The AI Is Still Bad?

I’m going to do an experiment this weekend, where I go back into my late exploration war I just barely won before the era transition, and watch what happens if I let the AI break through my lines (maybe use those melee attacks my Ming infantry theoretically have).
 
The AI scrambling to settle an iceball city halfway across the continent. This crappy AI behaviour isn't new to 7. 😉

City loyalty was such a fantastic feature in Civ VI. Not sure why they didn't bring it back for 7. It at least restrained the AI a bit.
 
Well, as the creators of the artificially intelligent mod showed, it seems quite possible to improve the AI's settlement behavior without mod tools and without the loyalty mechanic (which by the way took a long time to work properly in civ VI - I can remember getting crappy cities for free all the time, because the AI forward settled to places it couldn't support).
 
The AI scrambling to settle an iceball city halfway across the continent. This crappy AI behaviour isn't new to 7. 😉

City loyalty was such a fantastic feature in Civ VI. Not sure why they didn't bring it back for 7. It at least restrained the AI a bit.
It was great at keeping the AI settling more in line. It did however, make overseas expansion near impossible, without attacking everyone over there to maintain loyalty. The current distant lands sprint would not work at all with 6s current loyalty mechanics.

tldr, it would work fine for homelands but not distant lands.
 
Hard to make it work with Distant Lands.

Fair point.
It was great at keeping the AI settling more in line. It did however, make overseas expansion near impossible, without attacking everyone over there to maintain loyalty. The current distant lands sprint would not work at all with 6s current loyalty mechanics.

tldr, it would work fine for homelands but not distant lands.

Yeah, I get that. They have sort of painted themselves into a corner that way. Not that I think the distant lands mechanic is terrible but it wouldn't work with loyalty.
 
That's how I remember it, too..
With regard to the initial AI with Civ 4, it's not nearly as horrible as Civ 6 was all the way until the end (with all DLC).

I played Civ 3 through 5 for 5k hours each (6 for only 2k). I never had a prayer of beating Civ 4 on Deity from vanilla through BTS. Civ 6 it was a regular occurrence. Civ 5 Immortal was the top for me.

Maybe it's not the AI, but the way the level is calculated. That I could go with.

EDIT: And Acken's Civ 5, well, be prepared to play TWO levels down. So, it's possible to program the AI. The Civ developers just don't care.
 
Fair point.

Yeah, I get that. They have sort of painted themselves into a corner that way. Not that I think the distant lands mechanic is terrible but it wouldn't work with loyalty.
It might be possible to code a similar loyalty system to 6 for the AI, but disable said logic when it comes to distant lands. As long as cities arent flipping from it, it should only matter for settling purposes.
 
The mod that allowed min settling to 1 hex away worked spectacularly for me in CVI. Almost all civs designed complex snakes structures all 1 Hex apart. Combined city bombing was strong defensively speaking. Only a few civs kept cities spread apart a little bit. Japan especially I was always almost certain that it would produce out a snake structure. Maybe reduce the city min distance settling, and also enabling possible settling inside another player borders, even if it would trigger war (sometime only one settler remaining, last city capured, to prevent being wiped out Ai settled in place its last roaming settler where it was, and if by chance it was on a prohibit square, it would settle anyway. Or roam forever in a caravel untill someone find it and killed it if it was on water. 5 hex apart cities is frankly too much. Ai settlers probably forward settle bc of lack of prime land near them. So that simple mod should be worth a chance in the base game.
 
You don't need a game mechanic for the AI to not settle far away. Just fix its plot selection logic.
Exactly. 1 hex min distance mod was a godsent in CIV VI.
Not only it would fix distant forward settling, but it would allow 50+ cities per civilization, given 9 out of 10 settlements remains towns.
Each city would have 8 - 9 towns all closely packed. Also nice snake structures would emerge, even if one settlement is lost, the connection is barely broken. With the new city range mechanics makes it even hared for players to take down such structures. They are highly defendable with combined ranged defence.
 
Last edited:
The problem of forward settling has 2 parts:
1. Gameplay limitations, which could work in MP the same way as in SP
2. AI logic to follow those limitations

Loyalty was part of the first, limiting everyone and the game probably should have something additional, but so far just inevitability of wars is enough.
Second part is much easier and if I remember correctly, planned for the next patch on Mar, 25.
 
I will say the AI is more adept at winning the game than Civ 6. That's not saying much, I know. And oh boy are they ruthless on those city states/independent peoples. Bloodthirsty. Okay, they really have only been able to win one way in my games: science victory. They can get the great banker in economic victory, but I was watching Ashoka's great banker, and he was just doing stupid things with it.
 
I will say the AI is more adept at winning the game than Civ 6. That's not saying much, I know. And oh boy are they ruthless on those city states/independent peoples. Bloodthirsty. Okay, they really have only been able to win one way in my games: science victory. They can get the great banker in economic victory, but I was watching Ashoka's great banker, and he was just doing stupid things with it.
Economic victory requires significant amount of gold and influence (especially after patch), I think AI just find it hard to hoard enough. I was in this situation myself - while being at war with several AIs at once, I was throwing all my gold and influence there. Sure, I was able to reprioritize things and win economically, but I just had fun with war.
 
You don't need a game mechanic for the AI to not settle far away. Just fix its plot selection logic.
Adding one line that says "don't settle more than 10 tiles away from a current settlement" would improve the AI so much!
 
Adding one line that says "don't settle more than 10 tiles away from a current settlement" would improve the AI so much!
While also blocking distant land settling in many cases and creating weird settlement chains in old world.

AI logic is usually quite complex and need to take into account many parameters.
 
Some consolidated thoughts on settling (those are scattered here or there already, especially in the AI mod thread):

- Capt. Obvious: The AI should simply settle the good spots close to its territory as a priority. (This is improved by @notque s mod)
Interestingly, the (base) AI settles smarter on Fractal (imo). Less options = less confusion, maybe?

- The major AI problem seems to be the silly tile evaluation (mouse over the suggestions on the settler lense near borders - the AI doesn't get that resources are already taken by other civs) - this is a core logic problem that FXS needs to deal with. (They probably will - improvements are on the roadmap and it's the most glaring issue)

- Forward settling is not the problem in itself, it's actually encouraged by game mechanics. (Trading range + resource grabs in particular - but also settlement limits mean that you need productive settlements.) The AI should also be able to try that as a last resort when boxed in - and obviously any limitation needs to be shut down in Exploration.

- Loyalty mechanics didn't help the AI in Civ6 at all, quite the opposite - as a player you could often easily take cities by loyalty war. It's not easy to program AI to handle this. It got catastrophic in Dramatic Ages mode, where whole continents fell to Free Cities because the AI could not handle the loyalty system.
You could build a similar system just for tile evaluation, maybe? (But again: the mechanics are wildly different- you need to have a wide presence on the map to trade, not to mention Exploration win conditions.)
 
Last edited:
While also blocking distant land settling in many cases and creating weird settlement chains in old world.

AI logic is usually quite complex and need to take into account many parameters.
Distant land tiles are identified in the game. The problem with the current AI is that it puts values on some spot too highly resulting in these patterns.
They'll fix it because as proven by an already existing mod, which doesn't even have access to the internal logic, it can be done by changing a few values.
 
They'll fix it because as proven by an already existing mod, which doesn't even have access to the internal logic, it can be done by changing a few values.
I've looked at the internals of the RHQ mod and while I didn't study it in details, I could say I disagree with both of those statements:
1. The AI logic is written in decision trees inside XML, so mod does have access to internal logic
2. The mod changes much more than a few values, it changes the logic - I see significant parts of the original settling logic commented out and the new one written
 
What I meant is that to understand the tree the moder had to do a lot of trial and error. It also cannot change what those nodes do outside of modifying the input values. My point is that the Firaxis AI dev has even more power to modify the behavior and probably doesn't have to first understand how it works since they made it.
 
Back
Top Bottom