Unescorted Settlers (Deity)

[SeismoGraf]

The AI currently sends out settlers without escort even on Deity difficulty. I absolutely hate it. See attached screen shot for an example.

This Roman settler just begs to be kidnapped. Taking it would mean having a pretty easy game. But what's it good for? I could have selected a much lower difficulty, if I wanted less of a challenge.

Ambushing an escorted settler with several units is legit IMO.
But getting a "free city" on turn 11 Deity is just a joke!

Please change this

 

[Big J Money]

AI, in general, is in poor shape. It's going to affect all skill levels. This happened in Civ V too, right? But Settlers converted to workers so it wasn't as big a deal.

 

[KrikkitTwo]

Since they announced the escort function, I was sure the AI would use it for settlers and builders.

I was thinking they would do another smart thing and have settlers/builders just die when attacked. (since the player can escort them now) [you could even have the settlers give a defensive combat bonus to units linked with them/on their tile]

Suprisingly, they did neither

 

[Furycrab]

Since they announced the escort function, I was sure the AI would use it for settlers and builders.

I was thinking they would do another smart thing and have settlers/builders just die when attacked. (since the player can escort them now) [you could even have the settlers give a defensive combat bonus to units linked with them/on their tile]

Suprisingly, they did neither

The problem is an interesting one. Sending a unit with your settler isn't even something players do all the time. It's often a risk vs reward thing, and often the risk is very low. I imagine if you did AI simulation and one AI was always 100% adamant of having an escort on hand, and the other was a bit looser, sometimes exploring a little more aggressively but didn't always wait for the escort, the one taking the risk would win more often, or at least be in a better position more often, but every once in a while take a heavy hit to a bad decision.

Personally I find that the greater problem is that stolen settlers stay as settlers and you don't get a diplo hit as if you took a city and if you are in the early game you don't even take a hit for the war. If a stolen settler turned into a builder, you might not be able to snowball so hard off what would effectively just be lucky scouting. Maybe the Settler could become a useless "civilian" you can't delete and can't be used to settle until the war is over.

 

[Acken]

or it just become a worker.
or the AI just escorts it. With their ton of warrior I'm sure one is available.

AI vs AI performance based on risk taking is pretty much meaningless at the stage we are in.

 

[rrrwww]

this is another reason why deity is a joke

 

[Furycrab]

or it just become a worker.
or the AI just escorts it. With their ton of warrior I'm sure one is available.

AI vs AI performance based on risk taking is pretty much meaningless at the stage we are in.

One of the first things I said was they should probably change it to be a builder.

I don't know exactly what is leading to these stray settlers. I've caught a few myself and it does feel very bad and "easy" especially when you consider you get almost no diplo hit for doing so. I'm guessing it's committing the warriors it gets to other stuff, and later when the settler pops out and it makes the call that it's relatively safe. However I have no stats on how often it does this or how often it gets caught trying to do this even against players.

I know that AI vs AI is not the same as against a player but it's not meaningless. 100% waiting for an escort is slow and bad which means they would need to give the AI more bonuses to compensate. It's like asking an AI in civ 5 to 100% protect it's workers. You could, but it's completely impractical and you better give the AI a horsehocky ton more stuff if you want it to even remotely be capable of doing such.

I think it might not be assessing a players willingness to just yank a settler in the Ancient or Classical era properly, but in the 50+ games I've messed around in my first week, I've only managed to catch a "really" early settler a handful of times. However I only know my POV and have no clue what lead the AI to decide to make that decision. Most of the time, if it happens, I was just very lucky with a scout and the AI might not have even known the scout was in the area. I'm just annoyed I'm able to snowball really hard off of it, and don't really care if I just caught it with it's pants down if it means the other 19 games where I'm not lucky and catch the settler, it gets away with settling faster and thus offers me more competition with fewer bonus yields.

 

[CaiusDrewart]

Yep. They made two bizarre design decisions here. First, to have captured settlers stay settlers rather than turning into builders. I don't get this. Having them become builders is more thematic (these are people you've just enslaved!) AND is also better, less swingy gameplay. And second, how did they neglect to program the AI to use the "Escort" function with its settlers? Could it really have been that hard to program the AI to never send a settler out without a warrior attached?

Both of these were mistakes. Taken together, they're just disastrous. And this comes up a LOT. Playing on Deity I think I grab an AI Settler in the first 25 turns in roughly 50% of games.

 

[isau]

The funny thing about it is it's trivial to change this bit of code. I did it for the mod I personally use to play. Just applying the same logic to Settlers that is used for Great People fixes this issue with one line of SQL. I don't know why they pressed forward with this still in game. Sure hope it wasn't intentional.

 

[SeismoGraf]

or it just become a worker.
or the AI just escorts it. With their ton of warrior I'm sure one is available.

AI vs AI performance based on risk taking is pretty much meaningless at the stage we are in.

Agree. I've been swarmed with 8 warriors or more in the early game, so warrior availability should not be a problem at all.

And a point on "AI scouting": does the AI really compute its decisions on imperfect map information? Or does the AI "cheat" and look at the whole map? I don't mind the AI cheating, if it improves quality of play.

Playing on Deity I think I grab an AI Settler in the first 25 turns in roughly 50% of games.

I've stopped taking the AI settlers, but it leaves a bitter after taste. Gandhi DoW 10 turns after I generously and kindly ignored his suicidal settler.

 

[Acken]

I know that AI vs AI is not the same as against a player but it's not meaningless. 100% waiting for an escort is slow and bad which means they would need to give the AI more bonuses to compensate. It's like asking an AI in civ 5 to 100% protect it's workers. You could, but it's completely impractical and you better give the AI a **** ton more stuff if you want it to even remotely be capable of doing such.

I think it might not be assessing a players willingness to just yank a settler in the Ancient or Classical era properly, but in the 50+ games I've messed around in my first week, I've only managed to catch a "really" early settler a handful of times. However I only know my POV and have no clue what lead the AI to decide to make that decision. Most of the time, if it happens, I was just very lucky with a scout and the AI might not have even known the scout was in the area. I'm just annoyed I'm able to snowball really hard off of it, and don't really care if I just caught it with it's pants down if it means the other 19 games where I'm not lucky and catch the settler, it gets away with settling faster and thus offers me more competition with fewer bonus yields.

I think you're kind of missing the point of my post.
It's not that not-escorting the settler is always a bad idea. It's that I think an AI developer should probably make its AI play reliably before attempting more advanced concepts. If it was able to better predict stealing then it would be okay for it to sometimes lose one.

Also I reliably get at least 1 in almost every non isolated game so maybe it skews my perspective.

 

[SeismoGraf]

I think in general the AI does "everything it can" to make the player win. For example: not firing/attacking with units, not firing with fortifications, retreating a doomed unit instead of trading 1-for-1, not building a lot of ranged units. A decent AI would make taking (and then keeping) cities impossible with equal local force. Even without strength bonus. Wars would be an economic ruin and fighting the exception. Everything would revolve around the early land grab and other victory conditions.

Defensive wars are easy to win. If the AI was on the level of a good human Civ6 player you would not be able to take any cities from them unless you out-gun them massively. Would this be fun? Maybe, maybe not.

 

[Furycrab]

I think you're kind of missing the point of my post.
It's not that not-escorting the settler is always a bad idea. It's that I think an AI developer should probably make its AI play reliably before attempting more advanced concepts. If it was able to better predict stealing then it would be okay for it to sometimes lose one.

Also I reliably get at least 1 in almost every non isolated game so maybe it skews my perspective.

I guess it would. I'm not actively trying to steal a settler so I get one maybe every other game where an AI is close enough for my scouts to even get in there. Often it's just blind luck and I'm almost ok with catching the AI in such a way because if it was a real player, it would likely be almost the same. Probably doesn't help that I don't like to play Pangaea.

I imagine if I focused on catching the AI in a dumb move, I could bring up the rate at which I pull off these sorts of things. It's just one of those things where I don't find it enjoyable if my strategy involves exploiting the AI, so I just don't actively try. In 5, I was the guy that didn't enjoy extreme worker stealing or using tactics to fish out garrisoned units. I admit, I didn't win as often on Deity, or get the fastest win times because of these self imposed rules... However I'm probably a little closer to the vast majority of players who don't actively try to poke holes in the AI and I still enjoyed the game.

So it's like... should the AI devs give up on trying to give the AI more advanced concepts if it makes it play better against most players just because very few players are actively trying to completely break them?

Either way... in this case, I don't think it really matters, since I think the true problem is mostly that it's just too rewarding to successfully pull it off.

 

[Big J Money]

Really, both workers and settlers should just die when captured by civs.

And if a civ 'rescues' someone else's unit from a barbarian they can choose to return or kill it. But not take it.

But you can still rescue your own units from barbs.

 

[CaiusDrewart]

Really, both workers and settlers should just die when captured by civs.

And if a civ 'rescues' someone else's unit from a barbarian they can choose to return or kill it. But not take it.

But you can still rescue your own units from barbs.

People have suggested a small gold reward for killing enemy civilian units. I like it. It would basically be another form of pillaging.

 

[KrikkitTwo]

People have suggested a small gold reward for killing enemy civilian units. I like it. It would basically be another form of pillaging.

I like that... especially with making it different for barbs.
(the AI would still need to escort its settlers though)

 

[chazzycat]

What I don't understand is how they added an "escort formation" feature to the game, which would be perfect for the AI. It would be so trivial...just stick one of those million warriors they love to build in an escort formation with the settler and send them on their merry way...but they forgot to program the AI to use it

 

[Takfloyd]

Capturing settlers is fun, it shouldn't be removed. The AI just needs to protect them, or at least bother trying to recapture them.

 

[Windscion]

Best of all, make captured settlers into workers with two builds left (so still strictly worse than ones you create). Altho that might be beyond current modding tools, I dunno.