Stellaris Dev Diary #239 - AI++

[Offe]

Ok, but if I understand this correctly, that's only really an issue once a planet is full, i.e. when I can't (or don't want to) add any more "better" jobs. Then you'd have pops fill the amenity jobs instead of migrating away. And then I can manually limit the available jobs to the amount of amenities I want to have.

If I only have to make sure there's enough "good" jobs available to fill and the game takes care of employing enough amenity producers to stay in the positives, at least this ugly bag of mostly water is quite content.

Yes that is the intended system as for now until we can think of a good solution for how to solve that case when there is no other jobs left to go on

 

[Offe]

Okay, that's good to know. In fact, if I'm reading this correctly, this actually works well with the 3.3 admin sprawl meta.

If the AI considers the monthly trades as their long-term economic plan, then this gives players a good incentive to trade processed goods (CG and alloys) for raw materials, which not only trades at favorable rates, but will also encourage the AI to 'over develop' their raw resource economy to get to the minimum resource margins, when those margins are being consumed by your monthly trades. These will be resource districts and worker pops you don't need to pay the admin sprawl penalty for, even as your surplus of CG lets you employ more scientists.

In time/with diplomatic finangling, you can thus set up your neighbors/allies to be resource colonies, whose economy is fundamentally based around being a resource exporter to you, while your own economy can hyper-specialize to being a CG-based economy that trades. With the AI covering the admin sprawl of your resource colonies, that's fewer sprawl penalties for you, but comes at the cost/risk/necessity of defending your allies to maintain markets.

That's a really good synergy- and absolutely appropriate in macro-economics and sci-fi- and I hope you all can keep that going in the future.

This should work in theory yes

 

[Offe]

I hope the solution is less "make AI disable clerks to min max" and more "make clerks not terrible."

(I still think the game would be better off if clerks were simply removed completely.)

Yes I agree, this is why I chose not to disable them for the AI for this patch

 

[Offe]

IMO the issue is "passive" clerk generation. As long as building housing results in 1 (or 2 with prosperity) mandatory clerks you're going to have situations where you have clerks you don't want and didn't ask for. If you make clerks so good you never don't want them then getting them free with housing is a problem in the other direction.

Take clerks out of city districts and you can go hog wild making them good and rare or bad and numerous or whatever you want.

There are many ways of changing them, but it falls firmly outside of our AI focus so I will leave it up to our game designers

 

[Offe]

I might not have to use those AI improvement mods after this update!

I've seen a lot of people recommending star tech AI mod, I havent tried it myself but they have mandate to change what they want in order to fully turn the AI into a minmax monsters, so while we are working hard on making the AI put up much more of a challenge, mods will always be able to go even further

 

[Offe]

The AI that is in 3.2 is already much better than it was before. Now, it seems the AI is going to be almost as competent as Human players! Maybe they will even always pick psionic ascension since it is the best one!

This post was made by the spiritualist gang

I prefer cloning vats myself

 

[Offe]

Is there any chance that you will also look at the AI's "this is pretty good, but it can be even better" preferences in regards to Ascension perks?

Currently, the AI will not pick any of the Utopian ascension paths (Mind over Matter, Engineered Evolution, The Flesh is Weak) if it does not own 10 or more worlds. This is especially problematic in galaxy settings with few habitable worlds and no guaranteed habitable worlds - there just are not enough worlds per empire. On top of that, the rare AI that does manage to reach 10+ worlds is still very unlikely to randomly pick these Ascension perks due to them having average or low AI weights when compared to other Ascension perks. Since I play with galaxy settings like these for practical reasons (my computer is pretty good, but it could be even better), I have never seen an AI empire go for any of the three Ascension paths.

My suggestions would be to:

• Remove the 10-planet requirement altogether, which other players seem to agree on.
• Remove the ethic-based AI weighting for these three perks; the likelihood of materialists getting Psionic Theory and spiritualists getting Droids is greatly reduced to begin with.
• Give Mind over Matter a greater base AI weight than the other two ascension paths; it would not be unreasonable when considering the rarity of Psionic Theory.

This was brought up earlier as well, I haven't had time to look at any of the tradition tree and AP selection, something I hope to do Soon TM

 

[Offe]

B-Biological ascension over psionic?

H E R E S Y

Cloning is like robot assembly, but we are Humans so I would not recommend machine assembly of course.

 

[Offe]

It HAS to build Soldierjobs - otherwise it won't be able to compete with Human simply based on the differences in Fleet sizes in mid to post late game.
So please consider adding this to the economic plans! (And the code to properly decide when to increase Fleet capacity )

You could be right, I am no expert at this game but I have beaten 25x crisis spawning on year 2350 without any soldier jobs

 

[Offe]

May be a simple shift in Techprogress for hive minds (like a stright x% cut) would balance out the overall baölance at least until the early post late game?
Not as a permanent or final solution, but for now? (And for me at least it would make sense RP wise that hiveminds (organic or synthetic) are less likely to be innovative)

While you may be right, I try not to get too involved into the actual game balance itself. Rather I look at what I can do in order to try and make the AI work better with what it has available right now

 

[Offe]

This needs to be addressed or the overtime trade options with AI be disabled until fixed.
Basically a player can screw any AI he wants by gifting them ressources and then cut them of. Without the AI beeing able to predict this. This is far to exploitable.

If you can make the AI death spiral by doing this, you can upload the save game and I can see if there is a way that they can recover. With all the changes in 3.3 I would guess that they would be able to build themselves out of a "attack" like this, but I can't be certain until someone tests it

 

[Offe]

What is the ominous "GA Trade Buff" everybody is talking about?

Regarding the AI should shift focus away from Tech:
Maybe there should be a differnce between early/mid and late game (defined by the state of the AI empire) scaling economic plans?

So late game (when all reglar techs have been researched, or average level of repetables is like 10 so the limited ones are out of the way) -> stop scaling tech, start scaling energy instead.

Something like this?

In 3.2 Grand Admiral didn't boost trade value for AI empires. Only "resources" from jobs, trade value is not considered a resource.

Definitely possible.

AI scalable economic plans can be scripted with triggers that enable and disable them.

There isn't anything stopping us from doing this except you know that most people don't play 300+ year long games on grand admiral where this would be relevant and there is a lot of things to do

 

[Offe]

Btw.

I totally forgot to thank you (and the team) for the great work you are doing with the AI. I totally love that the time and effort is invested and find the results most promising!
Keep going!

(btw. i am german/franconian, so anything that is not criticisim is to be considered praise, actually giving praise therfore is exceptional! )

The AI could be even better!

 

[Caligula Caesar]

The AI tech picking fixes showed up some funny things, by the way. For instance, the AI was meant to favour researching robots 10x more than the average tech, but in fact, it favoured it 0.1x as much as the average tech. Woopsie!

 

[Offe]

As refreshing as it is to see the AI not spiraling into its own doom, why does it sit on 35K minerals at the cap, with +1K monthly surplus, and not set up automatic trades, or sell them in bulk to stock up in Rare and Strategic resources?

Always room for improvements

 

[Offe]

Praise the totally not robots coding gurus for this! Thematically I love playing Machine Empires but the constant need to shuffle and micro amenities was really grinding my gears. Can't wait to get to try it at some point

One question: How well AI can "downscale" if they for example lose bunch of planets in a war? I assume it should work better as now AI can actually destroy and replace stuff if needed.

Bit of an unexplored topic, so far I've mainly focused on trying to get the AI to build up and expand and conquer others. It is possible that it has become more fragile now since they tend to specialize planets more, but overall they are much more able to recover from death spirals. But yeah this is bit of a secondary priority as once you have beaten the AI once in a war, you will likely be able to do so again, but meanwhile your other neighbours are becoming ever stronger

 

[Offe]

You killed me here. I enjoy your dev diaries a lot with your humor.

A question: will we see those AI improvements used to improve automation features for players (tech, planet building, sectors...)? I mean, it's pretty good, but it can be even better!


Also... Is this using 3.2 admin capacity or is there a new iteration?
View attachment 799088

AI logic and player automation are not the same, I do not know how it works in detail but it will likely not affect it, but I can neither confirm nor deny if it matters.

I think all the screenshots and graphs are from before the admin cap rework

 

[Offe]

To elaborate further, does this mean Grand Admiral AI get the full 100% increase to trade that they did for all other resource jobs before? And in this case, as a sanity check, I presume this is calculated at the point of converting this trade to resources rather than on a per-job / per planet trade basis, to avoid things like inadvertently inflating the value of branch offices on AI worlds, or those worlds chances at hosting the galactic market hub?

Currently it is boosting the planet, so making branch offices there would get a buff yeah, could be something to look into further if it becomes a big problem. It was already something we had on our radar internally

 

[Offe]

Does this, uh, make sense in a context outside of vanilla, though? Let's say a mod adds a new resource, e.g. Gigastructures or similar. If the AI is forbidden to build things outside of their economic plan, will they simply never interface with this resource? If it's up to the modder to add to the economic plan, how can two different mods add new resources without a mod collision?

Additionally, AI can now specialize their planets - but will they take advantage of planet modifiers when doing so, or will an AI specialize a planet in minerals even though it has an energy bonus, for example?

It could never really build things that were not in the economic plan so it would not get any worse.

I think it is possible to mod in new resources and add them to the economic plan but I know very little about the mod capabilities of stellaris so it is just guesswork.

For this version of planet specialization it is just looking at the number of districts that have already been built on a planet. Modifiers that increase the max number of districts of a certain type will be more likely to be specialized into that type of world since they have a higher chance of having a large amount of those districts. But there is definitely room for improvements here, one step at a time

 

[Offe]

Quick question regarding job assignments - currently in 3.2, when you upgrade a capital building, jobs are re-assigned in such a way that you end up with an unemployed specialist and open worker jobs. Any chance the new assignment algorithm fixes this?

If you can replicate the issue on the open beta and upload a save game on the bug forum and ping me there I can look at it