Ideas for a killer AI

[Random Passerby]

It appears that I have misinterpreted the data. I looked at the example of Spain building marketplaces with "Never build wealth" on and assumed that Spain was looking at the list of possible roles for marketplaces, ignoring its use for wealth, and opting to build them for their happiness benefit. I'm now beginning to think that improvements may have one and only one role that the AI considers them for, and marketplaces are not considered wealth at all; similarly, never build culture does seem to have no effect whatsoever. It appears my results came from my testing setup, wherein the AI had some very impressive production.

It turns out that the AI weighs the production capacity of a city when deciding how soon to build a factory. Here's the test I devised to confirm this:

-The AI is given five ungrowable cities, with production of 5, 10, 20, 30, and 40 (the high production comes from resources). Diety level, accelerated production--so the AI can build things VERY quickly. Never build any units; often build production.
-The 5 shield city went through all the research and happiness culture improvements right off the bat, as usual.
-The 10 and 20 shield cities went through library and university, then went to factory, power plant, manufacturing plant.
-The 30 shield city built a library first, then factory, power plant, manufacturing plant.
-The 40 shield city built a factory right away, not bothering with any other improvements until it had its power plant and manufacturing plant done.

Eager to see if these results could be applied towards getting the AI to build more commercial type improvements, I went in and changed the AI from build often production to build often wealth/trade. In my haste, though, I had grabbed an earlier test and neglected to turn off unit builds. The result? The AI (with about a dozen cities) built one airport, one harbor, and then started pumping out guerrillas to attack me with, with a few odd libraries built here and there. Not a marketplace in sight.

This might explain the impoverished AI problem: if the AI doesn't have many luxuries, it wouldn't want to bother with a marketplace. If it has no marketplace, it can't build banks. I've never been in love with the marketplace's luxury effect--it makes it obscenely easy to keep your cities happy--and would be happy to get rid of it to try and jostle the AI into viewing marketplaces as a money-making proposition, but I'm a bit wary of crippling the AIs that DO have lots of luxuries by adding to their happiness problems. Well, actually, petrified of it, given that Soren has said that the AI only uses the empire-wide happiness boost of the luxury slider in order to fight the empire-wide war weariness unhappiness.

[alexman]

Originally posted by DaveMcW
Decreasing corruption is just as drastic as decreasing the AI production costs. If you want to make a "beyond diety" mod so be it, but I thought the goal of the AU mod was to make the AI play better without cheating.

Chieftain gets 100% of the optimal number of cities. Regent gets 90%. That's a 10% increase. Just for comparison, the Commercial trait is a 25% increase in the optimal number of cities, and you can barely see the difference. So it's certainly not as drastic as increasing production costs.

On the other hand, it IS cheating. But I think it's justified cheating to compensate for the bad AI FP placement. In the end, the AI empire with the bad FP placement will hopefully have as much corruption as a human empire with a good FP placement. One good thing about this cheating compared to production bonuses is that a OCN increase is felt more in the late game when the AI really needs help.

The goal of the AU mod is not to make the AI play better. That's Soren's job, and we have little control over it. The job of the AU mod is to make it appear as if the AI plays better, as Zachriel put it in another thread. Increasing the OCN would make it appear as if the AI had a better FP placement.

[Theseus]

Commenting on RP's post:

F-cking IRRIGATION! That's what !%##%$!$#!@%#$ needs to be changed.

[alexman]

Originally posted by Theseus
1. I would avoid changing Settler cost... this is more for me than for the AI, it just confuses me too much when going back and forth between mods and stock.

Fair enough. I thought this was a big change anyway. But what about attacking with the AU mod Infantry-on-steroids? Isn't this confusing too?

2. Double the entertainers' effect. WE just have to agree not to exploit this.

I'm not sure we have to, because it may not be exploitable. I believe that happiness improvements will still be the way to go, but we definitely need to playtest this.

3. and 4. I think trade and production should be emphasized... we do it, don't we?

Agreed. We didn't used to emphasize trade for all civs.

5. and 6. Leave alone.

Why? We humans don't look at maintenance costs (we build whatever needs to be built), so increasing them will not affect our strategy - only our pockets. The AI builds many units before improvements. So why not help them save some costs in doing so?

7. 8-attack for Infantry, yes?

Yes!

8. Repeat what we did before.
9. Cool!
10. Leave alone.

Why leave alone 10? We made this change in the AU mod. Do you build wealth more often as a result?

11. This is the most difficult one... I'd love to hear from Soren.

I'd love to at least know he's reading some of these threads.

12. What are the current fixes in the AU Mod?

Not settled yet. AU 1.05 had free maintenance. AU 1.06 has increased free unit support. Both had fasted workers.

[Theseus]

Well, all of the changes are sorta confusing... I just can't deal with any changes to Settlers!

5. and 6. I'd like to see more of PTW first... I think the AI civs should build MORE units, if anything (but that's my warmongerin' way!).

10. Sorry, I forgot.

12. Man, I just don;t know what to do with Communism, but we gotta do something. Unless people really disagree, I suggest we *try* making Communism the uber-government for how and why the AI civs (SI, whatever) seem to function. Now that the AI civs are getting 'polished' we seem to be focusing more on later-game performance... well, that probably means at least some warfare, and thus Commies. This'll be 1.07/PTW? Fine let's consciously experiment, and give Communism everything... in fact, is there a way to increase the likelihood of Factories?

[alexman]

Originally posted by Random Passerby
I'm now beginning to think that improvements may have one and only one role that the AI considers them for, and marketplaces are not considered wealth at all

Soren has already told us that buildings belong to several groups. He also said to be very careful when setting something as build-never. But from our tests it appears that build-never means that you never build something just for that category, but you build it if it falls in another category that you are allowed to build. Marketplaces belong in the Happiness, Wealth, and Trade categories, BTW.

It turns out that the AI weighs the production capacity of a city when deciding how soon to build a factory.

Your tests are very enlightening. Keep up the good work.

This might explain the impoverished AI problem: if the AI doesn't have many luxuries, it wouldn't want to bother with a marketplace.

Buildings belong to different categories, but have a certain amount of strength in each category. I think marketplaces are weak in trade and happiness, but strong in Wealth. In any case, did you repeat your experiment with units turned off? It would tell us much.

Again, thanks for sharing your experiments!

[Theseus]

alexman, we cross-posted... look for my response on the first page.

[Random Passerby]

Oops. I am an idiot. First, fiddling with the multiple-category aspects of improvements in the AI's preferences, I discover that the AI does weigh production capacity when deciding its build order; then when I start looking at marketplaces I go back and ignore both of these. Doy.

Anyways, I went through two more runs on variations of the production scale scenario: the first time, I just unchecked production for build often and put in wealth and trade instead. The results were negligible: the highest production city still started off on a factory, and second-highest went through library-university instead of library alone. All cities ignored marketplaces until they had finished just about everything else.

Then I smacked myself on the forehead and upped the production resources I gave them to have 25 commerce too. The AI built library first everywhere, and marketplace was 1-2 builds later. This is obviously much better.

One thing that is either promising or disturbing: in both scenarios, the AI built airports very, very quickly (I left those out of the above build orders on purpose, since they appeared to have a roughly universal weighting). All the cities I was looking at had strategic resources in their radii (but unconnected), so this wasn't such a bad idea in terms of regular game conditions... except that all the cities were on the same landmass. I'd do some more fiddling with the effects of resources and trade routes on build preferences, but I don't have any more time to spend fiddling.

[Killerdaffy]

6. The AI is predictable in its research path.

IF we decide to change tech costs, here a couple of ideas:
The general cost for tech shouldn't be changed without lots of playtesting, so nearly all changes are balanced within the same tree part immediately before or after. Generally some of the changes might be too small to make a difference but let's try this first.

Ancient
a) Mathematics down one to 7
b) Map Making up one to 13
c) Philosophy down one to 5 (it's cheap already but the AI ignores it anyway
d) Code of Laws up one to 11
e) possibly Literature cheaper

Medieval
f) Engineering down two to 34 (probably too little change)
g) Feudalism up two to 34 (why was it cheaper in the first place??)
h) Printing Press down two to 34
i) Education up two to 46
j) Chemistry down four to 56
k) Metallurgy up four to 68 (keeps the cost for the tree intact but make chemistry more attractive)

Industrial
l) Medicine down ten to 90
m) Nationalism up ten to 130 (probably isn't enough)
n) Electricity down 20 to 120
o) Scientific Method up 20 to 120 (this should make our ToE prebuild more difficult)
p) Atomic Theory down 20 to 180
q) Electronics up 20 to 200
r) Sanitation up ten to 90
s) The Corporation down ten to 90
t) Communism, Espionage would also be candidates for a raise

Modern
Hm, I'm missing experience here, I hardly ever play this far. It seems to be better balanced, though (and we shouldn't forget that our changes in the Spaceship Parts already have quite an impact).

[ducki]

Maybe it's because I'm inexperienced, and therefore, making "bad" decisions on tech research, but I always end up buying Philosophy from the AI.

Actually, I can't recall which factor is significant, but either since I started playing Regent/Monarch or possibly with PtW, the AI always beats me to Philosophy, Map Making, Code of Laws, and Republic.

Unless I realize within the first 40 turns that I'm alone on a landmass and beeline for Map Making, naturally.

Actually, I'm usually last to Mathematics as well.


Am I doing something terribly wrong, or is there enough randomness in research that some people will see the AI ignore certain techs and others won't?

[Theseus]

I think it's PTW... I still don't have a good handle on how research works.

[alexman]

From my limited experience with PTW in a real game, it seems to me that the AI diversifies more in its research and then trades with each other. That's why it seems like they have a higher research rate.

We need more experience with the game before making tech cost changes. However, based on my experience with 1.29f, this is the list I came up with, before even looking at Killerdaffy's list (note how the increases and reductions balance out in each age)

Bronze Working 3->5
Alphabet 5->3
Warrior Code 3->4
Iron Working 6->8
Mathematics 8->6
Philosophy 6->7
Literature 10->7
Horseback Riding 5->6
Polytheism 12->9
Monarchy 24->25
Construction 20->21
Currency 16->17

Feudalism 32->36
Printing Press 36->32
Engineering 36->32
Theology 40->44
Education 44->48
Chemistry 60->56
Astronomy 56->60
Metallurgy 64->60

Nationalism 120->150
Medicine 100->80
Electricity 140->100
Atomic Theory 200->160
Motorised Transportation 140->180
Mass Production 140->180
Radio 200->190

They are quite similar (even though there are some differences we need to talk about) which goes to say that we have figured out the AI research preferences even with only our own games as reference. Imagine what we could do with everyone's input on 1.04f...

[alva]

re: Korn 469

ptw blitz mod

original blitz mod

blitz mod part 2

[alva]

this is where I think they started discussing the settler thing in the original mod

http://apolyton.net/forums/showthread.php?postid=693278#post693278

*Alexman, it's quite alot to go through (+500posts)
A pm might be simpler

ps: they are not so much doing work on improving ai, but for balance issues and possible conflicts/exploits, it could be a good idea to exchange some notes

[Killerdaffy]

Originally posted by alva848
this is where I think they started discussing the settler thing in the original mod

http://apolyton.net/forums/showthread.php?postid=693278#post693278

*Alexman, it's quite alot to go through (+500posts)
A pm might be simpler

ps: they are not so much doing work on improving ai, but for balance issues and possible conflicts/exploits, it could be a good idea to exchange some notes

Generally the blitz mod is too far away from stock for us but there are some really nice ideas. Especially adding / differentiating basic units over eras (e.g. special forces) might be a possibility that is easier to implement with PtW's ability to load scenarios (without messing up the original game).

On a related note: After beating the crap out of the English again yesterday I'm starting to contemplate giving them a decent UU that is actually relevant in a regular game. English Longbows (5.1.1 40, replacing regular Longbowmen) or Redcoats (6.6.1 80, amphib, replacing Riflemen) might fit in nicely (they can keep the Man-O-War, makes them special without doing anything, really). It is getting kind of crowded in Medieval in terms of UUs but then that's the time to have a GA.

@alexman
It looks like we pretty much have the same ideas in terms of changing tech cost (with the notable difference of Philosophy?!?). Your changes are a bit more aggressive so we should test them some more. I'll try it once I'm finished with my current game.