Q&A With Merni - AI

[merni]

Do AI civilizations build up "threat" opinion modifiers against end game crises? Say, the AI revolt is murdering everyone, will the civilizations in its path band together in a coalition to defeat them?

Yes!

 

[merni]

Hi Merni

Wiz said in one of the early Blorg streams that species from the same group have a greater affinity when first contacted with species from their own group than if the the contact was from another species group. ie: if an Avian race meets another Avian race for the first time, they are more "friendly" than if they had met say a Reptilian race.

Is the reverse true also? Is there any in-built "rivalry" between species groups? For example do AI Arthropod races have more first contact disdain for Avian races than they would say for Mammalian races? If not, is this something that could be easily modded in?

Depends on their ethos as well but generally yes.

 

[merni]

In BaneWilliams stream, a single robot pop on a sectored planet decided it would be a good idea to work on a tile producing only food, which is of no use to a synthetic pop.
Clearly there need to be some specific AI script that govern priorities when a planet's pops solely are artificial.

This is a known issue. We are on it!

 

[merni]

Do AI empires also have their own situation log and will they research anomalies?

Yes they will!

 

[Cossack8559]

Hi Merni

Are the a.i alliances dynamic like in EUIV where they can be friends and enemies in the course of the same game or is it more static where they will remain allied until the end?


Also in one of the twitch streams, Shen declared war on a empire of equal strength... As he invaded the enemies outer systems there fleet was no where to be seen. After occupying a couple planets he moved his 4k fleet to the enemy home world where they had 6k sitting. The a.i fleet won the battle forcing shen to retreat, but despite the a.i having a larger fleet and having just won a major fight... they agreed to be vaslised as the war score had ticked up so much through occupations while the fleet sat at the capital planet. Is this working as intended or was this a bug from the pre-release version? The vassal who ended up with 13k fleet remained loyal to shen even when his fleet was only half the strength, i was waiting for them to declare an Independence war but they never did.

 

[StealthyBandit]

Does the AI Empires interact in the same ways as the Player Empires when they deal with their own sectors "mini AIs"?

Or do they manage all themselves directly?

 

[Safehold]

My question has to do with how the battle system determines which target to focus fire on.

If a battleship decides to shoot at a corvette, does that take into account weapon distance, damage of the enemy ship, evasion chance/to hit chance, and/or actual dps calculations of offense vs defense? This is without the addition of combat computers.

The reason I ask is because if the target priority is based only or more on range, there are ways to manipulate that in a fight. If it was based only on to hit chances and actual dps applied, it would be hard to impossible to manipulate to the player's advantage.

 

[neusaap]

Is the Stellaris ai allowed and/or will they often pick 'dangerous' tech such as sentient robotics and warp drive, or will they take the cautious route, leaving it up to the player whether organic life gets destroyed?

EDIT: I also wonder: Are there potential combat computers that use special tactics such as focus fire or only ones with 'basic' tactics, such as playing defensively or offensively?

 

[Mackus]

You can spawn predefined empires with the same species type, yes!

That's great news!

 

[Pigeoncount]

In what ways do the AI "cheat", if any. Like with the documented list of AI advantages in EU4.

I'm not opposed to the AI being given some advantages to compensate for their limitations (although, ideally, it would not be necessarily), I am just genuinely curious.

 

[Tim_Ward]

Can the AI be modified to work better with any changes made by mods?

 

[TheBeautifulVoid2]

Hi merni,

In the mid/late game, the human player will know which Fallen Empires are in the game and which are not. This will, I expect, affect their behaviour: for example, if there are no Keepers of Knowledge then the human player may decide that they can get away with researching forbidden techs that they would otherwise not research for fear of being roflstomped by Fallen Empire fleets.

Will AI empires have the same "yay, the teacher isn't watching, let's break all the rules" attitude?

 

[Tiaexz]

Need to stick the Merni AI war picture in the OP, seems fitting.

 

[Daddl]

Do you use some standard mathematical model to get all these different modifiers and values together to meaningful numbers? Or do you mess around with numbers until the result is satisfying?

 

[IntoTheStars]

Are you able to discover an anomaly on a location that someone else already surveyed? Or are they specific to the one location and always discovered when surveyed?

 

[corny.1234]

Hi merni,

Do the anomalies are drawn from a common pool from which every empire can get anomalies or does every empire has his own.

Practical, can you research a system and find anomalies in it even if it has been surveyed by an ai empire?

 

[Silas1775]

Anything you can share about the slaves never rebel issue? (the problem, difficulty to fix, eta, etc)

Thanks for answering questions!

 

[Jorlem]

Our AI is not comparable to a NPC in a general game since it has to play the game. These categories that you mention are not really applicable to our games. We need to create a variation of non deterministic AI:s so that they are not predictable. This can't be solved with a simple algorithm or a state machine. But I guess logical AI is the closest to what we do. We use the different ethos, traits, policies to create the variation in the decisions.

For some reason, the alert system is saying you quoted my earlier post. Is this a glitch I should report to the forum tech people, or did you just edit the post?

 

[Magdaki]

Our AI is not comparable to a NPC in a general game since it has to play the game. These categories that you mention are not really applicable to our games. We need to create a variation of non deterministic AI:s so that they are not predictable. This can't be solved with a simple algorithm or a state machine. But I guess logical AI is the closest to what we do. We use the different ethos, traits, policies to create the variation in the decisions.

Thanks for the reply. That is to some degree what I expected based on the observed AI behaviors in most strategy games. So would you say then that the objectives of an AI in a GSG (or a strategy game in general) is make rational, but possibly non-optimal, decisions while providing a personality to the game? I.e. so that a player will feel like you are playing against another civilization. That optimization is not the most important factor since such a system will feel robotic if it is to deterministic and ruin the immersion?

 

[LordPavel]

Semi-unrelated to Stellaris, but do you foresee using cloud-based computing to boost game AI and allow for machine-learning to be part of your toolset as an AI programmer? I'm thinking of Microsoft's Cortana and Azure based APIs...
Could allow for some awesomely good future game AIs.

Not sure that will be great. If a IA can learn to play against a human, this IA will not try to play fun, this IA will try to play with the objectif we give and dont care if an Human think this is a good or a bad game style.