[NFP] AI crazy about Science

[Eva-kun]

Kinda strange seeing Civ with custom districts not build many of them since they're spamming Campus districts.

 

[iammaxhailme]

Anyone wondering why AI is mostly focused on science for the last... let's say... half-a-year?
It is because of this line, added in Victories.xml file with September 2020 patch:
<Row ListType="ClassicalYields" Item="YIELD_SCIENCE" Value="150"/>
It increases the internal value of science yield by 150% and this stays for the entire game. It overshadows other yields, like Gold or Faith. AI almost stopped building Commercial Hubs. No more missionary carpets. And poor Great People are wandering around not even in Russia (they always did so in Peter's empire) but basically in every civ - because there are not enough Theater Squares to put their Great Works into.
Change this value to a reasonable 30-50 if you want to have a bit more diverse game.

This reminds me of the aliens colonial marines AI...

 

[aieeegrunt]

Kinda strange seeing Civ with custom districts not build many of them since they're spamming Campus districts.

If the AI has priority to spam campus districts set as high as this thread seems to indicate, it’s spam them even as it goes bankrupt and is forced to disband it’s army

Meanwhile the barbarians get hugely buffed by it

 

[Infixo]

By the way, @Infixo, I've glanced into the Expansion2_Victories.xml and saw that both laser projects are not listed among AI favoured projects, unlike other SV projects. Could this be reason that AI tends not to build those projects after it launches Exoplanet Expedition?
Would adding those two lines here be enough to change anything?

Most likely yes. Probably.

 

[goatdeer]

To think about this again, this could also be the reason behind Barbarian Men-at-Arms in the Ancient era.

By the way, @Infixo, I've glanced into the Expansion2_Victories.xml and saw that both laser projects are not listed among AI favoured projects, unlike other SV projects. Could this be reason that AI tends not to build those projects after it launches Exoplanet Expedition?

Spoiler :

Would adding those two lines here be enough to change anything?

Well, when this happens, I would have lost so many of my games haha! Every time when the AI launched its last rocket, I knew I had 50 more turns to patch things up. Now this could make things really exciting. That is, on the condition that the AI knows how to generate enough power or aluminum. PS: I do use Real Strategy by @Infixo and as far as I'm concerned, this has not affected their preference towards the laser stations.

I am too seeing a decreasing variety in districts among the AI. I can't seem to put my spy anywhere. Conversely I don't feel that the AI's science has improved somewhat. There are a couple of good science civs and the other ones are focusing on culture, religion or military. This has been very constant throughout my games as of the last year. This could also be the work of Real Strategy.

 

[Infixo]

PS: I do use Real Strategy by @Infixo and as far as I'm concerned, this has not affected their preference towards the laser stations.

Because they are not marked as favored in RST which is something I missed when updating the mod for GS. I will make that tweak in the upcoming update for sure, I am also curious to see if this helps.

I am too seeing a decreasing variety in districts among the AI. I can't seem to put my spy anywhere. Conversely I don't feel that the AI's science has improved somewhat. There are a couple of good science civs and the other ones are focusing on culture, religion or military. This has been very constant throughout my games as of the last year. This could also be the work of Real Strategy.

The whole point of RST is to make Civs play to their strengths, so if you see Civs that actually try various strategies then it means that RST is working as intended.

 

[Smity2k1]

Because they are not marked as favored in RST which is something I missed when updating the mod for GS. I will make that tweak in the upcoming update for sure, I am also curious to see if this helps.


The whole point of RST is to make Civs play to their strengths, so if you see Civs that actually try various strategies then it means that RST is working as intended.

Can't wait to play your new version, once I figure out why Civ keeps crashing every few turns after the last update...

 

[goatdeer]

The whole point of RST is to make Civs play to their strengths, so if you see Civs that actually try various strategies then it means that RST is working as intended.

Sure! My point was that the developer's science update discouraged the AI to build anything else but campuses, but its strategies are still influenced by RST, which I experience in my games and which I also find to be fun. That is probably why I am not seeing so many commercial hubs, since they do not gear towards a particular victory condition.

For people not playing with RST, the science generated by AI must go through the roof I suppose.

 

[Linklite]

For people not playing with RST, the science generated by AI must go through the roof I suppose.

I had noticed that while I easily outpace the AI in gold, culture and faith (if I'm playing the religious game) and I used to outstrip them in science, for a while I've really struggled in science to be clear leader. I've even had problems being top 1 or 2 in the early game.

The AI doesn't seem to use it though, I usually still end up fighting outdated units.

 

[Pfeffersack]

The AI doesn't seem to use it though, I usually still end up fighting outdated units.

I suspect a combination of lacking founds and/or priority to use them for upgrading and partly also miss-managing strategic ressources (issues withnot settling aggresively enough towards them, failing to consequently improving them and the speed-independent AIs purchase limit of 40 for the early strategic ressources on Epic/Marathon gamespeed)

 

[Kotofei]

I DID notice that the Commercial Hubs and missionary carpets were missing, and yes, that's VERY disappointing. I will absolutely make that tweak to mitigate it. There are plenty of things wrong with the AI but that wasn't really one of them.

The AI building more campuses didn't make a difference, it just means my own tech snowballs faster after I capture them.

If I'm reading this right, the AI is supposed to go balls-to-the-wall science in the Classical Era, then let up? That jives with observation. Could I make it continue the science push by copying this line into later Eras, editing the Era name appropriately?

I guess that could backfire and make the AI an unstoppable science superpower on Deity, but hey, it'd be something different at least. I could always just disable SV.

 

[PSG]

This hilarity brought to you by the Department Of Zero Playtesting

I am sure they playtest things thoroughly, but they probably use casual gamers or worse yet console-gamers to test things.

 

[Askia Muhammad]

I'm curious - where would I find these AI priority lists etc?

I'd like to adjust them and see how it alters their gameplay. Specifically, I want to raise their valuation of cities to hopefully provoke some aggression beyond the classical era.

Also, with these priority trees - do all AI utilize a single set? or are there custom ones for groups of civs? If there aren't would it be possible to make custom priority trees for groups of civs (I think it would be too hard to make a civ by civ one if thats even possible) like say warmongering civs (Alex, Genghis etc) have a higher valuation of cities, a preference for domination, and higher value for unit building etc, whilst Korea and science civs could have the current design with higher value on campus?

 

[Socrates99]

It used to be over emphasis on holy sites that prevented the AI from keeping up. That's why Kongo AI was always one of the best competitors. It wasn't really because it didn't spam campuses. I personally rarely spam campuses, it's often the second or third district in my cities.

The better move to make would have been to make it prioritize a CH or Harbor in every city first. Might not have always been optimal but if programming AI is as dumbed down as prioritizing one district over another itd be the best catchall choice. It'd give it the gold and trade routes that provide culture, science and faith along with gold. Then it could use the gold to build other stuff. The side benefit being that it could maintain a standing army. Another side benefit is that the best campus adjacencies are often in the second or third ring. If it's not the first district theres a good chance the AI will have a better choice by the time it's ready for its second district.

 

[MrRadar]

Well, I've added laser projects to the AI favoured items:

Spoiler :

and ran a game on a standard size continents map and standard speed, Deity AI, no modes, and as a spectator (Better Spectator Mod). I've also reduced the science yield that @Infixo mentioned in the OP (<Row ListType="ClassicalYields" Item="YIELD_SCIENCE" Value="150"/>) from 150 to 40.

The results were that as long as AI had enough aluminium, they did build multiple Lagrange lasers after launching Exoplanet, as soon as they got the required tech, but they were building only Lagrange, no love given to the terrestrial laser stations.

The end game turned into quite an exciting race: Robert launched EE first but he did not have aluminium, so his ship travelled at 1 ly, Mvemba launched second, and had just enough aluminium for 1 Lagrange, so his ship travelled at 2 ly, and it seemed that he would win. Then Gilga launched and started to build Lagranges like crazy, as he had quite a stockpile of aluminium. He built them until he travelled at 7 ly... and on t350 won the race! Seondeok also launched and was building lasers, but it was too late for her.

Spoiler :

So it looks like, AI in Civ VI has indeed been coded not to try to win too soon.

 

[Infixo]

@MrRadar Nice test and awesome results
If I had to guess why AI doesn’t build Terrestial Stations it would be probably for the same reasons that I rarely do that - its easier to get aluminium and also power mgmt is more complex that just stockpiling resources.
AI improves resources so aluminium accumulates „by itself”. Then as you observed - civs that have it - build lasers. Others can’t buy it because this is probably not coded.
And terrestial station is more complex -AI would need to analyze power reqs and prepare IN ADVANCE required infrastructure e.g. Wind farms, hydrodams, etc. This is much more complex than just stockpiling aluminum. Chances are we will never see AI building a terrestial laser.
To test that one would need to play a bit as AI, prepare an infra that provides an abundance of power and then see if AI can detect that and build a terrestial laser.

 

[MrRadar]

@MrRadar
And terrestial station is more complex -AI would need to analyze power reqs and prepare IN ADVANCE required infrastructure e.g. Wind farms, hydrodams, etc. This is much more complex than just stockpiling aluminum.

Is it indeed so difficult for the AI? AI does spam solar plants and windfarms late game, far too many for what's required for their cities, judging by the looks of it. And as AI continues to build multiple spaceports, a few cities could have some energy to spare for a terrestrial laser, couldn't they? Coastal areas do turn into offshore windfarm carpets. Maybe again some line is missing here in some file?

 

[kaspergm]

What happens if you add the Laser project but NOT the Lagrange project to the favoured list?

 

[MrRadar]

What happens if you add the Laser project but NOT the Lagrange project to the favoured list?

Well, I've removed Lagrange project from the AI favoured items, left Terrestrial in, and spectated another game on the otherwise same settings. The results were... breathtaking

Freddy launched EE first and after a short while proceeded to build two... Lagrange station projects, despite having plenty of surplus power:

Spoiler :

Where he got the aluminium, I have no idea: as far as I saw, he had 0 aluminium before and after those Lagrange lasers, so perhaps he did trade for it?
So he built two of them and reached 3 ly and there were no more German laser projects.

In the meanwhile, Teddy and Seondeok also launched their EE, after Freddy but before Freddy's Lagrange lasers, so they were falling seriously behind. At this point I was almost convinced there was some laser code lacking, but then, oh holy miracle, when Freddy's spaceship was more than halfway there, America and Korea woke up and did indeed start building a couple of terrestrial lasers:

Spoiler :

And they were chaining them, one after another in the same city, although renewable power was already at the limit there, but then fossil fuel power would have probably kicked in again. But they were too late, got only to 2 ly, when Freddy arrived at t335

Spoiler :

So AI has the ability to build both types of lasers, did not go for terrestrials in the first game though, for some reason.

 

[Dotsworthy]

What difficulty setting are those victories on? Could it make a significant difference on difficulties like Emperor and Immortal?

The other thing that seems to have been implemented is that the AI get a significant discount on great scientists after a human player claims one, from renaissance era onward. I couldn't however find this in any of the XML files so possibly hidden away in the source code.