I am fighting a coalition war.
https://www.twitch.tv/marcoan7onio_/v/96313758
Last edited:
AI sending large stacks overseas to conquer strategically unimportant land (for example, trashing my already bankrupt colonial nations) still happens this patch.
Kind of hard to see what you could do to change it - I imagine greatly increasing prioritization of capital-containing continent land, but that is cheesable by moving capital somewhere else.
Kind of hard to see what you could do to change it - I imagine greatly increasing prioritization of capital-containing continent land, but that is cheesable by moving capital somewhere else.
Weirdest AI behavior I noticed with 1.18 in my Ottomans game was when I declared war on Ethiopia (which had conquered a large chunk of Egypt) and I was able to carpet siege half their country (every last non-ZoC province) with only two mercenary units of infantry while the entire rest of my army was away fighting the Timurids and Italian/German minors in a two-front coalition war. Ethiopia didn't put up any resistance at all after my main army captured their only fort in Egypt and then left for another front, even though they could have stack wiped my feeble 2000 infantry easily with sheer numbers, even with their much lower military tech. I think I even saw their main army at one point, and it didn't attack me at all.
I suppose it's actually understandable behavior if fog of war prevented the AI from knowing where my main army was and Ethiopia thought it was a trap, but I've definitely never seen the AI be that cautious before in any prior patch.
I do like, though, that the AI no longer seems to bankrupt itself with mercenary spam like in prior versions.
I suppose it's actually understandable behavior if fog of war prevented the AI from knowing where my main army was and Ethiopia thought it was a trap, but I've definitely never seen the AI be that cautious before in any prior patch.
I do like, though, that the AI no longer seems to bankrupt itself with mercenary spam like in prior versions.
Funny situation.
Castille declares war to France. They called me to help.
Castille + Aragon + Austria(me) + BOH+HUN(PU) against only France.
We were winning the war, with some difficult, and suddenly I see around the map and.... 32k stack of Castille just stayed at Sevilla! Yeah AI, you start a war, but your soldiers stayed at home. Nice move...
Castille declares war to France. They called me to help.
Castille + Aragon + Austria(me) + BOH+HUN(PU) against only France.
We were winning the war, with some difficult, and suddenly I see around the map and.... 32k stack of Castille just stayed at Sevilla! Yeah AI, you start a war, but your soldiers stayed at home. Nice move...
I saw many times my allies with more troops sending them in very small stacks against my enemy to suicide. Lost many wars like that.
This could even be optimal behaviour, if war exhaustion of CN would increase its liberty desire. IMHO no need to tweak the AI here, but tweak game mechanics: let CN/Vassal WE increase its liberty desire.
Thanks for link, I will have a look later this week (even if what I can do nowadays is limited to harassing Gnivom).
Not pissing of the player systematically when allied would certainly be one qualifier of many.