[R&F] AI Pantheon choices

[Victoria]

With regards to the decision making, does the game actually score tiles in a city to evaluate the value of wonders like Petra, Chichen Itza or St. Basils Cathedral, or does thee AI just rank the wonders generically and build them at random if a suitable tile is available and nothing otherwise ranks higher?

The AI has wonder preferences by civ, including negative preferences, I posted them a long time ago.
The decision beyond this small list will be based on their current agenda so a religious agenda will build that type of wonder if one is available. If two are available I believe it randomly chooses between them, I’m fairly sure I have seen the rare occaision of a random decision for wonders,

I would have to look deeper to be surer but that’s fairly true I believe but is anecdotal.... @Magil do you know any more as a wonder mod?

 

[kaspergm]

The AI has wonder preferences by civ, including negative preferences, I posted them a long time ago.
The decision beyond this small list will be based on their current agenda so a religious agenda will build that type of wonder if one is available. If two are available I believe it randomly chooses between them, I’m fairly sure I have seen the rare occaision of a random decision for wonders,

Ok, so no actual tile evaluation. That's quite disappointing, but correlates well with all those zero-yield Petras/Chichen Itzas we see all the time (have yet to see a zero yield St. Basil's, but I'm sure it will be even worth it, given how it doesn't even require a tundra tile to build it (is that an oversight!?).

 

[Victoria]

Ok, so no actual tile evaluation

The tree is see target wonder and does an awful lot of decisions but nothing clear about the value of wonder - more to consider than just tiles

Leaders.xml has the wonder requirements I mentioned by civ

 

[Infixo]

I am necroing the thread to share few observations in regards to the OP.

I did some testing recently, very similar to what @Victoria described in the OP.
So, 7 games, 40 turns each, to see what pantheons AI will choose. I also gathered the scoring for all pantheons from the turn when AI was making a choice. I have 29 sets from 29 decisions.

1. Nothing has changed in GS in comparison to R&F. AI still chooses:
BELIEF_EARTH_GODDESS (5/7 1st choice)
BELIEF_SACRED_PATH
BELIEF_GODDESS_OF_THE_HARVEST
BELIEF_GOD_OF_THE_FORGE
BELIEF_DESERT_FOLKLORE
And basically in that order.

2. What surprised me the most is the fact that 13 pantheons are (almost) always scored at 0. There was 1 single case for 3 of them out 29 where the scoring was non-zero. So, more than half pantheons will never be chosen by AI. Even if you remove those 5 top-picks. Among those 13 are such cases as: BELIEF_GOD_OF_THE_OPEN_SKY, BELIEF_GODDESS_OF_FESTIVAL, BELIEF_STONE_CIRCLES, etc. Pantheons that will give quite a lot of value, certainly not 0

3. Is there a problem with the valuation? Nope. I am pretty sure the valuation works correctly. The problem is that it scores the current situation, not the potential (or future). This is a huge design flaw because pantheon is chosen once, very early, and cannot be changed. So, for the choice to be meaningful, it has to consider the future.
Right now it does not. All valuations are done based on the current situation. And since this early game the civ usually doesn't have any plantations, mines over strategic resources (btw, those are not even yet revealed), camps, etc. thus their valuation will be zero. On the other hand, all feature based pantheons (like +1 faith to HS from tundra or desert or rainforest) are scored against features that are already on the map. Same with the god of harvest. AI scores all possible removals and it gets quite a high number (no matter if they will ever be removed, this is the opposite case).
BELIEF_RIVER_GODDESS was scored >0 only once (out of 29) because there was already a HS in place close to a river.
If the civ is not building a Wonder at the moment - it will never choose BELIEF_MONUMENT_TO_THE_GODS, even if once makes it +500% production to all wonders, ever.
Why AI doesn't take divine spark which is considered as one the best by humans? Because there are no districts that would justify it at the moment. Pantheons are chosen arount T15-40. AI can have a holy site, maybe a campus, but rarely. So, divine spark was scored 5x (out of 29) for 30-50 points. Compare that to usual valuations for top picks - they are around 300-500.
Earth Goddess - obvious, appeal is there on the map, so AI just counts how many tiles are viable.
I can go on, there are many more examples like this.

4. The bad news is that since this is a conceptual design flaw and fixing it would most likely require to rewrite entire pantheon scoring, I don't think Firaxis will ever do that.

 

[iammaxhailme]

I am necroing the thread to share few observations in regards to the OP.

I did some testing recently, very similar to what @Victoria described in the OP.
So, 7 games, 40 turns each, to see what pantheons AI will choose. I also gathered the scoring for all pantheons from the turn when AI was making a choice. I have 29 sets from 29 decisions.

1. Nothing has changed in GS in comparison to R&F. AI still chooses:
BELIEF_EARTH_GODDESS (5/7 1st choice)
BELIEF_SACRED_PATH
BELIEF_GODDESS_OF_THE_HARVEST
BELIEF_GOD_OF_THE_FORGE
BELIEF_DESERT_FOLKLORE
And basically in that order.

2. What surprised me the most is the fact that 13 pantheons are (almost) always scored at 0. There was 1 single case for 3 of them out 29 where the scoring was non-zero. So, more than half pantheons will never be chosen by AI. Even if you remove those 5 top-picks. Among those 13 are such cases as: BELIEF_GOD_OF_THE_OPEN_SKY, BELIEF_GODDESS_OF_FESTIVAL, BELIEF_STONE_CIRCLES, etc. Pantheons that will give quite a lot of value, certainly not 0

3. Is there a problem with the valuation? Nope. I am pretty sure the valuation works correctly. The problem is that it scores the current situation, not the potential (or future). This is a huge design flaw because pantheon is chosen once, very early, and cannot be changed. So, for the choice to be meaningful, it has to consider the future.
Right now it does not. All valuations are done based on the current situation. And since this early game the civ usually doesn't have any plantations, mines over strategic resources (btw, those are not even yet revealed), camps, etc. thus their valuation will be zero. On the other hand, all feature based pantheons (like +1 faith to HS from tundra or desert or rainforest) are scored against features that are already on the map. Same with the god of harvest. AI scores all possible removals and it gets quite a high number (no matter if they will ever be removed, this is the opposite case).
BELIEF_RIVER_GODDESS was scored >0 only once (out of 29) because there was already a HS in place close to a river.
If the civ is not building a Wonder at the moment - it will never choose BELIEF_MONUMENT_TO_THE_GODS, even if once makes it +500% production to all wonders, ever.
Why AI doesn't take divine spark which is considered as one the best by humans? Because there are no districts that would justify it at the moment. Pantheons are chosen arount T15-40. AI can have a holy site, maybe a campus, but rarely. So, divine spark was scored 5x (out of 29) for 30-50 points. Compare that to usual valuations for top picks - they are around 300-500.
Earth Goddess - obvious, appeal is there on the map, so AI just counts how many tiles are viable.
I can go on, there are many more examples like this.

4. The bad news is that since this is a conceptual design flaw and fixing it would most likely require to rewrite entire pantheon scoring, I don't think Firaxis will ever do that.

So do you think the AI are evaluating open sky, etc low because they are counting how many stables they have, NOT how many cows they have? etc

That may explain why I see so many civs take dance of the aurora or desert folklore when they only have a tile or two, when in their place I definitely would have taken festivals, stone circles, etc

 

[Infixo]

So do you think the AI are evaluating open sky, etc low because they are counting how many stables they have, NOT how many cows they have? etc

Yes, definitely.
I have an idea for a solution, kind of a workaround, that would enable AI to make better choices but I wanna make some tests first to see if it is actually possible what I am thinking of. I would like to see AI making more versatile choices, that is for sure.

 

[Karmah]

I discovered Earth Godess recently (in a Maori game) ,found it really really good. Not sure how I'd compare it to harvest , and that's saying something in fact. Before this particular game , I did not know it existed , I guess it was always gone before I could get a pantheon.

Before that it was always open sky for me ,if I can't get god of the harvest (most of the time that was the case) . With GS and a bit of nerf on chopping and good governor choices besides Magnus , it might not be my number one anymore.

 

[iammaxhailme]

I definitely prefer earth goddess to harvest, mostly because I don't like chopping. I find the optimal chop game to be quite boring and not fun to have to constantly minmax

 

[Trav'ling Canuck]

I discovered Earth Godess recently (in a Maori game) ,found it really really good. Not sure how I'd compare it to harvest , and that's saying something in fact. Before this particular game , I did not know it existed , I guess it was always gone before I could get a pantheon.

Before that it was always open sky for me ,if I can't get god of the harvest (most of the time that was the case) . With GS and a bit of nerf on chopping and good governor choices besides Magnus , it might not be my number one anymore.

When GS came out, there were quite a few of comments that Earth Goddess is probably the best of the pantheons now, but since the AI always takes it first, it's very difficult to get.

PS Thanks @Infixo for the follow up investigation on how the AI evaluates Pantheon selection!

 

[Abaxial]

I often find that Earth Goddess is not available, even when no AI has actually picked it. Something odd there.

 

[Phoenix1595]

It is no wonder why the AI is programmed to nab Earth Goddess first. Earth Goddess is a must. I find that in many of my games, the first turns are geared toward a race to Earth Goddess. The AI shoots for it even when not playing religion. Unless you are surrounded by marshes and rainforests, or plan to play a game that would make Captain Planet weep, Earth Goddess has a tremendous return for faith. Even if you do not plan to play religion, the rapid accrument of faith is great for GP buys.

This, of course, means it will be nerfed soon.

I do wish pantheons were culturally linked to each civ, much like civs have a preferential religion now. Of course, the problem is that starting areas do not always correspond to what a civ’s conceivable pantheon might be, given how pantheons are often linked to specific geographic conditions.

 

[zxcvbob]

I have not played Civ 6 in almost a year. Started a game today, and I couldn't remember what pantheons are best. So I took Divine Spark because I know it's good even if it's not the best. And I seem to be getting *all* the great scientists. I think Earth Goddess and Goddess of Harvest were available (I got a pantheon really early, not sure why, maybe from a goody hut)

 

[Abaxial]

Well, one could divert this thread and ask what pantheon is preferable when Earth Goddess has been taken. Divine Spark is nice, but I also like Fertility Rites and Religious Settlements.

 

[Mahi]

Unfortunately I always return to Divine Spark as well. It just seem to be the best all around pantheon to me. In all honesty I think Pantheons are broken. There are a few panthenons that are so good it is ridicolous and then there is the rest, which are quite neglectable.

In my opinion all "powerful" pantheons should only be be boosted by specific geographich/resources tiles (desert tiles/pastures, etc.). Rest of the pantheons should only be mediocre boosts. Top three pantheons should be nerfed at least.

 

[Victoria]

I discovered Earth Godess recently (in a Maori game) ,found it really really good. Not sure how I'd compare it to harvest , and that's saying something in fact.

As maoris do not chop much and are a coastal civ (high appeal) then earth goddess is the best choice specifically for them, and they can get it quite often due to their ability to find CS very fast.

When GS came out, there were quite a few of comments that Earth Goddess is probably the best of the pantheons now

the number of tiles that give good faith is limited and you do not get in the long run anywhere near as much as the harvest goddess if you are a chopper.

It seems everything in this game is based on a sliding scale from OP to crap. This means a good start snowballs your OP value. The game is more luck that judgement resolved to a degree by restarts and reloads.

 

[Infixo]

Unfortunately AI doesn't choose Earth Goddess because it is good but because its scoring routine is seriously bugged. It scores as if all tiles in the city were valid for the pantheon. With that bug, even the crappiest pantheon would be chosen by AI.
For those interested in details: https://forums.civfanatics.com/thre...atter-when-ai-evaluates-earth-goddess.644591/

I did some more testing in the meantime and there are even more errors in pantheon scoring subsystem. Like, a lot...
Here: https://forums.civfanatics.com/threads/1-0-0-317-more-issues-with-ai-scoring-pantheons.644923/
AI probably would be better off just picking them at random.

 

[Bibor]

I think the tile bonus pantheons go up and down in value (compared to, say, GotH) depending on map size. If you're playing normal or smaller, there's way too few resources of the same type around you (apart from maybe quarries or pastures) to make them worthwhile. I never tried it, but in Civ5 (and to some extent in Civ4) the map generator tends to clump up resources by type, especially on a continents script, so you'd have like an area with 10+ Sugar tiles.

 

[blackcatatonic]

In a recent game as Canada, I lucked out massively and got a relic from an early goody hut, so bagged Earth Goddess for the first time ever. Amazing pantheon! I can see why the AI likes it so much. Then I had a quandary with Mali, where again it was available (because desert) but not worth taking because appeal was crappy (because desert) - luckily Desert Folklore was still available.

Usually Divine Spark is my go-to for most Civs/situations.

 

[Phoenix1595]

In a recent game as Canada, I lucked out massively and got a relic from an early goody hut, so bagged Earth Goddess for the first time ever. Amazing pantheon! I can see why the AI likes it so much. Then I had a quandary with Mali, where again it was available (because desert) but not worth taking because appeal was crappy (because desert) - luckily Desert Folklore was still available.

Usually Divine Spark is my go-to for most Civs/situations.

It is ironic that Mali almost always will score the first pantheon choice, guaranteeing a chance to nab Earth Goddess, yet the very desert tiles that got them the early faith to begin with are usually not suitable for EG. For Mali, Desert Folklore will generally be the better choice if faith generation is your game.

 

[Victoria]

Denial is a valid and often vital tactic @Phoenix1595
One has to wonder why they do not go for folklore if they now have a heavier desert bias.