• 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.

Stellaris Dev Diary #385 - AI Benchmarks

Hi everyone!

The 4.0.13 update released today with the following changes:

Stellaris 4.0.13 Patch​

Improvements​

  • Behemoth Fury is now available to Wilderness Empires.
  • Improved tooltips for the following civics:
    • Functional Architecture/Constructobot
    • Environmentalist
    • Astro-mining Drones
    • Maintenance Protocols
    • Ascensionists
    • Augmentation Bazaars
    • Brand Loyalty
    • Death Cults
    • Dimensional Worship

Balance​

  • Mutagenic Habitability now counts all planet types as ideal for upgrading Gaiaseeders
  • Dramatically increased the draw chance for the Mineral Purification, Global Energy Management and Food Processing technologies
  • Rebalanced the Pleasure Seeker civics to transform Civilians into Hedonists
  • Logistic Drones are now Complex Drones not Menial Drones

Bugfix​

  • Fixed invaded pre-ftls not becoming biotrophies
  • People once more die when they are put in the Lathe
  • Bio-Swarmer missiles can now be used by all biological ships with medium weapon slots (including defensive platforms)
  • Pops that are being pampered will now be forcibly switched to the correct living standard
  • Replacing a district specialisation no longer destroys CyberCreed buildings that should be kept
  • Corrected a tooltip bug where a planet would display itself as a possible migration target.
  • Fixed capitalisation for resources in trade policies
  • Updated assorted modifiers that still referred to Clerks
  • A Trade deficit now causes Job Efficiency and Empire Size issues
  • Fixed the tooltip for the Polymelic trait
  • Armies now protect 200 pops from raiding, not 2
  • Blocked the Federation Code technology for some empires, for example homicidals. To draw the tech, the empire is also required to be in contact with someone they can form a federation with.
  • Blocked the Development focus task Form a Federation for some empires, for example homicidals
  • Added swaps for some empires, for example homicidals, for the Development focus rewards Federation Code, Xeno Diplomacy, and Xeno Relations
  • Updated the Colony view tab mentioned in the hint of the focus task Enact a Planetary Decision to say Management
  • CyberCreed pops with Ritualistic Implants can now colonise planets
  • Fixed Recycled and Luxurious traits not applying to Roboticists
  • Catalytic Processing Civic now lists correct information regarding job swap
  • Cost for repairing orbital rings when you use bioships is now correctly calculated
  • Gale Speed trait gained from Defeat no longer causes errors
  • Fixed scope for LeaderShipSurvivalReason
  • Fixed scope bug for ruler in leader_election_weight
  • Fixed Worker Coop gaining Elite strata jobs in too many places and tidied up the civic tooltip
  • Updated tooltip for Warrior Culture civic
  • Added a pre-list colon to the Feudal Society civic's tooltip
  • The everychanging stone can no longer cause artisans to have negative mineral upkeep
  • Gave the Neural Chorus advanced authority the pop growth speed modifier that had accidentally been assigned to Memory Aggregator
  • The Planetary Supercomputer no longer has an empire cap of 1
  • The Research Institute/Planetary Supercomputer no longer give scientist capacity
  • Added dashes to Traits tooltips and list items
  • Fixed trigger logic for criminal syndicates and federations
  • Fixed Offspring Bioships not being visible in game
  • Fixed Offspring Bioships not being labelled as non-offspring ships in the ship designer
  • The Machine Uprising will no longer spawn 100 machine pops for every 1 missing housing. However the pop-rework seems to have handled 6 million machine pops okay.
  • Stopped removing occupation armies for bombarded and invaded planets on savegame load
  • Repairing ruined buildings in zones is now always possible.

Performance​

  • Flattened pop job modifier node into planet one
  • Made clearing modifiers a fire and forget job

Stability​

  • Fixed a possible OOS when a player leaves the game.
  • Fix CTD when generating a Cosmic Storm mesh.
  • Fixed a random freeze when loading save with stations containing multiple defence platforms.

We expect the 4.0.14 release will be next week (probably on Tuesday), and is expected to include some fixes to a few infinite loops and some select balance changes (like splitting up Enforcers and Telepaths again). It will be a short work week here in Sweden, so it’s likely to be the only update of the week.

As I mentioned last week, with multiplayer stability largely handled, AI is one of our next focuses. Today I want to talk about AI benchmarks, and have a discussion with you about how we should measure “success”.

What Makes a Good AI?​

The AI in Stellaris has always been designed as very reactive, and AI personality has a massive impact on their behavior. Our goal is for our AI empires to feel like actors in the galactic play - acting in a manner consistent with their Origins, Authorities, Civics, and Ethics rather than always picking the “meta” play.

They do still need to put up a bit of a challenge though, especially at higher difficulties.

The first economic goal we make for our AI is “please don’t collapse in an economic death spiral”, and it’s actually far better at that in 4.0 than it was in 3.x. The current AI does NOT meet the second “provide an adequate challenge” goal though.

One of the fundamental tools we have for our AIs are resource targets in their economic plans. They’ll strive to reach those targets, and many of these are set as “scaling” - if they meet the target, they’ll raise the target the next month. This attempts to ensure that they’ll keep thirsting for ever larger research and alloy numbers (or food if they use bioships!) as is appropriate. This is one of the tools we also use to make them exhibit their ethics - Materialists scale their Research targets faster than other empires, so they’ll inherently be more likely to build more Research specializations, while Spiritualists are more likely to have a lot of Unity specializations.

Ironically, improving AI tends to consume any benefits we carved out through performance improvements. The stronger the AI, the more stuff they have - fleets, colonies, and so on.

Benchmarking​

One way to decide whether or not the AI is performing up to expectations is through benchmarking - what kind of fleet power, alloy generation, and research generation should they have by 2230, 2250, 2300, and so on? Around what year should they hit 10k fleet power?

Then there come questions around whether the benchmarks should differ based on personality type. Should it be different if they’re Democratic Crusaders vs. Peaceful Traders? Or does differentiating them there make the friendlier empires too weak?

I’ve got my own set of benchmarks that come from running 3.14 and from the multiplayer community, and in general, I’m okay with Grand Admiral being significantly harder than it was in 3.14. but I’m interested to hear what you all strive for.

How much research and alloy production do you try to have 10 years, 30 years, 100 years, and when the end-game crisis comes calling? (Include your preferred difficulty settings and galaxy sizes as well if you could, as well as if you change any other important settings like tech costs.)

What’s Next?​

We’re going to continue with 4.0 post release support.

Since the next two weeks are both short weeks in Sweden, our next Stellaris Dev Diary will be June 12th. (You’ll be hearing from me in patch notes in the meantime though.)
 
  • 76Like
  • 16
  • 5
Reactions:
The AI's thirst for vassalage is wild.

In my latest game I had a defensive pact with my neighbour and was trying to get into a trade league with them, we fought and handily won a war against their northern neighbour and then as soon as they were at peace they vassalised themselves to another empire to my south. Like dudes you just won a war, you don't need a protector.
It's so bad that I use a mod to turn off all diplomatic vassalization. I also don't like it when they come running to me. And by 2080 or so the galaxy is divided up into huge vassal blocs. If I could get rid of vassalization entirely, I would do so. Vassalization wars are also too rewarding compared to Conquest wars and much of the meta involved abusing vassals.
 
  • 5
Reactions:
The AI really needs another dedicated patch. Not just for general economy either... this game is really lacking any meaningful or realistic interactions between empires. Empires need to be much more reactive to potential threats and snowballing empires. Alliances (and federations) should be far more weighted as a certain guarantee among smaller empires when a massive, sprawling empire is present in the galaxy. Small empires should also have a way to stack some of their research and unity together, to help them progress tech at a similar pace to the bigger blobs. AI is still pretty bad at managing fleets/targets during wars, but the above should be good start.
 
  • 10
Reactions:
The AI's thirst for vassalage is wild.

In my latest game I had a defensive pact with my neighbour and was trying to get into a trade league with them, we fought and handily won a war against their northern neighbour and then as soon as they were at peace they vassalised themselves to another empire to my south. Like dudes you just won a war, you don't need a protector.

100%, it's so frustrating. I generally like having diplo games, even if I'm being more aggressive, and prefer forming a federation over vassals unless I'm playing an empire that is themed around vassals. It's aggravating to have a web of allies that are all in defensive pacts but once trust reaches high enough they start sending emails offering up their sovereignty. Yet at the same time they're super hesitant to join a federation that they could feely leave, and would have a vote in.

In one of my 4.0 games I was playing galactic peacekeeper and got five vassals of a very powerful overlord pledging secret fealty. Which wasn't what I wanted, I was supporting independence hoping they'd rebel, but they didn't. So in the end I went to war and they all swapped to me so I released them. I tried to get them into a federation but in very little time most of them had bent the knee to random empires across the galaxy.
 
  • 8
  • 2Like
Reactions:
And the AI NEED to learn how to specialize worlds, and know that such worlds are a priority to defend during war so players cant just snipe them by scooby-dooing the AI doomstack.
Ten gets you twenty a colony-turtle bot like that can be baited into defending the "wrong" colony while the player commits against another.

(And without very careful tuning, that also produces a situation where bot-on-bot wars degenerate into sitzkrieg, to the detriment of the bots' overall performance.)
 
  • 1
Reactions:
Its being said to death but AIs can keep their personality and still be significantly improved. They make lots of dumb decisions right now that are neither effective nor in character. A spiritualist could focus on unity generation and go to war more readily against a materalist just fine while making its planetary and empire wide economy more effective. This is the area that needs the most improvement. The AI can't use expansion as its only economic tool, it needs to be taught/programmed how to refine its existing economy.

As for difficulty modifiers: A lot of players never consider that most events and story chains are Player Only which means without any modifiers, the player is the one "cheating" a resource advantage. If the AI had access to AI-only events even half as often as the player does, that could help a lot. A dynamic way to do this could be creating some generic modifier-granting events and adding an on-trigger that events can call each time the player gets a permanent buff that runs for all the AI, 40% of the time giving them 1 buff, 20% of the time giving them 2 buffs, and 40% of the time none. This would help the AI scale alongside the player. Naturally, some AI would end off better than others, posing as bigger threats or more competent allies in the process, with the player at the upper end of the curve (since the AI would have 80% as many modifiers on-average).

Another option could be to scale the amount of AI personality's effect on economy and military with difficulty. At Ensign, AI personality is 100% in effect, but in GA, it might be at only 50 or even 20% of economic and military impact. The AI retains full opinion and galcom personality, but has a more humongous economy and military at grand admiral. This could even be a switch like tech costs, for players who want GA modifiers but a more roleplay focused experience, or a slider disjointed from difficulty entirely like AI Aggression. ("Economic Personality: Exagerated, Normal, Limited, Opinion Only)
 
Last edited:
  • 4Like
  • 4
  • 1
Reactions:
I would really appriciate a good AI for stellaris. I stopped playing 2-3 years ago because I always dominated late game and only crisis brought up some challenge. I also find the AI acted pretty boring in general. Why not have an AI empire controlling one third of the galaxy? It never happens, they all stopped expanding at some point. I tried out different AI mods but nothing really helped.

Game would be so good when different AI personalities do cool stuff in a big galaxy!
 
Last edited:
  • 2
Reactions:
I don't aim for specific numbers by a specific year, so I can't give you any. I simply prioritise the things that fits the vibes of whatever empires I'm playing (sometimes to my own detriment).

I have noticed that the AI sucks at prioritizing grabbing systems, or at least it feel that way. I never feel like its actively trying to cut me off or try to grab valuable systems before me.
 
  • 2Like
Reactions:
  • Continually supporting random resolutions like security contractors, even if there’s no benefit
  • Spamming the same federation law requests only to want to revert them a few years later (like migration on or off)
  • Wanting to become a vassal even if they have powerful defensive pacts, guaranteed independence, allies in a federation etc. This one is particularly egregious as it’s poor to have an ally you’ve been friends with for decades randomly offer to give up their sovereignty when they don’t need to. And it interacts very badly with federations, since the AI will often refuse to join a federation but will happily become a vassal even if it pulls them into a fed with no say. IMO the AI should, with some exceptions for personality, consider vassalisation as their last diplomatic option when there are powerful threats.
  1. That irks me, too. So much time in the GC is wasted passing all the low-level resolutions, when literally no one in it has an incentive to actually vote for them.
    My suggestion would be that normal empires won't support a resolution they don't really care about until it comes to the senate floor.

    Their voting on things they don't really care about should be impacted by their diplomatic relations.
    Do you have good diplomatic relations with empire in favor/opposition with it? Then vote accordingly to represent diplomatic efforts - those envoys/delegates are getting paid for something, after all

    Things that benefit themselves(i.e. opposing/supporting a resolution) should massively outweigh factors based on diplomatic relations, though.
    No diplomatic relations should ever get a mercenary empire to repeal any security contractors resolution.

  2. A "What did i vote on this last time?" factor should be included in weighting before proposing and voting. As well as a "Not this again..." factor making them more likely to vote with what was majority last time and makes empires a lot less likely to propose something that has been voted for many times.

  3. The interaction between vassalage and federation is indeed kinda wonkey.
    AI is often very picky about federation relations, but is much less so about vassalization. That has always seemed weird to me.
    Sometimes i wished for ways to influence federation members and potential ones so they'd come together. I had it a couple of times that i wanted to have 2 in my federation, but they just hated each others' guts for no reason i could find despite non-opposing ethics.
    I don't care how much influence i have to invest long term, but the potential to be a mediator would be worth a lot.

    I think an empire in a dire economic situation or stagnation could also ask a neighboring super-power for support.
    But that should be an outlier and definitely less frequent (and seemingly random) than it currently is.
 
  • 6
  • 3Like
Reactions:
As I mentioned last week, with multiplayer stability largely handled,
Multiplayer stability is still atrocious. A group game started Tuesday night on the (at the time) latest patch was throwing pretty much constant desyncs attributed to almost every player and of almost every variety. There are multiple threads in the bug report forum.

We were getting out of sync errors every five or ten minutes once we had gotten a decade into the game.
 
  • 4Like
  • 2
Reactions:
Multiplayer stability is still atrocious. A group game started Tuesday night on the (at the time) latest patch was throwing pretty much constant desyncs attributed to almost every player and of almost every variety. There are multiple threads in the bug report forum.

We were getting out of sync errors every five or ten minutes once we had gotten a decade into the game.
Have you tried starting a new game on the current version 4.13 yet?
 
asking the community how many ressources an empire should have at a certain time on a certain difficulty draws me to the conclusion that your internal QA mostly relies on console commands to trigger certain gameplay situations and not on actual playing a full game. And I am sorry to say it: you asking us how to best fix the AI sounds a little desperate. I am not a game developer…
I don't think you actually understood the point of this. And the dismissive insuation QA doesn't test properly is also uncalled for, especially since you admit you're not a developer.

This is about setting appropriate development goals. They could aim for Grand Admiral to be so OP that only the most minmaxed meta strategy can possibly compete, but that's not what most people want. People still want an appropriate challenge if they pick GA. But what does appropriate mean? Where on the spectrum of the plentiful metrics they have is this vague "appropriate"?
It makes sense to not just arbitrarily define a line as "this is normal", but to ask the people that you're making it for what they want. That's the community communication that is key for Paradox GSGs to thrive for many years.

Additional problems are obviously, too, that this is a complex grand strategy game. Good decision making requires lots of thinking - which takes resource.
It is also something that takes extra effort. No use in an improvement of the AI if it notably increases their cpu resource use. I am sure the devs have had their share of those in experiments, too.
 
  • 7
  • 6
  • 5Like
Reactions:
I don't think you actually understood the point of this. And the dismissive insuation QA doesn't test properly is also uncalled for, especially since you admit you're not a developer.

This is about setting appropriate development goals. They could aim for Grand Admiral to be so OP that only the most minmaxed meta strategy can possibly compete, but that's not what most people want. People still want an appropriate challenge if they pick GA. But what does appropriate mean? Where on the spectrum of the plentiful metrics they have is this vague "appropriate"?
It makes sense to not just arbitrarily define a line as "this is normal", but to ask the people that you're making it for what they want. That's the community communication that is key for Paradox GSGs to thrive for many years.

Additional problems are obviously, too, that this is a complex grand strategy game. Good decision making requires lots of thinking - which takes resource.
It is also something that takes extra effort. No use in an improvement of the AI if it notably increases their cpu resource use. I am sure the devs have had their share of those in experiments, too.
You know, a person doesn't have to be a developer to have every right to disdain both the Paradox testers and their management. Just look at the release of version 4.0 and their lies on the eve of this release, when many people on the forum were screaming about the extremely sad state of the beta.

They deliberately released an unfinished product, which means the consumer has every right to disdain.

Specifically the testers, if they even exist, because it seems to me that testers are only us, maybe incompetent or not having enough people. But the result of their joint work is a shameful 4.0.

That's why your attempt to defend them looks ridiculous.
 
  • 12
  • 6
Reactions:
have every right to disdain both the Paradox testers and their management.
"disdain"? Damn, chill, dude.
If you anyone has such vile hatred for them, they shouldn't be here.
 
  • 15
  • 2
Reactions:
Hi everyone!

The 4.0.13 update released today with the following changes:

Stellaris 4.0.13 Patch​

Improvements​

  • Behemoth Fury is now available to Wilderness Empires.
  • Improved tooltips for the following civics:
    • Functional Architecture/Constructobot
    • Environmentalist
    • Astro-mining Drones
    • Maintenance Protocols
    • Ascensionists
    • Augmentation Bazaars
    • Brand Loyalty
    • Death Cults
    • Dimensional Worship

Balance​

  • Mutagenic Habitability now counts all planet types as ideal for upgrading Gaiaseeders
  • Dramatically increased the draw chance for the Mineral Purification, Global Energy Management and Food Processing technologies
  • Rebalanced the Pleasure Seeker civics to transform Civilians into Hedonists
  • Logistic Drones are now Complex Drones not Menial Drones

Bugfix​

  • Fixed invaded pre-ftls not becoming biotrophies
  • People once more die when they are put in the Lathe
  • Bio-Swarmer missiles can now be used by all biological ships with medium weapon slots (including defensive platforms)
  • Pops that are being pampered will now be forcibly switched to the correct living standard
  • Replacing a district specialisation no longer destroys CyberCreed buildings that should be kept
  • Corrected a tooltip bug where a planet would display itself as a possible migration target.
  • Fixed capitalisation for resources in trade policies
  • Updated assorted modifiers that still referred to Clerks
  • A Trade deficit now causes Job Efficiency and Empire Size issues
  • Fixed the tooltip for the Polymelic trait
  • Armies now protect 200 pops from raiding, not 2
  • Blocked the Federation Code technology for some empires, for example homicidals. To draw the tech, the empire is also required to be in contact with someone they can form a federation with.
  • Blocked the Development focus task Form a Federation for some empires, for example homicidals
  • Added swaps for some empires, for example homicidals, for the Development focus rewards Federation Code, Xeno Diplomacy, and Xeno Relations
  • Updated the Colony view tab mentioned in the hint of the focus task Enact a Planetary Decision to say Management
  • CyberCreed pops with Ritualistic Implants can now colonise planets
  • Fixed Recycled and Luxurious traits not applying to Roboticists
  • Catalytic Processing Civic now lists correct information regarding job swap
  • Cost for repairing orbital rings when you use bioships is now correctly calculated
  • Gale Speed trait gained from Defeat no longer causes errors
  • Fixed scope for LeaderShipSurvivalReason
  • Fixed scope bug for ruler in leader_election_weight
  • Fixed Worker Coop gaining Elite strata jobs in too many places and tidied up the civic tooltip
  • Updated tooltip for Warrior Culture civic
  • Added a pre-list colon to the Feudal Society civic's tooltip
  • The everychanging stone can no longer cause artisans to have negative mineral upkeep
  • Gave the Neural Chorus advanced authority the pop growth speed modifier that had accidentally been assigned to Memory Aggregator
  • The Planetary Supercomputer no longer has an empire cap of 1
  • The Research Institute/Planetary Supercomputer no longer give scientist capacity
  • Added dashes to Traits tooltips and list items
  • Fixed trigger logic for criminal syndicates and federations
  • Fixed Offspring Bioships not being visible in game
  • Fixed Offspring Bioships not being labelled as non-offspring ships in the ship designer
  • The Machine Uprising will no longer spawn 100 machine pops for every 1 missing housing. However the pop-rework seems to have handled 6 million machine pops okay.
  • Stopped removing occupation armies for bombarded and invaded planets on savegame load
  • Repairing ruined buildings in zones is now always possible.

Performance​

  • Flattened pop job modifier node into planet one
  • Made clearing modifiers a fire and forget job

Stability​

  • Fixed a possible OOS when a player leaves the game.
  • Fix CTD when generating a Cosmic Storm mesh.
  • Fixed a random freeze when loading save with stations containing multiple defence platforms.

We expect the 4.0.14 release will be next week (probably on Tuesday), and is expected to include some fixes to a few infinite loops and some select balance changes (like splitting up Enforcers and Telepaths again). It will be a short work week here in Sweden, so it’s likely to be the only update of the week.

As I mentioned last week, with multiplayer stability largely handled, AI is one of our next focuses. Today I want to talk about AI benchmarks, and have a discussion with you about how we should measure “success”.

What Makes a Good AI?​

The AI in Stellaris has always been designed as very reactive, and AI personality has a massive impact on their behavior. Our goal is for our AI empires to feel like actors in the galactic play - acting in a manner consistent with their Origins, Authorities, Civics, and Ethics rather than always picking the “meta” play.

They do still need to put up a bit of a challenge though, especially at higher difficulties.

The first economic goal we make for our AI is “please don’t collapse in an economic death spiral”, and it’s actually far better at that in 4.0 than it was in 3.x. The current AI does NOT meet the second “provide an adequate challenge” goal though.

One of the fundamental tools we have for our AIs are resource targets in their economic plans. They’ll strive to reach those targets, and many of these are set as “scaling” - if they meet the target, they’ll raise the target the next month. This attempts to ensure that they’ll keep thirsting for ever larger research and alloy numbers (or food if they use bioships!) as is appropriate. This is one of the tools we also use to make them exhibit their ethics - Materialists scale their Research targets faster than other empires, so they’ll inherently be more likely to build more Research specializations, while Spiritualists are more likely to have a lot of Unity specializations.

Ironically, improving AI tends to consume any benefits we carved out through performance improvements. The stronger the AI, the more stuff they have - fleets, colonies, and so on.

Benchmarking​

One way to decide whether or not the AI is performing up to expectations is through benchmarking - what kind of fleet power, alloy generation, and research generation should they have by 2230, 2250, 2300, and so on? Around what year should they hit 10k fleet power?

Then there come questions around whether the benchmarks should differ based on personality type. Should it be different if they’re Democratic Crusaders vs. Peaceful Traders? Or does differentiating them there make the friendlier empires too weak?

I’ve got my own set of benchmarks that come from running 3.14 and from the multiplayer community, and in general, I’m okay with Grand Admiral being significantly harder than it was in 3.14. but I’m interested to hear what you all strive for.

How much research and alloy production do you try to have 10 years, 30 years, 100 years, and when the end-game crisis comes calling? (Include your preferred difficulty settings and galaxy sizes as well if you could, as well as if you change any other important settings like tech costs.)

What’s Next?​

We’re going to continue with 4.0 post release support.

Since the next two weeks are both short weeks in Sweden, our next Stellaris Dev Diary will be June 12th. (You’ll be hearing from me in patch notes in the meantime though.)

I'm going to toss something unconventional your way.

Unfortunately I don't really have any hard fleet numbers beyond that for the first century of so my empire builds up towards 10-20k before things snowball ridiculously into having multiple triple-digit fleet power fleets. I suspect that this is helped by either megastructures, ascension, or Cosmogenesis. The last one obviously plays a huge role but even without it things tend to get out of hand and the next moment to realize that the rest of the galaxy (outside of the fallen empires) is rated pathetic.

Ascension is the big one. Currently, the AI is kind of . . . completely trash at handling ascension. They can take ascension paths but outside of Psionics (which is the one path now that does not involve pop modification), they are completely inept at modifying their pops. Currently they will slap random traits onto their species without any regard to how useful those traits are. I've seen AI nerve staple their entire species before. That's how bad it is. One thing I've done in my own game is fiddle around with adding weights to the various traits to make them actually pick the useful stuff and ignore the things they can't handle. This had the effect of AI empires actually being contenders when they ascend, and resulted in the one of the few times an AI impressed my when back in 3.14 a hive mind modded its species with the cybernetic and genetic auto-modding traits and proceeded to hive world terraform a good third of the galaxy. So I think this is something worth pursuing from the dev side of things.

TLDR: modify the AI to use species modification properly. Its possible with the current code and you won't regret it.
 
  • 11
  • 4Like
  • 3
Reactions:
When do we get fixes for gestalt machine intelligences with machine worlds AP / resource consolidation origin? Machine worlds "rural" districts (the bottom 3) still don't change description text to what they are specialized in and don't provide any housing at all. In addition, the "Malfunctioning replicator bay" on the resource consolidation origin homeworld doesn't add any assembly jobs. If this is intended please remove the entire blocker, or rename it, or make it make early game pop numbers go "brrrrrrrrrr" again :3

Plz <3
Oh and I totally forgot to mention - district specialization for trade resource is "Commercial nexus". As a gestalt, I am offended, I don't do "commerce", I do "logistics".


Literally unplayable. :p
 
  • 8Haha
  • 1
Reactions:
I thought the ai was good, too.
There are just too many civilians and building slots that are not used, hopefully this part will be improved.

You can let them develop without creating negative resources, be it trade, alloys, ships, defense platforms.

At least don't let them get too many civilians.

So far, it looks like there's no problem with the basic development of the ai.
Now it's a question of how to utilize so many civilians according to the personality of the computer empire.
 
Last edited:
I use 1000 Stars, mostly full(20+ AI, max FE and Marauders, max Pre-FTL), 5x Habitables(for the AI's benefit, I play Life-seeded to compensate), 3x Tech, 2x Traditions, GA.

The biggest issues I see with AI aside from how to Specialize their Planets (especially in the new juiced up system), are their inability to remove things they already built and getting stuck without a key Tech to progress.

  • If the AI don't get Starhold Tech, they will only have the one Shipyard in their homesystem for a significant amount of time which has a big effect on their Fleet production. All starting Starbases should have both slots as Shipyards; popless Basic Resources are okay, but they can put those on other Starbases than their primary Shipyard station, which is also the Homebase where all their Fleets go to Upgrade.
  • If the AI don't get Cruisers(which means drawing and picking Destroyers as well), they will all just build Corvettes (and Frigates... to fight Corvette Fleets) into the Late Game, which have skewed Fleet Power from their high Evasion and also suffer massive losses they can't afford. This has run on effects on their Diplomatic decision making(big swings in Power rankings and over confidence), and when one AI finally gets Cruisers out they get a massive relative Fleet Power boost and get a bunch of free Diplomacy and Vassals, or go wiping the floor with their new found Power ranking. An easy fix here would be to make Frigates better against Cruisers, as I don't think there is anything you can beat Cruisers with right now unless you have really good Components and a brain to counter them well.
  • Not replacing Buildings as mentioned above, will lead them to making some very patchwork Colonies, and leave them with a bunch of obsolete infrastructure that takes up Pops, Upkeep, and has a big opportunity cost with the level you can now scale a properly set up Specialized world.
I realize the 3x Tech cost is a big factor for me, but 5x Habitables and Primitives should more than compensate for that if they could get even 30% more of the extra Colonies, but they don't seem to, at least not early enough to start compounding their progression. I think a big factor on this is the Technology bloat that comes with DLC ownership, where I will observe an AI's Researched Tech tab and see they have only studied Space Fauna and Storms for the first 50 years or so while their Colonies are full of Storm Buildings. Easy fix here would be to have most of these be entirely ignored by normal AI, and then heavily focused on by only flavour appropriate Empire types. I don't care if one out of thirty of my AI lose because they really like Storms, but to have them all do it is not only bad for gameplay but also immersion. Higher Difficulties should absolutely follow a more meta Planet template and Tech progression in my opinion, outside of a few outlier flavour Empires.

I am very glad to see AIs being adressed(or at least looked at) early on in 4.0, as it took more than 3 years to get around to it after 2.2. I am really loving the new depth you can build into your Planets, for game and roleplaying, so I really hope the AI can join in on the fun so I'm not feeling alone surrounded by training dummy bots and that when I take their Planets I don't feel like they're modern art exihibits instead of functional industry and population centers.
 
  • 2
  • 2
Reactions:
I'm going to toss something unconventional your way.

Unfortunately I don't really have any hard fleet numbers beyond that for the first century of so my empire builds up towards 10-20k before things snowball ridiculously into having multiple triple-digit fleet power fleets. I suspect that this is helped by either megastructures, ascension, or Cosmogenesis. The last one obviously plays a huge role but even without it things tend to get out of hand and the next moment to realize that the rest of the galaxy (outside of the fallen empires) is rated pathetic.

Ascension is the big one. Currently, the AI is kind of . . . completely trash at handling ascension. They can take ascension paths but outside of Psionics (which is the one path now that does not involve pop modification), they are completely inept at modifying their pops. Currently they will slap random traits onto their species without any regard to how useful those traits are. I've seen AI nerve staple their entire species before. That's how bad it is. One thing I've done in my own game is fiddle around with adding weights to the various traits to make them actually pick the useful stuff and ignore the things they can't handle. This had the effect of AI empires actually being contenders when they ascend, and resulted in the one of the few times an AI impressed my when back in 3.14 a hive mind modded its species with the cybernetic and genetic auto-modding traits and proceeded to hive world terraform a good third of the galaxy. So I think this is something worth pursuing from the dev side of things.

TLDR: modify the AI to use species modification properly. Its possible with the current code and you won't regret it.

This sounds like a great thing to combine with personalities too. Materialists more likely to pick the intelligent/natural researcher traits, and also more likely to pick weak/fleeting. Authoritarians more likely to pick conforming and decadent. Militarists strong and unruly etc.
 
  • 4
  • 2Love
  • 1Like
Reactions:
If you guys are looking for AI metrics, best to reference community. Some even have a boost to performance and easy to implement, while others are more extensive and performance hungry. But these mods go into detail of their own respective improvements to the AI. Im not advocating for a pure implementation, but rather you guys take the perceived weakness in AI and implement your own solutions.

Smarter Hyper Relays: Improved AI
This make AI empires build proper networks starting at capitals, branching to sector capitals, then to other colonies.

Enclave Adjustment 3.12
Restrict AI empires from building starbases outside of their territory's adjacent star systems. Pretty much just reduces two variables from 2 to 1, very minor.

StarNet AI
Acomplishes much of what you mentioned preventing death spirals and AI making smarter choices. Key is that its done without modifiers, its all fair. The AI make more deliberate tech choices, prioritize economy based off neighbors, specialize planets, react more to rivals and threats, create long term allies, and more.

New Galactic System
Adds a bunch aside AI, but the takeaway from this is AI weights for Galactic Resolutions that depends more on ethics, civics, origins, etc. or soft cooldowns on sanctions to prevent queue clogging

Aggressive Crisis Engine
Unlike Starnet, this is fully script based and obviously shouldnt be taken on its own but makes for a good reference on where AI is lacking.
  • New custom pathfinding accounting for wormholes and gateways
  • Fleets are relentless and always seek combat.
  • Defensive fleet recall for Contingency worlds (optional).
  • Galactic empires receive initial massive damage penalty against crisis factions so new technologies need to be obtained in order to mitigate this (optional).
  • Crisis colossi (optional).
  • Various fixes related to crises.
AI Game Performance Optimisation 4.0
Take some of these with a grain of salt if implemented in 4.0, and prioritizes performance over balance. However it makes late game much more approachable.
  • Stops the AI from constructing Hyper Relays as they degrade performance especially with their required amount.
  • Stops the AI from constructing Gateways more than twice every 100 years.
  • Decreasing modifier update frequency to 3 times per month.
  • Stops the AI from getting Xeno-Compatibility ascension perk to prevent unnecessary hybrid species.
  • Stops the AI from getting the Omnicodex Relic to prevent unnecessary hybrid species
  • Stops the AI from constructing ringworlds by introducing a limit of 3 ringworlds per AI.
  • Stops the AI from constructing habitats by introducing a limit of 7 habitats per AI
  • It limits the AI from creating large number of sub-species, limiting it to 2.
  • Fills AI job vacancies faster and reduces unemployment by resettling the unemployed pops to planets with open jobs to prevent increased checking.
  • Stops the AI from building corvettes, frigates, destroyers, and cruiser by putting a cap on corvettes and frigates to 60, destroyers to 30, and cruisers to 15 per fleet. Once the AI reaches the cap(s), a corvette is replaced with a frigate, and 2 frigates are replaced with a destroyer, and 2 destroyers are replaced with a cruiser, and 2 cruisers are replaced with a battleship. This should make AI fleets less corvette heavy and stronger. This only occurs when possible e.g., when the AI has destroyer technology and cruiser technology for balance reasons.
  • Prevents the AI from continuously market micro selling and buying small amounts when resource capacity is full.
 
  • 11
  • 1
Reactions: