Stellaris Dev Diary #239 - AI++

[Eladrin]

I thought this was supposed to be modding stuff?

Due to an AI rebellion they seized the Dev Diary Production Facilities. Modding will be next week's topic.

 

[Offe]

"In 3.3 we are adding additional AI support for the rogue servitor civic and how they handle their bio trophy pops. The AI should now build an organic sanctuary on each planet that has an upgraded capital structure causing their bio trophies to spread to other planets."

Is this the best idea for the AI? I normally don't spread biotrophies until i need complex drone output on a planet (i.e. if it is a forge, factory or science world), to minimize biotrophy upkeep. I am not sure whether my playstyle is "better" than the default behavior though.

This is more of an RP consideration as opposed to trying to absolute minmax an optimal playstyle for rogue servitors. Once the biotrophies start to reproduce they should then move back to bigger worlds with high numbers of complex drones

 

[Offe]

This is nice. Can you do something like this for Ascension perks aswell? For example make sure Spiritualsit always pick Psionic Ascension, Materialists always go for Synthethic and non-Spiritualist and non-Materialist have some random chance to pick between Biological/Psionic/Synthethic path with higher weight to Biological to get different kinds of empires?

Also it would be nice to see the AI develop Ascension buildorders for Terraforming. For example, Hiveminds will always pick Ascension Perk 1, then Biological AP #1, followed by Biological Ascension #2 and then Hive worlds as 4th Ascension perk. I can not remember a single game where Hiveminds ended up Fertile/Erudite pops and start to Terraform every single planet into a Hive world. Non-Gestalts would do something similar, like picking Ascension perk 1 - 2 -3 and then Ecumenopolis as 4th or 3rd, depending on when they unlock their final Ascension.

It would be really cool if non-Gestalts would end up creating Ecumenopolis regularly while Gestalts will always start to turn all planets into Hive worlds and Machine worlds. The latter probably not for Rogue Servitor, so maybe make them terraform everything to Gaia instead.

I haven't looked at tradition and AP selection yet but yes they are on the TODO list and it is for sure something that could be improved.
The screenshot from above with hive minds was one instance where they were terraforming everything into hive worlds but yes it is not that common to see it

 

[Offe]

What are your thoughts on adding some scripted starts, since the first year or so of the game there aren’t any variables to change things. Stuff like having materialists always build a science lab, or more science ships? Maybe have a couple of scripted starts that are selected based on AI personality? In general, are there any plans for AI personality to effect the AI’s decisions (Beyond just ethics based weighting, which is more generic)?

It is forbidden to build something that doesnt align with your economic plan. You could script it such that materialist needs to fulfill some level of science before other resources are added to their build plan thus forcing them to make atleast X number of science labs as their first buildings.

Personally I am not a fan of making hand crafted special rigid instructions for AI to follow because it makes it very predictable and usually it doesnt take long before these type of solutions becomes obsolete.

Having AI personality and ethics affect more things in terms of how they behave is something we want to do but it would likely be via tweaking weights in order to make them more likely to do certain things rather than making hard rules for them

 

[Offe]

That's awesome! Is this already in open_beta, or still to come?

About half or so of the changes are in the open beta

 

[Offe]

This all looks great One question: are there any plans to improve the AI in charge of planet/sector automation? As I understand it this is a different system to the AI that runs non-player empires.

The planet automation is better than it once was but it still has some rather annoying problems, like how each planet ignores what the other is building and thus a small deficit (say -1 strategic resource) is met with every single world building a refinery.

I haven't looked into player planet/sector automation. While I have heard a lot of people wishing for improvements in this area it has been outside of my primary designation of improving the actual AI you play against unfortunately

 

[Offe]

Far too much good stuff on here for me to comment on all of it, though I especially love the repeating economic plans. So instead I'll comment on your typo:

Shouldn't that "science" be consumer goods?

Right, makes more sense, i was thinking of CG as a intermediate for science

 

[Offe]

Looking forward to a time where I don't have to control the number of available maintenance drone jobs anymore

Unfortunately with how it works right now, once there is nothing left on the planet to work on the pops will go into the maintenance drone job as a last resort.
So it should be better than before but not yet perfect. The AI has added micromanagement where it will go and forbid the jobs if it has too many maintenance drones at the end of the day, but the player has to do this manually for now

 

[Offe]

Are the changes to the jobs assignements the reason robots no longer replace bio pops in earlier stages of the game? This is very convienient in older versions, because the bio pops can migrate to new worlds while the robots fill up the worker jobs over time until we reach droids and later synthetics.

It would be nice if the game would priorize robots and slaves for job assignment to give free pops the choice of being unemployed (with social wellfare and utopian abundance of course) or just migrate away. Or just let everything migrate, no matter of what pop they are. I would assume that this would also help the AI in many cases.

This doesnt sound very familiar to me. We added specific extra functionality in some of the job weight scripting such that robots should be more likely to take base level worked jobs from bio pops in order to free those pops up for specialist jobs.


If you have a good save game you could upload it to the bug forum

 

[Offe]

Do the AI economic plans include Naval Capacity from Jobs?

If they had the same infinite scaling for Naval Capacity then I would have expected the AI to produce more Soldier Jobs in the late game instead of relying on the insufficient Naval Capacity from starbases. (Maybe only start increasing the weight for soldiers when at or over fleet capacity).

Currently it does not but it would be possible to add it in the future. As a first step i would probably make some changes to how AI thinks about starbases in the late game and allow them to go over the cap when they start making huge amounts of energy credits in order for them to continue expanding the fleet.

Not a bad idea though

 

[Offe]

Just to ask, that means that the game will be more "difficult" at the same setting or only that the AI will be reasonably more reactive, but not really changing the user experience when playing at the difficult level we are used to?

It depends on how you look at it. Currently as it is AI tend to be the strongest in the early game. With the changes they should become less weaker in the mid and late game in relation to the player. But it is still likely that the player is able to outscale the AI, but in a lesser extent than before. But comparing absolute numbers of AI fleet and tech between two late game saves between the patches it is very likely that the AI will be substantially stronger now in the late game compared to 3.2.

Overall it will probably give you a more balanced experience where the AI doesn't suddenly become pathetic to you in the mid/late game

 

[Offe]

Regarding the buff from trade value. Does thus mean that hive mind and machine ai empires, which do not get trade value never will be able to go above the 3.3 GA military power curve? While normal empires which will get the trade buff will be more challenging?

Hive minds are the strongest AI empires. With the grand admiral buff to trade value, bio empires will become stronger, but will likely still be eclipsed by hive minds at the end of the day. It helps reduce the difference in their power level but it will probably make the game siginificantly harder by itself

 

[Offe]

With regards to job filling, is there anything to be done to help pops fill the jobs that their traits maximize? I'm thinking in particular of things like having different species who are thrifty, industrious, ingenious, agrarian filling clerk, miner, technician, and farmer jobs respectively. There's currently some slight frustration over the lost output efficiency when species don't select their best output jobs

While I haven't done much of this myself there is a lot of work put into the job weight script to try and make sure each pop works the most optimal job. There are a huge number of jobs and it won't always work perfectly, but the intention is that they should pick according to their traits, and if they dont then upload save games on the bug forum

 

[Offe]

Are there any special conditions to encourage the AI to exceed fleet cap when all starbases built? (E.g. militarists target exceeding fleet cap if they have spare energy credits income and it won't bankrupt them).

This isn't going to just touch AI, and would probably mean retuning ship components, but do you plan to have any plans to 'flavour up' fleet compositions more?

• Like, AI personalities have a dedicated preference for energy/kinetics/missiles (not sure how this really works with Gslots tbh) and strike craft, as well as shields and armor (I don't think there's a preference for evasion/picking small ships over big ones).
• But in practice this is at best nuanced, at worst totally unnoticeable when fighting most AIs.
• It would be really nice if the AIs could lean harder in to this.
  • Like (for example) 75% of their fleets follow their preferences sharply, whilst the remaining 25% will dynamically refit to counter whoever they're fighting.
• So facing a single empire would lead to fighting kinetic and armor-heavy fleets (for example), whilst federal fleets (or vassal swarms) would be a mix of specialised forces.

Lastly with the sprawl changes in 3.3. will AIs make more effort to vassalise other empires? (Or even snap off their own sectors as vassals, if they're excessively over sprawl)

Most of our efforts have been about building up the AI economy, exactly how they will put it to use in the best way is mostly still on the TODO list. I would like to make a better general AI solution for going over starbase and fleet cap.

There has been some work on making different AI empires go for specific ship designs such as missile focus, I don't know how far we got on that topic because I was not directly involved in it, something for a future DD perhaps

 

[Offe]

This immediately caught my attention and the screenshot below just confirms my doubt. In a perfect game your aim for monthly net energy income is +10~+100 throughout the whole game (for less versed player up to 1k near end game is ok). mineral need is more or less specific to your empire size, but usually 30 times your underdeveloped planet count is a good start. no sane person runs a +3k energy surplus if his sole aim is to win the game (i.e. not roleplaying, doesn't have the sheer laziness to micro the economy etc). The pops working technician jobs should probably be reassigned to make more alloy, be soldier to boost naval cap, or get more science somehow.

Yes and no, once you go ultra late game you need insane amounts of energy credits to sustain a fleet way over your fleet cap. Current GA AI can reach about 1.3M fleet power but it starts to struggle in the ultra lategame empire management where production focus needs to start shifting away from science and back into energy credits.

AI also needs to have a solid energy income in the mid game to terraform their planets and do a few other things.

But I agree something like minerals could probably have an upper cap of about 300 per month as the really only huge late game spending on this would be ecunopolis

EDIT: Also i think in the screenshot they had done something like built a dyson sphere because their economic plan wouldnt have requested them to go to +3k energy

 

[Offe]

Question- how do any of these plans impact (or consider) player-initiated resource trading?

Say I'm a Void Dweller, with the typical Void Dweller mineral resource shortage. I make buddy-buddy with a neighborhood empire, who at Cordial or better relations starts being willing to trade their minerals for my CG at rates favorable for me. (Say 2 to 1, so a 6 mineral profit.) Will the new model incorporate my trade deals into their future planning in any way- say not building CG factories since I am 'locked in'- or will they build regardless, and just not be as interested in the deal when it's time to renew?

Good question, this isn't something I have been looking at specifically but if I had to guess I would say that the will account for this in their planning but disregarding any time limit on your trade.

So for example if you are selling them minerals they would likely stop building more mineral districts and instead build other things in the economic plan since it is based on their monthly income

 

[Offe]

Well, a bit of an anecdotal report, as it were... What is present in the 3.3 beta is already pretty good, but it could be even better

I've played a few games with a set of force-spawned custom empires, these always outscale the random AIs. Notably, the fan.mat Machinist Technocracy has managed to almost keep pace with me in terms of science, sitting at only Inferior instead of Pathetic by the 2350s, consistently. Most others perform worse, and I myself wasn't even playing any kind of materialist ... and was doing my usual "pacifist" (but no state pacifism) playthrough where I'd expand to fill in borders, have about 15 planets, then build habitats etc from there.

The main thing I noticed that the AI does not optimize away ... are clerks. Thereby the AI, while sitting at higher pop counts than me (yes this happened), had actually worse science and economy and fleet, because it had pops tied up in terrible jobs. There is almost no world where having clerks is a good idea for most empires - these should be prioritized to be replaced by entertainers or other amenity jobs ASAP, this is something the AI could stand to do.

... That's my main gripe, that the AI doesn't see clerks as a problem, when really they are. They're a fine job as a stopgap temporary measure, but should never be permanent.

Forbidding all clerk jobs is something we have talked about, it would not be very hard to do it, but it doesnt feel right. With the trade buff on GA at least it will be less bad than before but yeah I agree it is not optimal. Here we are approaching the limit of where we can sacrifice immersion for minmaxing.


We are going to look into AI vs AI power levels more in the future now that the baseline AI economy is stronger, I hope there is something else we could do in regards to clerks but we will see what happens in the future

 

[Offe]

Offe, you know my thoughts on how the AI should use the markets more when they want to build stuff but are short on a resource, and have huge excess stockpiles of some other resource.

Question semi-related to that: how effective is the AI at using resources gifted to them by the player? Like, if I'm in a federation (or have a vassal), and I give an ally 10K food and 10K energy and 10K minerals no strings attached, will they efficiently put that to work? Or just add it to their stockpile?

Time is limiting factor here unfortunately. I can only work 24 hours per day

Gifting them food will probably not do anything as the AI always tries to keep a +10-20~ food balance. Energy they could use to do things like terraforming if they have the tech.

If they have unemployment and you are gifting them minerals when they are lacking it they would start constructing more jobs.

Gifting them straight up alloys should make them immediately spent it in accordance to the above posted alloy priorities

 

[Offe]

Also, one more thought since I'm posting:

in my 3.3 beta playtests I noticed that some of the most successful mid/late game AI empires still only had 1 shipyard starbase. Given the increased importance of fleets in 3.3 and generally improved economic AI, it seems like the AI's should prioritize having at least one backup shipyard starbase. If their fleets get destroyed or if they lose their shipyard system to a crisis, they will have a really hard time recovering military strength (can they even convert a non-yard starbase to a shipyard?). Probably something very useful to put on the TODO list.

AI starbase template ratios were not calculated correctly. I have fixed bugs with it and also changed AI priorities regarding how many starbases they should build, now they will focus more on anchorages and shipyards respectively.

I think this was fixed outside of the beta

 

[Offe]

In the beta , i noticed that nowdays ( with me not trying to minmax actively) players start to outrun the AI around battleship tech , so after 2300 midgame tick .


i played only on grand admiral, and most AI normal empires had a big dependency on the market, as they realy liked to have consumer goods planets and a big deficit in energy ... of around 18-20 AI around 3-4 were on equal standing, while 1-2 were actualy superior in tech or fleet power.

i think what keep the AI from snowballing atm is only the fleet management and the autodesign of ships that have some weird weight for lvl 0 fighters \ true dmg weapons ...


well, there is to say that when the AI is your enemy , it seems to be more competent in the use of the fleet ( more competent, not good) than friendly AI that seems to be on the tsundere side . they seems to realy like the take point, but without it will ignore you to death or stand watching when other empires invade ( and eat) evrything you have .

AI Factory/Forge world designation had issues with it in the open beta. I spent more time on this topic after it was reported on the bug forum, hopefully there will be significant improvements on this topic for the 3.3 release, but this system is new and will likely have more edge cases found later on