Socrates99
Bottoms up!
Maybe, but I've heard way too much about crunch times and unreasonable demands made on the plebes in the gaming industry to chalk it up to ineptitude. As the saying goes, the fish rots from the head down.
[NFP] AI crazy about Science
- Thread starter Infixo
- Start date
Aristos
Lightseeker
I never discarded any of them... that is my point, in fact... why disregard the most simple explanation, which is ineptitude? Again, look at the general demeanor in some of them...
Not necessarily. Sometimes it rots from all angles.
Archon_Wing
Vote for me or die
- Joined
- Apr 3, 2005
- Messages
- 5,257
If anything AI science focus is nowhere near sufficient especially late game.
But not building trade hubs is pretty bad, even if I like stealing all the Great Merchants.
Although even more bizarre is that it shouldn't be hard for AI to build.... 2 districts? But they often don't even in large cities.
But not building trade hubs is pretty bad, even if I like stealing all the Great Merchants.
Although even more bizarre is that it shouldn't be hard for AI to build.... 2 districts? But they often don't even in large cities.
Last edited:
DizzKneeLand33
Fall from Heaven 2 still rocks
Part of the issue is that some brilliant soul that was programming the game thought that by simply raising science focus that all science problems would be solved. But, the original designers of the game made it so that science for some civs is increased by first increasing faith/culture/etc -- without the base districts for that type of civ, they will never compete.
I mean, you can increase science quicker in some instances through gaining a religion than by spamming campuses.
MrRadar's numbers are just creating a beautiful game again. I suggest you all use them, and then you will see what we mean (the ones who aren't on a console, of course -- if you are playing on a console, I suggest changing over to where you can mod the game).
I mean, you can increase science quicker in some instances through gaining a religion than by spamming campuses.
MrRadar's numbers are just creating a beautiful game again. I suggest you all use them, and then you will see what we mean (the ones who aren't on a console, of course -- if you are playing on a console, I suggest changing over to where you can mod the game).
MrRadar
Deity
- Joined
- Nov 8, 2014
- Messages
- 2,318
You know what's funny? Whatever are the science preference numbers: post-April patch, or rolled-back pre-April ones or corrected to have increasing bias with later eras, AI finish times looks like same-ish. AI behaves erratically and lacks focus in any case.
Although it seems that increasing science yield preference in the later eras may be the way worth exploring more. In this case AI has time to establish other infrastructure, including faith and gold economy, and culture, and then raising science focus helps AI to reach 600-800 bulb output per turn or so late game, they have lots of units and more AIs attempt a SV. Arguably this may also funnel them towards a SV in detriment to other types of victories, but as it is very difficult for the AI to stumble upon those victories earlier than SV anyway, maybe this sort of damage is not so big.
I ran another game with increasing yields like one of the tests before, with these settings
Spoiler :
and AI does the best for the yields, in comparison to other tests. Trajan blobbed and won SV on T324.
Classical is reduced to 15, Medieval left unchanged at 15, and I added three lines for Renaissance 25, Industrial 50 and Modern 100.
It would be interesting to experiment more with different values and see how they change things. If anybody would like to give it a go, it is very easy, just get Better Spectator Mod from the Steam Workshop, perhaps Enhanced Camera as well, for better overview, and you're good to go. The game plays itself, no need to press Next turn, it just stops during Congress votes, then you return to the map and press Shift+Enter to continue. Those pauses are also good opportunities to click and look around the map in peace and quiet. It is very revealing to watch AI only game. And saddening too.
And, @Aristos, I don't know how to write a releasable mod, my level of 'modding' is to open those files with the Notepad and change some numbers or copy paste some lines. That's it. I know this is not the best way, but I also backed up original files just in case
Anyway, for the endgame, it would be good to teach the AI not to build a carpet of GDRs that are then running around low on uranium all the time, and that's just one among million other things.
By the way, the Deforestation bug that accelerates Climate Change phases screws AI even more than I thought. They start the Flood Barriers, Deforestation jumps from 0% to 50%, then before too long phase 7 is reached and AI is stuck with astronomically expensive FB like this:
Spoiler :
And AI will not switch away, AI build to the end and complete them.
DizzKneeLand33
Fall from Heaven 2 still rocks
Can someone tell us how to eliminate the climate change deforestation bug, or is that hard coded?
Kotofei
Warlord
- Joined
- Oct 26, 2020
- Messages
- 139
I tried a couple spectator games, and, this.
Did you figure out how to make it build fewer wonders? I saw the AI waste a lot of critical early-game production on ones it doesn't need.
aieeegrunt
Emperor
- Joined
- Jan 8, 2021
- Messages
- 1,611
Depends on whether it’s a parameter that can be got at with Notepad Tech
MrRadar
Deity
- Joined
- Nov 8, 2014
- Messages
- 2,318
I've reported it a year ago in the Bug subforum, but no avail, so far.
No, I'm afraid this is beyond my abilities
https://forums.civfanatics.com/threads/real-strategy-ai.640452/#post-15324159No, I'm afraid this is beyond my abilities
Section about pseudoyields. Wonders are affected by PSEUDOYIELD_WONDER.
MrRadar
Deity
- Joined
- Nov 8, 2014
- Messages
- 2,318
I've varied the numbers again and ran another test. Here's Kongo's tech tree close to the endgame:
Nanotechnology is tainted in the eyes of the AI. Kongo goes for Off-world mission first.
Scotland did exactly the same. Scotland had yet to research Plastics, but they went for Future tech first. Then Plastics. Then Future tech again. Then Synthetic Materials. Then Future Tech. Then Composites and Nanotechnology with no Future tech in-between this time.
Kongo did something very similar. They both and someone else were on the Moon around t250-260, but then AI actively sabotaged their own progress until about t300, and then they got on with the space race again. And that's pretty much consistently happens every game, from what I've seen.
You can tweak those yield preferences however you want, it looks like AI is indeed coded to avoid winning too soon, certainly not before t300 on Deity. The only difference is, that in this case AI do have a tonne of units at all times and there's a lot of activity, albeit mostly futile.
I've also deleted Lagrange preference, left only terrestrial lasers in, and this time there were positive results. As soon as EEs were launched, all the civs who launched immediately started to build them en masse and one after another. And, as it is with the resource consuming units, AI does not know when to stop, they overbuilt them to the point of power failing and spaceship slowing down again, but here AI can come back and build some more renewables and bring those cities online again. There was a race, Kongo launched first, but Scotland managed their power better and won a SV on t328.
Spoiler :
Nanotechnology is tainted in the eyes of the AI. Kongo goes for Off-world mission first.
Scotland did exactly the same. Scotland had yet to research Plastics, but they went for Future tech first. Then Plastics. Then Future tech again. Then Synthetic Materials. Then Future Tech. Then Composites and Nanotechnology with no Future tech in-between this time.
Kongo did something very similar. They both and someone else were on the Moon around t250-260, but then AI actively sabotaged their own progress until about t300, and then they got on with the space race again. And that's pretty much consistently happens every game, from what I've seen.
You can tweak those yield preferences however you want, it looks like AI is indeed coded to avoid winning too soon, certainly not before t300 on Deity. The only difference is, that in this case AI do have a tonne of units at all times and there's a lot of activity, albeit mostly futile.
I've also deleted Lagrange preference, left only terrestrial lasers in, and this time there were positive results. As soon as EEs were launched, all the civs who launched immediately started to build them en masse and one after another. And, as it is with the resource consuming units, AI does not know when to stop, they overbuilt them to the point of power failing and spaceship slowing down again, but here AI can come back and build some more renewables and bring those cities online again. There was a race, Kongo launched first, but Scotland managed their power better and won a SV on t328.
Spoiler :
Archon_Wing
Vote for me or die
- Joined
- Apr 3, 2005
- Messages
- 5,257
It's kind of a result of spreading misinformation that gets parroted everywhere. It's a popular thing to do some reductive crap without nuance (eg "Spam Campuses and win") while missing the subtleties. And yes, there's plenty on these forums here. A lot of people can certainly complain, but they don't actually know why it's happening!
And we'll skip the obvious stuff like not building hubs/Izs, as that is really bad and has no excuse.
It is true that we win games by pushing science. It is also true that the campus is a very efficient district and you need a lot of them because wide. But it is not true that it starts campus, campus, campus and many good players do not start out spamming campuses. If one pursues this method, then they will find themselves missing out on Eurekas both from researching "too fast", and the fact that eurekas are also determined by getting inspirations. It's often a good idea to pursue culture (but not necessarily from theater squares) to get policies and maybe even certain wonders. And of course expansion over infrasture early game.
Then late game you also need the Royal Society/and t3/t4 govs too.
It's honestly pretty not intuitive at all. And even I have been guilty of spreading this spam campus meme. And also while it's easy to say "Firaxis is big dumb dumb", a lot of people weren't exactly able to spell it out like I just did either. So.....
Kotofei
Warlord
- Joined
- Oct 26, 2020
- Messages
- 139
This seems to work, thanks. The AI appeared to build more useful stuff instead of, like, Meeknashi Temple.
It'll take some tweaking on my part, though. I set it to -95 for the ClassicalYields, which worked, but then the AI's hit it hard as soon as they hit Medieval, to make up for lost time. My gut is that scaling it by era something like -90/-60/-30 would work.
This seems to be the biggest takeaway of your research. (outside of a lucky religious victory)
Socrates99
Bottoms up!
Its so common in the strategy and tips section to have to tell people to worry about campuses later. In the ancient and classical it's all about infrastructure, getting the t1 govt and expanding. I'll only build campuses early if I have great adjacencies. I might lock a campus production cost in (something players do that AI doesn't) but switch to other priorities.
I feel like my deity advice could be copied and pasted to half the threads there "expand, defend, get Political Philosophy asap..."
Pietato
Platonic Perfection
This better be a joke. Bethesda has never released a completed game in their life, and consistently hand-off broken games to their fans to fix.
aieeegrunt
Emperor
- Joined
- Jan 8, 2021
- Messages
- 1,611
Yup, and they were smart enough to realize that it works BECAUSE they fostered that devoted mod fan base and bent over backwards to accomodate it
I and millions of other gamers bought New Vegas, Skyrim and Fallout 4 despite the hilarity of Fallout 3 and Oblivion at launch because we knew the modders would be fixing it at a launch AND adding a ton of content for at least a decade.
Literally millions of copies of Skyrim Special Edition sold 5 years aftet launch and the expanded modding potential was the primary driver.
Then they went Monetization First with 76 and the resulting debacle bankrupted them
Just about any Bethesda title at launch had fewer issues and better AI than what Civ6 has now
THAT is the real joke
Larsenex
King
Great thread. I am going to update my game clear all my saves and write in all the code suggestions listed here.
I do not play on Deity but all of these should help the ai across all difficulty levels.
Mr Radar, thank you for the time you spent investigating this issue and double thanks for the xml changes which you have shown do make the game better.
I do not play on Deity but all of these should help the ai across all difficulty levels.
Mr Radar, thank you for the time you spent investigating this issue and double thanks for the xml changes which you have shown do make the game better.
Aristos
Lightseeker
MrRadar
Deity
- Joined
- Nov 8, 2014
- Messages
- 2,318
As it was found out, these three lines were modified/added in the April patch in the Victories.xml file and they gimp the AI the worst:
The simplest change to bring AI back to what it was before the patch would be to delete the 0 in the first line (150 becomes 15) and to delete the second and third line entirely.
If you want to bring up AI science yields gradually in later eras, so that more AIs would attempt a SV, I added three more lines:
<Row ListType="RenaissanceYields" Item="YIELD_SCIENCE" Value="25"/>
<Row ListType="IndustrialYields" Item="YIELD_SCIENCE" Value="50"/>
<Row ListType="ModernYields" Item="YIELD_SCIENCE" Value="100"/>
That gave quite a nice result. I'm just uploading the whole modified Victories.xml file which includes these extended modifications as well, in case this makes things easier.
Pietato
Platonic Perfection
Do you work for the CCP?
Similar threads
