Is the AI worse now?

[Verx90]

i observed already 4 games to endgame , and played 2 games grand admiral aggressive .

the AI had no problem at economical-fleet power-tech growth . ofc there were empires that did better than others , around 2300 i always kinda outpaced the AI .

ATM the problems are :

1. the AI has a bug with defensive platforms where upgrading them take around 14 years.
2. the AI doesn't upgrade shipyards to cittadels ( negating possibility to build titans\juggernauts\colossus early) and depend on the MEGA shipyards (they build it when they are realy strong) . realy, the AI don't upgrade to cittadels enought bastions-shipyards .
3. AI fleet use , they use fleets realy badly , send suicide fleets against stronger opponents , do not help theyr own fleets when they are attacked , do not run away from fleets (even when they see them in the galatic map) but only if they are in the same system , etc etc.
4. federations use they fleets even worst , 1 AI at using fleets is always better than 2-infinite AI using theyr fleets in war.
5. ascension are picked randomly , traditions are picked randomly.tech are picked randomly. this mean that a player will always outscale the AI eventualy by simply picking the"right" decision for the moment-future plans. even just completing esploration IDEA and picking Technological ascension give a plyer 20% research speed over any AI
6. the AI relation with starbases , they build bastions randomly in theyr system , and do not try to fight with them when possible \ use them to keep a fleet safe\retreat towards it , and for some reason, they ALWAYS BUILD residence on theyr bastion, PLS they need to build shields nerf or weapon range.
7. the AI does not prepare theyr fleets before declaring war . this mean that after a declaration of war , there is always a pause of sometimes YEARS before they even start .
8. AI do not build enought gateways , AI empires with more than a certain N of systems , should realy build gateways

the rest , is an inevitability of the fact that is an AI . players will always outscale them , because the complexity of the game make the work way too hard for the AI, even more because players can ignore the RP of the game, the point 5. is essentialy why an AI will never keep up ( economicaly-tech) with a player . because we have plans , the AI simply pick something randomly .


a fix for point 5. would be to give determined ethics-civic combinations a "build" to follow of traditions and ascensions . but this would make the game much more repetible.

a way to amortize this , would be to give us the possibility to "pre-build" the AI of a faction that we create . a sort of "build my nemesis - ally - neutral that will do something.

atm , i've an empire HIVE MIND DEVOURING SWARM that always manage to become the strongest (fleet power- economy - ) of the game and almost keep up with tech .
if only i could give it what to pick , i could create a perfect nemesi for evry game , even with the HORROR movie that is the fleet management of the AI .


edit: if you have problems with lower difficulty, i suggest you to go scaling difficulty with grand admiral . ALWAYS PICK AGGRESSIVE AI, or the AI will not conquer when she can and you will always end up the only blob of the galaxy.

 

[pmchem]

I suspect PDX playtest by hosting huge MP matches at the office and hence they never play vs AI

if so, that is just the dumbest way to test since it's a small minority of actual playerbase experience

AI has been terrible for who knows how long. It's the reason I stopped buying Stellaris DLCs and encourage you guys to do the same.

Good AI programming might not be very flashy or sell products, but it's still the engine that drives the whole thing.

yup

 

[pmchem]

ATM the problems are :

1. the AI has a bug with defensive platforms where upgrading them take around 14 years.
2. the AI doesn't upgrade shipyards to cittadels ( negating possibility to build titans\juggernauts\colossus early) and depend on the MEGA shipyards (they build it when they are realy strong) . realy, the AI don't upgrade to cittadels enought bastions-shipyards .
3. AI fleet use , they use fleets realy badly , send suicide fleets against stronger opponents , do not help theyr own fleets when they are attacked , do not run away from fleets (even when they see them in the galatic map) but only if they are in the same system , etc etc.
4. federations use they fleets even worst , 1 AI at using fleets is always better than 2-infinite AI using theyr fleets in war.
5. ascension are picked randomly , traditions are picked randomly.tech are picked randomly. this mean that a player will always outscale the AI eventualy by simply picking the"right" decision for the moment-future plans. even just completing esploration IDEA and picking Technological ascension give a plyer 20% research speed over any AI
6. the AI relation with starbases , they build bastions randomly in theyr system , and do not try to fight with them when possible \ use them to keep a fleet safe\retreat towards it , and for some reason, they ALWAYS BUILD residence on theyr bastion, PLS they need to build shields nerf or weapon range.
7. the AI does not prepare theyr fleets before declaring war . this mean that after a declaration of war , there is always a pause of sometimes YEARS before they even start .
8. AI do not build enought gateways , AI empires with more than a certain N of systems , should realy build gateways

stuff like those would be good to report as individual official bug reports just in case something actually gets fixed via 1% random chance of a QA guy looking at it

 

[HugsAndSnuggles]

ascension are picked randomly , traditions are picked randomly.tech are picked randomly. this mean that a player will always outscale the AI eventualy by simply picking the"right" decision for the moment-future plans. even just completing esploration IDEA and picking Technological ascension give a plyer 20% research speed over any AI

Not entirely random, especially traditions. Still, the main problem here is balancing: if weights are supposed to be balanced around RP, choices should provide comparable benefits.

 

[Verx90]

stuff like those would be good to report as individual official bug reports just in case something actually gets fixed via 1% random chance of a QA guy looking at it

i already reported the BUG about the defensive platforms ( i was dead when i saw all AI starbases stuck at upgrading defensive platforms for YEARS, just to start again because the AI found a new tech or the auto-generator decided to change something )

all the other , i don't think are considerable as bugs . they work as ""intended"" by the programming behind them , it is simply still far from perfect.

And considering how it works , they are working on it by years at this point , but are slowed down by constant rework on economy\military\FTL system .


even just the change to the "DOOMSTACK" fleets moved the AI fleets management\use back by 2 years of developement.

 

[MrGuyPerson]

AI fleet use , they use fleets realy badly , send suicide fleets against stronger opponents , do not help theyr own fleets when they are attacked , do not run away from fleets (even when they see them in the galatic map) but only if they are in the same system , etc etc.

I've noticed issues with fleets as well. I was playing a Grand Admiral game and was trying to finish up a war. While my fleets were on one side of my empire, my neighbor on the other side declared war. The only entrance to my territory was guarded by a bastion and Bubbles. The AI sent in his fleets and while they were in the system, I sent Bubbles away so it wouldn't get killed. The AI gave chase until Bubbles escaped. I figured it would then go after my star base, since it had enough power between its three fleets. Instead, it kept two fleets one jump away and then flew its other fleet back and forth just out of range of my star base. I'm really not sure how the decision tree worked there but it was very strange. It gave me plenty of time to regroup and prepare in a situation where I should have been in trouble.

 

[Verx90]

I've noticed issues with fleets as well. I was playing a Grand Admiral game and was trying to finish up a war. While my fleets were on one side of my empire, my neighbor on the other side declared war. The only entrance to my territory was guarded by a bastion and Bubbles. The AI sent in his fleets and while they were in the system, I sent Bubbles away so it wouldn't get killed. The AI gave chase until Bubbles escaped. I figured it would then go after my star base, since it had enough power between its three fleets. Instead, it kept two fleets one jump away and then flew its other fleet back and forth just out of range of my star base. I'm really not sure how the decision tree worked there but it was very strange. It gave me plenty of time to regroup and prepare in a situation where I should have been in trouble.

I THINK if they can dodge your BASTION they are programmed to do so , if you do not research the bastion FTL block , the AI will never engage your bastions if they can't win with 1 fleet easy .

i had wars where the AI proposed peace ASAP as the tech was researched , because it didn't have the fleet power to defeat my Bastion on theyr border .

 

[MrGuyPerson]

I THINK if they can dodge your BASTION they are programmed to do so , if you do not research the bastion FTL block , the AI will never engage your bastions if they can't win with 1 fleet easy .

i had wars where the AI proposed peace ASAP as the tech was researched , because it didn't have the fleet power to defeat my Bastion on theyr border .

I'm pretty sure I had that researched, though I could be wrong. Even if I hadn't researched it yet, the AI's fleet never tried to leave my system; it just flew back and forth within the system and never sent in backup that was one jump away. Then when I finally sent over my fleets, he just sent his other fleets away and never forced a large engagement.

 

[blind-oln]

I think the AI is worst now. I least I'm pretty sure I didn't become as good in a span of a day. Before 3.01 AI was easy to defeat but at least it had some fleet and keep up was trying to keep up with my economy. Now, with 3.01, all AI players in my game are just pathetic and keep losing to rebelions every 20 years. Now there is a spamfest of 1 planet size empires that want me as a protector.

 

[Verx90]

I think the AI is worst now. I least I'm pretty sure to not become as good in a span of a day. Before 3.01 AI was easy to defeat but at least I had some fleet and keep up with my economy. Now ALL AI players in my game are just pethetic and lose to rebelion every 20 years. Now there is a spamfest of 1 planet size empires that want me as a protector.

i suggest you to increase difficulty . i've never saw something like this in my games.

 

[blind-oln]

i suggest you to increase difficulty . i've never saw something like this in my games.

I will in my next game, but I'm playing on the same difficulty as before.

 

[Verx90]

I will in my next game, but I'm playing on the same difficulty as before.

unluckly , i started to play Grand admiral scaling on\off max aggressiveness from 2.6 , so i've no idea of how the AI performed before at different difficulty.

even more important , the trick is the number of AI in the game , you need to have a decent number of AI , to stop you from expanding for free, and to give war target to other empires \allies .

on a 800 galaxy i suggest around 15-18 empires , on a 1000 galaxy you need around 27-30 AI empires .



in my observation in 2.8 , around 2-4 AI over 30 ( starting AI , usualy end up being 37 ingame) end up becoming quite strong , still would get outscaled by a player \ killed before getting there , with around 5-6 AI having a significant fleet power .
in 3.01 nothing realy changed , there are still around 2-3 AI that are on par with a player till around 2300 ( no change in tech cost etc. only dates of midgame-lategame by 25-50 years ) with a 4-6 AI empires with a significant fleet power .

those are over 30 + AI empires . it was "bad" before, it is as bad in this patch with a plus of the economy change that may offset a player .

 

[pmchem]

i suggest you to increase difficulty . i've never saw something like this in my games.

the problem with just "increasing the difficulty" is that it masks real, fundamental problems by just letting the AI cheat. a lot of issues will never be fixed by devs if people just universally change that slider. then it just adds technical debt so that when the slider no longer helps, the game is revealed to be broken in all sorts of troublesome interconnected ways.

 

[sotmh]

As a note to Paradox marketing dept. If your prerelease marketing centrally features a huge multiplayer streamer game I'm just going to assume the AI isn't good enough to showcase the game in single-player streams.

 

[mial42]

I think the AI is worst now. I least I'm pretty sure I didn't become as good in a span of a day. Before 3.01 AI was easy to defeat but at least it had some fleet and keep up was trying to keep up with my economy. Now, with 3.01, all AI players in my game are just pathetic and keep losing to rebelions every 20 years. Now there is a spamfest of 1 planet size empires that want me as a protector.

Check for food and CG deficits. Oh, and way too many freaking defense platforms.

 

[Verx90]

the problem with just "increasing the difficulty" is that it masks real, fundamental problems by just letting the AI cheat. a lot of issues will never be fixed by devs if people just universally change that slider. then it just adds technical debt so that when the slider no longer helps, the game is revealed to be broken in all sorts of troublesome interconnected ways.

mhh.. i think we have to understand that the AI will never be on par with experienced player.

the difficulty is there to make it more changeling and competitive, it is never going to be on par . this is true for almost ANY videogame that is centered about strategy against AI , i think a game need a level of semplicity on par with chess to have competitive AI , and even there it usualy have no " grand plans" and decision making sufficent to adapt in meaninfull ways .

Programming AI is not simple, is not free , and take times.

as i stated before, they worked on it for years, and are still working non-stop on AI ( on evry single patchnotes you can see the sectiction about AI ) , the whole system was reworked too many times to have an AI programming keeping up .

even then , the AI work , it has no adaptibility , or diplomatic decision making apart from declaring wars that it think to be able to win by fleetpower comparison and the inability to deliberately decide to ignore a deficit in order to obtain a more important resource\results is nowhere inside the AI tendency . This mean that evrytime you decide to ignore a food deficit and just buy it , you are taking an advantage over the AI , that try to fix that deficit ASAP .

you can see the AI as reactioning , they are trying to give it a sort of " plan " at least economicaly , but the AI still try realy hard to have a stable economy with no deficit ( i'm not only talking about resources, but tech-administration too ) .

the increased difficulty is an abused meccanism that any strategy game actualy use and used . and lets not talk about AI with simply preset actions .


it would be a dream to have a difficulty being based on AI capacity and just limiting the AI APM on lower difficulty . but atm the best you can have is giving AI advantages. because evrytime you make an action that result in an investiment for future gains , you are outscaling the AI . be it rushing for a system choke point, or investing in early game unity \ tech + ascensions + tech speed .


The AI could be better , but it is already more good than the last time the economy was reworked.

and to go back to the discussion origin , it didn't particolary change from 2.8 . as long as i remember , AI with no bonuses get snowballed by anyone that played the game more than twice . because we humans have things called abstract thinking , and the AI doesn't learn from evry game.



but if you want to add a learning AI to stellaris , i'm totaly ok with it.

 

[Olterin]

It seems that the baseline AI needs some help with how to manage their population. Things of note, in no particular order:
-Amenities, they're important to keep out of the red. AI pls, stahp having rebellions all over
-Clerks. We know it's a bad job, the AI does not. Just ... ugh. When I integrate vassals the first thing I do is shut down their clerks and make sure those pops move to productive jobs ASAP.
-Resource shortages. I've had vassals sit on a food shortage for years upon years upon years, when I had left them (a released sector) with enough of a functional economy to not be unable to recover from a food shortage
-Building upgrades. They're just not worth it, like, ever. Well, almost. They're slightly worth it if they allow you to concentrate pops on a more productive world, but this only kicks in when you hit the growth wall and your main way to improve the economy is to improve the throughput efficiency. Yet AI upgrades all the things even when it really really shouldn't

... Seriously, stability and amenities. Pls fix. It's gotten beyond comically bad when I start a game with 9 empires and by 2300 there are 20+ (and no, no primitives, those get cleaned up, it's really almost all rebels)

With all that having been said ... With higher difficulty, i.e. the AI having baseline more productive pops, most of these issues cease to be noticeable in that the AI remains vaguely competitive for a good while. But yeah, these are issues that arise on Ensign, when the AI gets to deal with the same tools as the player.

 

[pmchem]

It seems that the baseline AI needs some help with how to manage their population. Things of note, in no particular order:
-Amenities, they're important to keep out of the red. AI pls, stahp having rebellions all over
-Clerks. We know it's a bad job, the AI does not. Just ... ugh. When I integrate vassals the first thing I do is shut down their clerks and make sure those pops move to productive jobs ASAP.
-Resource shortages. I've had vassals sit on a food shortage for years upon years upon years, when I had left them (a released sector) with enough of a functional economy to not be unable to recover from a food shortage
-Building upgrades. They're just not worth it, like, ever. Well, almost. They're slightly worth it if they allow you to concentrate pops on a more productive world, but this only kicks in when you hit the growth wall and your main way to improve the economy is to improve the throughput efficiency. Yet AI upgrades all the things even when it really really shouldn't

... Seriously, stability and amenities. Pls fix. It's gotten beyond comically bad when I start a game with 9 empires and by 2300 there are 20+ (and no, no primitives, those get cleaned up, it's really almost all rebels)

With all that having been said ... With higher difficulty, i.e. the AI having baseline more productive pops, most of these issues cease to be noticeable in that the AI remains vaguely competitive for a good while. But yeah, these are issues that arise on Ensign, when the AI gets to deal with the same tools as the player.

yup, the AI needs to be able to handle the basic economics of planets in order for stellaris to actually be fun

 

[fourteenfour]

It certainly builds more defensive platforms than before but in the end the AI is starved for production for the same reason the player is, the new population hack. Think of it this way, every handicap put into the game hits the AI harder because unless specifically coded to handle the situation it just goes along as if everything is normal. The AI still wants to grow pops that won't because the growth tax exceeds what you can get from buildings and such on day 2 and just gets worse. It doesn't know it gets worse so it tries anyway. Assembly buildings are the worst offender as their rates are fixed but their costs are not, they are a trap now

 

[xeleth12]

At this point I don't think they even test their game anymore.
The AI obviously can't handle the changes to pop growth, its reductions/nerfs, and it was already bad at it before. That it's so bad at it now, it shows that it didn't get much attention/testing. That such a change messed things up, again? Absolutely no surprises there.
I know scripting/coding an AI, in such a game especially, is a tough task. But PDX have been having troubles with the AI for quite a while now, and some modders somewhat fixed it / improved it (maybe could've taken inspiration from them?) but with this update it just looks like they ignored the AI almost completely or at least didn't give it much love. Without surprises, the changes to pops really did a number on the AI. Of course there'll be problems, like John Hammond said right? But to this extent? No good.
It's the third (I think?) rework/overhaul of how the pops work. That can't be good either, as everytime huge, massive even, chunks of the game have to been reworked to accomodate the changes, repercussions through the whole game and its various systems. Of course it will create issues, but the more you rework that system the more the little problems left behind here and there through the various systems from the last changes will start to accumulate, and accumulate.
I mean we still have 3+ years old bugs in the game, some close to being gamebreaking (fixed by modded but not PDX btw), so yeah, this is a bad sign in my opinion and will take months to be fixed/adjusted.