Improving tiles, bad AI behavior inherited from Civ5

[Magma_Dragoon]

Playing a continents game, I just finished steamrolling ARC on the other continent. I noticed most of her tiles were unimproved, and I thought to myself "no way did they do this again".

They did. Just like civ5, the AI loves to build settlers and hates building workers. AI Brasilia has a fairly high value, but his obsession with military rather than infrastructure in the early game causes him to run out of steam like Montezuma before him. This value should be 10 for every leader.

LEADER_RUSSIA|FLAVOR_TILE_IMPROVEMENT|5
LEADER_INDIA|FLAVOR_TILE_IMPROVEMENT|5
LEADER_POLYSTRALIA|FLAVOR_TILE_IMPROVEMENT|5
LEADER_AFRICAN_UNION|FLAVOR_TILE_IMPROVEMENT|6
LEADER_FRANCO_IBERIA|FLAVOR_TILE_IMPROVEMENT|6
LEADER_BRASILIA|FLAVOR_TILE_IMPROVEMENT|7
LEADER_ARC|FLAVOR_TILE_IMPROVEMENT|5
LEADER_PAN_ASIA|FLAVOR_TILE_IMPROVEMENT|4

 

[Roxlimn]

Technically, it was inherited from Civ 4 - that's the game where the tile improvement behavior of the AI was really, really legendarily bad.

 

[Strategist83]

Automated workers sit at the root of many of the AI's issues. This was exactly the same in Civ V. Automated workers concentrate too much on connecting cities with roads and connecting strategic resources at the expense of everything else.

A minor update to this and/or having the AI build 2-3 times as many workers as it currently does would mean a far, far stronger AI.

 

[KrikkitTwo]

Just increasing the workers would be enough.
(AI flavor 0 should be 1 worker/city..5 should be 2.5 workers/city..10 should be 4 workers/city)

 

[omniclast]

Seems like the tile improvement issue is worse than in Civ5, since a lot more workers are needed and some improvements take forever.

 

[Thormodr]

Technically, it was inherited from Civ 4 - that's the game where the tile improvement behavior of the AI was really, really legendarily bad.

The problem is magnified greatly in Civilization 5 and its progeny Beyond Earth. To have 1UPT in the game, one of the nasty side effects is that it takes forever to improve tiles. At least in cIV, the AI when it put its "mind" to it could finish improving a tile in a reasonable amount of turns.

 

[Bandobras Took]

To be fair, I didn't notice the AI doing anything spectacular in SMAC, either. Civ 3 is probably the only time the AI has done a semi-competent job of tile improvement, and that's because there were only two options.

 

[Thormodr]

To be fair, I didn't notice the AI doing anything spectacular in SMAC, either. Civ 3 is probably the only time the AI has done a semi-competent job of tile improvement, and that's because there were only two options.

Indeed. The AI does fairly well when it is not overwhelmed with choices. Hence why it is so terrible with 1UPT.

 

[m15a]

Indeed. The AI does fairly well when it is not overwhelmed with choices. Hence why it is so terrible with 1UPT.

What does worker actions have to do with 1upt? How does it increase the number of choices?

 

[Bezurn]

Indeed. The AI does fairly well when it is not overwhelmed with choices. Hence why it is so terrible with 1UPT.

This is not the 1upt thread you are looking for... its over that-a-way.

Besides, workers stack with military units so there is no bottle neck for them as long in their own territory (barring scouts from open borders).

In relavance, I do often see the AI not improving tiles, and the AI's that are doing well are the ones that do improve the tiles.

 

[Roxlimn]

Frankly, I've always thought that stacking workers to hyper-improve tiles was kind of a player exploit in other games. I certainly didn't see the AI doing that to any meaningful degree.

 

[Fabio1701]

I would have expected that 1upt would have improved the worker programming. after all, I remember in civ3 seeing a stack of 50 AI workers moving in unison across the map doing nothing. Just put the to sleep until you need them! This was one of the reasons why the game turns took so long for the AI, moving stacks of millions of unemployed workers.

Once, I attacked the stack and killed them all just to make the game speed up. and it was an ally.

It cant be that hard to create a script where a city builds a worker when it reaches size two and that worker does nothing but improve the one city it was built in?

 

[Ryika]

If you just change the flavor, then every AI will get broke in the late game, because it will spam terrascapes like crazy. Probably needs to be combined with some economic bonuses - CivBEHandicapInfos.xml doesn't have an improvement%-modifier just for the AI, so ... yeah. That's that. Could of course lower the building maintenance or something, but that would have a high impact on other things as well.

I think without teaching the AI how to use highend-improvements properly instead of just picking the one with the overall best yield, there is no "quick and easy" solution.

 

[Drowsy Emperor]

The AI seems to be very chaotic. In Civ IV I can see it doing little else than connecting strategic resources, roads and a few basic improvements... but sometimes I can also see it throwing down cottages everywhere and using up most of its tiles.

Regardless, it doesn't play a very efficient game and since lost turns are just magnified down the road ends up hopelessly behind without cheats.

I don't know why they couldn't program the AI to make good bases. Not hyper efficient, specialized cities, but all around good cities. It seems like a rather routine thing to do compared to making it competent at warfare.

 

[Velasti]

It is. They could have used a genetic algorithm and a neural net to train the AI to place sensible quantities and qualities of tile improvements

 

[Thormodr]

This is not the 1upt thread you are looking for... its over that-a-way.

Besides, workers stack with military units so there is no bottle neck for them as long in their own territory (barring scouts from open borders).

In relavance, I do often see the AI not improving tiles, and the AI's that are doing well are the ones that do improve the tiles.

The point stands. When the AI runs into situations where it gets overwhelmed with options (improving tiles or 1UPT) it tends to get befuddled.

 

[Gort]

Maybe the system would be easier on the AI if they used the Call to Power system of Public Works. No "worker" units, you just spend gold to buy improvements in your territory. It would help the AI no end just to not have workers to lose to aliens/barbarians.

 

[Thormodr]

Maybe the system would be easier on the AI if they used the Call to Power system of Public Works. No "worker" units, you just spend gold to buy improvements in your territory. It would help the AI no end just to not have workers to lose to aliens/barbarians.

Yeah. There is some merit in that. That was one of my favourite things in the Call to Power series. That and playing as Canada.

 

[HorseshoeHermit]

The flavor shouldn't be improving tiles, it should be valuing yields.
Yields = good. Yields = the good thing. It's what you have cities for.

I can appreciate the personality flavours of the AI, but at some level there's just some basic ways the AI has to play the game. Moreso in BE where there's really nothing gained by having Rejinaldo be a space Moctezuma.

 

[Magma_Dragoon]

Total number of tiles improved by the end of the game would not be affected by unit stacking, only how fast each individual improvement is completed.

If you just change the flavor, then every AI will get broke in the late game, because it will spam terrascapes like crazy. Probably needs to be combined with some economic bonuses - CivBEHandicapInfos.xml doesn't have an improvement%-modifier just for the AI, so ... yeah. That's that. Could of course lower the building maintenance or something, but that would have a high impact on other things as well.

I think without teaching the AI how to use highend-improvements properly instead of just picking the one with the overall best yield, there is no "quick and easy" solution.

Well we could just adjust the flavors for terra-

Why is the Improvement_Flavors table empty?! Is this why they build so many terrascapes?