AI Citybuilding suboptimal - improvement suggestions


The AI has odd citybuilding practices. The following is merely exemplary: Often, the AI seems to build an inordinate amount of conclaves -- which would be justifiable, in my eyes, if it were to use these for purposes, primarily, of generating research or mana. But examining these cities over time often reveals that they quite frequently use these for training units -- while at the same time, the capital (for example, Tarth's Hope) is a dedicated fortress, which it uses for quite a few turns for ... generating wealth. Additionally, the Conclaves seem light on studies and labs, and are often heavy on gidlar-generating structures. Currently looking at a few conclaves Kraxis has scattered close to me, the nearest, Kerstey, is level 2, has a Study and a Black Market, has 2 units trained in its queue, and the third in the queue is Produce Wealth.

I think this needs some attention. The AI could do some or all of the following:

- make more fortresses, build more unit-improving structures in fortresses, use fortresses primarily for generating units,

- make fewer conclaves, build more research- or mana-generating structures in conclaves, use conclaves primarily for generating research or mana, train fewer units in conclaves

- build more food- or wealth-generating structures in towns, use towns primarily for generating food or wealth, train fewer units in towns

10,910 views 12 replies
Reply #1 Top

I noted this behavior as well.

My suggestions are slightly different from yours:



The AI could do some or all of the following:

- make more fortresses, build more unit-improving structures in fortresses, use fortresses primarily only for generating units,

- make fewer conclaves, don't build conclaves with only 0 or 1 essence, build more research- or mana-generating structures in conclaves, use conclaves primarily only for generating research or mana, never train fewer units in conclaves

- build more food- or wealth-generating structures in towns, use towns primarily for generating food or wealth, train fewer no units in towns

End of quote

Reply #2 Top

I believe the AI trains units more often in fortresses in 1.3. I'm not sure I agree with never training units in non-fortresses; I can think of four exceptions:

1: I quite often train Pioneers in non-fortresses

2: In extremis I will train units in non-fortresses, either to increase the number of units I can train at once, or to quickly provide some defensive units. A high essence town or city with Heart of Fire etc. and maybe some other bonuses (e.g. Sacrificial Altar) can provide decent units. Also, the AI on higher difficulty levels doesn't have the same gold maintenance issues that I have;  I tend to favour quality over quantity with troops, but the AI doesn't need to worry about this so much.

3: Sometimes a non fortress city is just better placed to train a unit and get it to the right place quickly.

4: If for no other reason, the AI needs to train units in non-fortresses if it doesn't have any fortresses, either because it's near the start of the game or because it's lost its fortresses in a war.

I agree that wherever possible the AI should train its military units in fortresses, though.

Reply #3 Top


I agree, hence "more" and "less", "primarily" etc. in my original post. I also think the AI could build more fortresses and less conclaves in general. I also believe the AI should be much less likely to turn villiages without essence into conclaves, and less likely to turn villiages into conclaves with essence =< 1.

Reply #4 Top

Ok, ok, maybe I wasn't explicit enough: don't build conclaves with only 0 or 1 essence -> way less Conclaves, many more Fortresses and some Towns. More Fortresses -> more units -> no need to build them anywhere else.

Quoting merlinme, reply 2

I believe the AI trains units more often in fortresses in 1.3. I'm not sure I agree with never training units in non-fortresses; I can think of four exceptions:

1: I quite often train Pioneers in non-fortresses

2: In extremis I will train units in non-fortresses, either to increase the number of units I can train at once, or to quickly provide some defensive units. A high essence town or city with Heart of Fire etc. and maybe some other bonuses (e.g. Sacrificial Altar) can provide decent units. Also, the AI on higher difficulty levels doesn't have the same gold maintenance issues that I have;  I tend to favour quality over quantity with troops, but the AI doesn't need to worry about this so much.

3: Sometimes a non fortress city is just better placed to train a unit and get it to the right place quickly.

4: If for no other reason, the AI needs to train units in non-fortresses if it doesn't have any fortresses, either because it's near the start of the game or because it's lost its fortresses in a war.

I agree that wherever possible the AI should train its military units in fortresses, though.

End of merlinme's quote

1: Good point. Pioneers can be trained anywhere.

2 - 4: Good points all. But ultimately, the AI will never win by 'outgunning' a smart player. Those troops only provide a fun challenge for us (well, for me at least). The only way for the AI to win would be by researching 'I win'-techs like Dance with Dragons and the repeatable techs and/or using lots of spells. The AI building lots of units when it has almost lost, as points 3 and 4 suggest, doesn't really serve any purpose. I'd prefer if the AI actually resigns if it is losing badly. But that's another story.

Reply #5 Top

Quoting onomastikon, reply 3

 I also think the AI could build more fortresses and less conclaves in general.
End of onomastikon's quote

I assume (although I could easily be wrong) that the AI's tendency to make villages into particular types is influenced by its personality. So I would expect militaristic AIs to build more fortresses, research favouring AIs to build more conclaves. The developers have said in the past that this is why (for example) certain factions build more of a certain type of building than might be expected. Personally I think the AI builds sub-optimal amounts of production boosting and unrest reducing buildings, but I believe this is a design choice to favour AI personality rather than optimal strategies. On high difficulty levels the AI gets such huge bonuses it builds nearly everything anyway and has massive resource surpluses more or less regardless of its strategy, so this is a relatively minor issue for me in my games. Perhaps the rules for which settlement type to choose could be tweaked though, because I agree building conclaves on 1 or 0 essence sites doesn't seem a good idea, regardless of sovereign personality type.

Reply #6 Top

Quoting Fallenchar, reply 4
The AI building lots of units when it has almost lost, as points 3 and 4 suggest, doesn't really serve any purpose.
End of Fallenchar's quote

I understand your reasoning, but this is very much a matter of taste. Personally I just think it's more "realistic" if the AI tries to defend itself properly. Apart from anything else it can't know if it's "almost lost"; it can't really know if you have to send your army somewhere else to tackle another target, or if you've taken heavy casualties and it has a chance to rebuild. Also, if it builds up some Epic garrisons then I get to have some Epic battles. :grin:   It makes a lot more sense to me for the AI to try to build units to defend itself than to build the Tower of the Witch in 20 turns time.

If you wish AIs would quit sooner you can modify the chance they'll accept Surrender demands (although I've never tried this, I don't know well it works). Also a better defensive AI doesn't really affect Quest or Spell of Making victories; you don't have to crush the AI into the dust unless you want to.

Reply #7 Top

Quoting merlinme, reply 6


If you wish AIs would quit sooner you can modify the chance they'll accept Surrender demands (although I've never tried this, I don't know well it works). Also a better defensive AI doesn't really affect Quest or Spell of Making victories; you don't have to crush the AI into the dust unless you want to.
End of merlinme's quote

Actually, on higher levels, you do have to crush the AI into the dust. Sovereigns never surrender, not even if you set surrender to 'often'.

 

Reply #8 Top


On 1.3 challenging, I have gotten the enemy to surrender consistently (3 times) when he's down to one city under "normal" surrender option.  Haven't tried higher than that difficulties.

Reply #9 Top


I've recently been noticing this in 1.3 (which maybe was also in 1.2 but I had missed it perhaps) which seems to make AI citybuilding even worse: I've seen AI Towns have "Produce Wealth" and completely incompatible Enchantments going; one AI Town actually had both Aura of Might and Aura of Grace.

I really think the AI citybuilding needs another, closer, look by someone dedicated to doing AI.

Reply #10 Top

Well, one thing that I would suggest. The AI on Challenging is often grinding it's production to a halt due to unrest. It starts off fairly strong and then gets stuck. It seems to happen most with AI's that are militaristic that keep cities they conquer. They either need to raze them (easiest) or be better at using appropriate buildings/enchantments. Of course, it drives me a little batty that most of the AI's ignore viable essence squares and plop down just about anywhere. A 3/3/1 is over all better than a 4/3 or 3/4.

Reply #11 Top

Quoting fenwe, reply 10
Of course, it drives me a little batty that most of the AI's ignore viable essence squares and plop down just about anywhere. A 3/3/1 is over all better than a 4/3 or 3/4.
End of fenwe's quote

Here come to help the Insane level of difficulty )))) The problem lies even in not choosing better tiles but even in low priority of expansion. AI is not quick enought to build more and more pioneers. Even more, these pioneers may stand or walk purposelessly before they finally manage to settle.

Reply #12 Top

Quoting Fallenchar, reply 7
on higher levels, you do have to crush the AI into the dust
End of Fallenchar's quote

The AI will generally surrender to me if I demand it and it only has one city left. I play on Ridiculous with other settings default. I think I've had it surrender to me with two or three (small) cities left, but I wouldn't absolutely swear it. I assume it depends on relative power level; by that stage I've generally taken the AI's capital and best cities and killed all its decent armies.