Improving AI performance by giving them a custom race they can use well

from Elemental Forums

I posted a few times about a game I had where the AI went absolutely nuts in the power ranks.  It recently came up again in another thread and I was thinking about the differences between that game and all the other games I have played where they generally perform terribly.

The notable things about that game that I can think of were:

1) The AI had half the map uncontested

2) They were playing a race I designed

3) When I played that race, in the game I designed lots and I mean lots of different henchmen units.  I didn't retire the old versions, just kept making new ones.

4) In the game where they went nuts, they had tons of henchmen.

So that got me thinking.  I don't know how the AI selects what class to make their henchmen into, I assume it is random.  That would mean about 20% of those would end up as commanders.  It would also mean that a lot of the units they make and put into cities to defend them would end up as henchmen.  

Purely by choosing commander they would get a decent unrest reduction, and with a few levels they could potentially generate a whole lot more unrest reduction.  Potentially for their entire empire.

Being henchmen, they don't die, they just rack up drawbacks.  If they did get teleported to a city from that, they would instantly drop it's unrest from basically anything to 0.

I am wondering if:

1) Anyone has had a similar experience

and

2) If anyone would be interested in trying to win a game after making an AI friendly faction and during the course of that game design a whole lot of henchmen units and then let the AI play it afterwards and see if the same sort of effect would happen.

Alternatively, has anyone made/played a mod where the AI unrest is permanently 0?

 

18,487 views 12 replies
Reply #2 from Elemental Forums Top

I find that the best way for me to win is for me to not engage the opponents until really late in the game.

AIs would have no chance if they didn't declare war on me until turn 200.

I don't know if I have ever lost a game where I survived past turn 100.

I think it can only penalize the AIs to drag the game out.

Generally speaking if their power graphs are linear and cap early vs my graphs that are quadratic and have no cap, every turn I delay helps to either close the gap or to increase how far they are behind.

I usually just bide my time until I am quite powerful and then just start crushing AIs from the weakest on up.

The Political Machine 2024
Reply #5 from Elemental Forums Top

What I meant is that it is advantageous anyway for me to wait until turn 200 to attack.

For that matter my victory is a tiny bit surer if I do it on turn 201 instead of turn 200 and a great deal surer if I do it on turn 220 instead of turn 200.

Waiting is a benefit to me rather than a benefit to the AI.

Instead, the worst time for me to go to war with them would logically (and in reality) be turn 1.  If I survived the rush I would likely be so far behind as to get dogpiled by the other AIs anyway.

In other news, while I like the idea of new unit designs it kinda feels like rearranging deck chairs on the titanic.  If they make unit A it is no different if they make similarly powerful unit B.

OTOH, what I was talking about has the potential to completely change the entire game for the AI.  Potentially enabling them to make as many cities as they want and have 0 unrest in all of them even without using unrest reducing buildings.  It would only work for Altair, but if people did the strategy and experienced similar results, it would definitely be an interesting verification.

Reply #7 from Elemental Forums Top

Quoting Raiddinn, reply 2
I find that the best way for me to win is for me to not engage the opponents until really late in the game.

AIs would have no chance if they didn't declare war on me until turn 200.

I don't know if I have ever lost a game where I survived past turn 100.

I think it can only penalize the AIs to drag the game out.

Generally speaking if their power graphs are linear and cap early vs my graphs that are quadratic and have no cap, every turn I delay helps to either close the gap or to increase how far they are behind.

I usually just bide my time until I am quite powerful and then just start crushing AIs from the weakest on up.
End of Raiddinn's quote

This is the part where if ai's were programmed to go after the player when he is weak there would be a lot more challenge to the game. But, as it is the ai is nothing more than a delaying action for the player. It's allowed to let you play till you win. Which is boring and a waste of time. If about 4 factions came at you in the early game (like in Spartan on hardest or impossible levels) it would be so much more fun defending onesself. As it is you get these 100-200 turns to do what you please mostly and only a moron or idiot would lose with that many turns to catch up or build up.

Reply #8 from Elemental Forums Top

Any of you guys know how the AI chooses which units to build?

I know there's a preference you can set in the custom faction leader's xml file (AIMilitaryStrategy or something to that effect) but that just seems to make them focus on one weapon (clubber=blunt, piercer=spears, swordsman=swords, and archer=bows), with an extra focus on mounted units.

But how does it choose between units within those very broad categories?  Does it always try and build the most powerful units if there's enough resources available?  Does it always build from one design or try and get a mix of units with different traits?  Does it always ignore the generic unit designs if there are any custom ones?

Reply #9 from Elemental Forums Top


I posted a few times about a game I had where the AI went absolutely nuts in the power ranks.  It recently came up again in another thread and I was thinking about the differences between that game and all the other games I have played where they generally perform terribly.

The notable things about that game that I can think of were:

1) The AI had half the map uncontested

2) They were playing a race I designed

3) When I played that race, in the game I designed lots and I mean lots of different henchmen units.  I didn't retire the old versions, just kept making new ones.

4) In the game where they went nuts, they had tons of henchmen.

So that got me thinking.  I don't know how the AI selects what class to make their henchmen into, I assume it is random.  That would mean about 20% of those would end up as commanders.  It would also mean that a lot of the units they make and put into cities to defend them would end up as henchmen.  

Purely by choosing commander they would get a decent unrest reduction, and with a few levels they could potentially generate a whole lot more unrest reduction.  Potentially for their entire empire.

Being henchmen, they don't die, they just rack up drawbacks.  If they did get teleported to a city from that, they would instantly drop it's unrest from basically anything to 0.

I am wondering if:

1) Anyone has had a similar experience

and

2) If anyone would be interested in trying to win a game after making an AI friendly faction and during the course of that game design a whole lot of henchmen units and then let the AI play it afterwards and see if the same sort of effect would happen.

Alternatively, has anyone made/played a mod where the AI unrest is permanently 0?

 

End of quote

All these complains are solved in Patchwork mod. AI settles and builds quickly, and fights deadly. https://forums.elementalgame.com/450250/page/1/

Reply #10 from Elemental Forums Top

Quoting BuzuBuzu, reply 8

Any of you guys know how the AI chooses which units to build?

But how does it choose between units within those very broad categories?  Does it always try and build the most powerful units if there's enough resources available?  Does it always build from one design or try and get a mix of units with different traits?  Does it always ignore the generic unit designs if there are any custom ones?
End of BuzuBuzu's quote

As far as I can tell, it heavily favors custom units.  Basically to the exclusion of everything else.

Reply #11 from Elemental Forums Top

Quoting Raiddinn, reply 10
As far as I can tell, it heavily favors custom units.  Basically to the exclusion of everything else.
End of Raiddinn's quote

Which is a Good Thing (tm) because the stock units (I mean the ones that unlock with the techs, not the pre-designed AI units) are generally not very well optimized.

Reply #12 from Elemental Forums Top

I believe PrimalSavage experimented in his AI mod with trying to get the AI to build stronger stacks. In my first game playing with the PrimalSavage AI, Yithril turned up with one Epic stack, which I'd not seen before, and it was quite early in the game. I've not seen so much of an effect in other games since.

It is also something of a trade-off, as if the AI puts all its eggs in one basket (or stack) then you can target that stack with strategic spells, which might be harder if the AI built two Deadly stacks.

In general the AI does seem to favour player custom unit designs. It's not a game changer, it generally makes battles harder to win rather than impossible, but it is definitely a bit more challenging.

Yithril seems to be steamrollering other AIs even more than normal since I gave it some optimised Juggernauts. I've not had to fight against them yet, that may change in my current game.