Q&A With Merni - AI

[LastLeviathan]

Couldn't find another thread, so conglomerate here for all your AI question needs. And we appreciate it, @merni !

I totally love talking to you guys! Please start a Q&A for the AI in the forum if you want to know more about that.

https://twitter.com/merni_/status/728266053835460608


Not sure if this has been asked before but how do AI deal with Fallen Empires? Will they try and trade with them? Plot against them? Will they accidentally conquer a Holy World for example or actively stay away from annoying the Fallen?

 

[KillSteenKill]

How do endgame crisis AI interact with each other? Will they be always hostile to each other?

 

[Cynical Dreamer]

Hi Merni,

Do you honestly feel that the AI is working perfectly fine right now, or do you think it still needs a few days (and patches) before working as intended ?

 

[bcsmall]

Two questions:

First, mostly just because I'm curious, does the AI know that my empire is "the player?" By that I mean, does the AI interact with the human player differently than it does with other AI's?

Second, the AI in CK2 has very clear modifiers depending on your difficulty level. The "accept matrilinear marriage" proposal for example, clearly showed that the chance of acceptance would be lower if you played a higher level. Is the Stellaris AI similar in that regard? Are modifiers in place at higher difficulty levels to make it more difficult for the AI to accept propositions? I guess, more generally, what makes the AI "harder" at a higher difficulty level?

 

[ZenthWolf]

I have a technical question about AI concerning a stupid mod idea I have. When I read this post by Wiz, I began to suspect I might end up feeling "bored" by the idea of AI personality defined very statically this way. I understand there is more than this file that goes into the calculations, but for now, this is what I have to work with. When I saw this I thought, "Why not introduce some variance into this file?" In my mind, if I create a bunch of copies of each personality, but with a small amount of randomization, this could be potentially very interesting.

My question then is this: if I create a supremely large number of possible personalities, will this effect performance? Will the game keep all personalities in memory, and think about one empire as "being a Hedgemonic Imperialist" or does it keep a copy for each empire separate (i.e. here are the traits for the Federated Zracon Nations). The former, I suspect would cause issues with scaling, but the later, in principal, wouldn't feel the impact of having millions of personalities as it keeps one per empire.

 

[merni]

Couldn't find another thread, so conglomerate here for all your AI question needs. And we appreciate it, @merni !



https://twitter.com/merni_/status/728266053835460608


Not sure if this has been asked before but how do AI deal with Fallen Empires? Will they try and trade with them? Plot against them? Will they accidentally conquer a Holy World for example or actively stay away from annoying the Fallen?

They will not annoy them if they don't have the capacity to actually deal with the consequences. So they will act like a human but possibly more cautious.

 

[TheBeautifulVoid2]

Hi merni. What is the heuristic for AIs deciding to enslave or free pops?

 

[merni]

How do endgame crisis AI interact with each other? Will they be always hostile to each other?

Yes they will. And if something is interrupting their expansion they will try to deal with it

 

[Wuz]

@merni
I noticed in Arumba's videos that there is a lack of focus fire in the combat AI.

Is this something that the dev team has considered adding as an "approach option?" He raised a few good points about it.

 

[merni]

Hi Merni,

Do you honestly feel that the AI is working perfectly fine right now, or do you think it still needs a few days (and patches) before working as intended ?

I would be a lousy programmers and game developer if I was ever content with the current state of the AI or any other part of the game for that matter. I think you are a better judge of whether the AI is good enough. I would never answer yes to that question.

 

[merni]

Two questions:

First, mostly just because I'm curious, does the AI know that my empire is "the player?" By that I mean, does the AI interact with the human player differently than it does with other AI's?

Second, the AI in CK2 has very clear modifiers depending on your difficulty level. The "accept matrilinear marriage" proposal for example, clearly showed that the chance of acceptance would be lower if you played a higher level. Is the Stellaris AI similar in that regard? Are modifiers in place at higher difficulty levels to make it more difficult for the AI to accept propositions? I guess, more generally, what makes the AI "harder" at a higher difficulty level?

The AI knows if you are human or AI. Most of the time it will not affect it's behaviour but sometimes, for example when it is allied to you and you are in a war it will act differently with the goal of making it easier for you as a player to plan your strategics.

The second part I think @Wiz can give a better answer to. Generally since economics is a very important aspect of getting powerful in our games the difficulty level can easily be tuned in aspect to that.

 

[Knotz]

Does the AI at all care about whether the research they're doing is flagged 'dangerous?'

 

[merni]

I have a technical question about AI concerning a stupid mod idea I have. When I read this post by Wiz, I began to suspect I might end up feeling "bored" by the idea of AI personality defined very statically this way. I understand there is more than this file that goes into the calculations, but for now, this is what I have to work with. When I saw this I thought, "Why not introduce some variance into this file?" In my mind, if I create a bunch of copies of each personality, but with a small amount of randomization, this could be potentially very interesting.

My question then is this: if I create a supremely large number of possible personalities, will this effect performance? Will the game keep all personalities in memory, and think about one empire as "being a Hedgemonic Imperialist" or does it keep a copy for each empire separate (i.e. here are the traits for the Federated Zracon Nations). The former, I suspect would cause issues with scaling, but the later, in principal, wouldn't feel the impact of having millions of personalities as it keeps one per empire.

I do not believe additional personalities will affect the performance but if it would you can rest assured we would fix it.

 

[Mackus]

If I force spawn dozen predefined by me AI empires, whose primary species have the same name, picture, traits (such as dozen different Humans, or dozen Blorgs), will game recognise them as the same species (for purposes of xenophobia, habitability of homeworld, etc.)? I ask because different empires of same species were shown to be supported in game (mostly rebel split-offs), though it wasn't shown if they are supported at gamestart.
I think they ought to, given that there is semi-cannon split-off Commonwealth of Man.

 

[Jorlem]

@merni
I noticed in Arumba's videos that there is a lack of focus fire in the combat AI.

Is this something that the dev team has considered adding as an "approach option?" He raised a few good points about it.

I was watching those vids too, and at the start of episode seven, when he is fighting the mining drones, the drones do a fairly decent job of focus firing on one of his ships at a time, so I wonder if it might be a weapon thing. Arumba's ships had missiles, while the drones had lasers, so perhaps lasers can focus fire, while missiles get to ignore shields?

 

[Drow7]

How reactive are AI in fleet designs?

And will they avoid bad matchups or they need to get trashed first to "adapt" to their enemies?

Will they have several designs per ship size for multiple responses or just single design they constantly change?

 

[Trithemius]

I was watching those vids too, and at the start of episode seven, when he is fighting the mining drones, the drones do a fairly decent job of focus firing on one of his ships at a time, so I wonder if it might be a weapon thing. Arumba's ships had missiles, while the drones had lasers, so perhaps lasers can focus fire, while missiles get to ignore shields?

Seems like a balance thing to me really - targetting profiles might be unlocked by battle computer upgrades? Or it could just be because they don't want to make human micromanagement unbeatably good for the simulated players? Or it could be because the game is not representing a tactical level of play?

 

[merni]

Does the AI at all care about whether the research they're doing is flagged 'dangerous?'

The AI chooses tech through scripted weights. I don't remember specifically how it handles dangerous techs but I would assume they have a low weight.

 

[merni]

Hi merni. What is the heuristic for AIs deciding to enslave or free pops?

It is based on the personality and policies of the AI.

 

[TheBeautifulVoid2]

It is based on the personality and policies of the AI.

Thanks!

What does that mean for Sector AIs? Will they enslave and/or free pops that were free and/or enslaved before those planets were placed under sector control?