The key to AI greatness

[Theseus]

First off, I totally applaud the effort. As I've stated before, I think the challenge for any worldgame developer in meeting the needs / desires of the spectrum from builder to warmonger inherently limits the AI civs to a lower level of performance than I'd like to play against... thus, the search for ways to generate killer AI civs.

Thus far, I've tried approaches to this OTHER than going into the editor. In general, just using player settings, I've found that a) differentiated quality of starting territory and b) distance between civs seem to be the greatest factors. The strong get stronger, indeed. Where I've been frustrated is that often the successful AI civs don't push there advantage as much as one of us would.

My guess is that these build preferences are where the answers lie... this is actually much more important to me than most of the modding discussions I've seen thus far, which have primarily focused on units, buildings, and techs.

Some comments:

1) FUNDAMENTAL CHANGES

Alexman sort of went wild in re-setting build preferences. While I think thats fine for scenarios, I am mostly interested in improving the basic game, so I would tend towards minimalist tweaks. I also think that gameplay balance is incredibly difficult to manage... again, hats off to Soren and company, and it's my guess that just a few changes to build preferences will edge certain AI civs over into a killer state (again, predicated on their initial environment in a given game).

2) KILLER HUMAN CIVS

Basically, I want the killer AI civs that arise to play with a combination of the tactics / strategies that people discuss here and at CFC.

(I can just see it: "Oh no, it's a Horse-Samurai-Cav-Tank Army!!! %$@!@$@#@% Must be the Theseus code in the AI!!")

Seriously, I want the AI to do the same things we do... oscillate, don;t go offensive without barracks, upgrade units, attack along hills / mountains, follow Vel's fanatical insistence on Marketplaces, get Factories up ASAP, etc.

I'm not asking for a re-write of the code, but I think, with the patches to date, we can tweak build preferences to get there.

If I had to pick the things WE do that the AI civs don't, but that I think build preferences can help: Barracks (except I don;t know where these fall in the preferences), offensive land units, and treasury management.

3) WHAT ALREADY WORKS?

Before trying to make specific suggestions, I have a question for all: What AI civs already seem to be the closest to killers?

For me, it's been Egypt, France, Germany, and Greece...

What are their existing characteristics, and are there any commonalities? For instance, all four have a preference for Culture!! WTF? What role does that play in their successes?

Can we get some discussion on this point, before trying to reach a consensus on what changes might be most beneficial?

[FrustratedPoet]

Originally posted by Theseus

For me, it's been Egypt, France, Germany, and Greece...

For me it is Persia almost every time.
The last 3 big games I've played on Regent have seen me dominate one continent and have to fight a strong Persian empire on another continent. Strange that it has worked out like that.

[Arrian]

Hmm,

In my experience:

Stong: France, Greece, Egypt and China (note 3 industrious civs in there)

Medium: Aztecs, India, Babylon, Persia, Japan, America, Russia & Iroquois (hmm, lots of religious civs)

Weak: Rome, Germany, Zulu, England

What do I notice? Well, without even looking at build preferences, 3 out of 4 "weak" civs are aggressive militarists with a mediocre 2nd trait. England just sucks. Three out of four "strong" AI civs are industrious, and all have medium-low aggressiveness ratings. In other words, they will attack only when they are stronger. Greece and China have UU's the AI can't help but benifit from, and France's is well-timed. Egypt's is really only powerful in the hands of the human.

Now, I have seen some of these civs in different categories in particular games. I've seen a stong India and a weak Greece. But I've never seen a strong Rome, Zululand or England. France, given time to develop, is almost always solid.

-Arrian

[Jaybe]

What are their existing characteristics, and are there any commonalities? For instance, all four have a preference for Culture!! WTF? What role does that play in their successes?

What do cultural buildings do? They produce either WORKING (as opposed to specialist) citizens, especially in metropolises, or they produce science.

[Arrian]

Greece and Egypt have build orders which match their traits:

Greece = Sci/Comm -> Sci/Trade/Culture
Egypt = Rel/Ind -> Prod/Culture/Growth

France and China don't match up perfectly:

Fr = Ind/Comm -> Happy/Trade/Culture
Ch = Ind/Mil -> Growth/Prod/Sci

Perhaps their industriousness and medieval UU is what saves them.

India is one of the few (only? I need to look again) civs with 4 "build often" orders: Growth, Wealth, Trade, Culture. This suggests, to me, that if you give them some time w/o being attacked, they could do well. In a recent game I didn't play out, India was heinously big in ancient times. Of course, I was attacking them...

-Arrian

[Theseus]

Now we're talkin'....

About other civs: I've seen Rome powerful once. I've seen India get veeery large, but without power. I agree about China, although they never seem to be among the final superpowers.

Arrian, I didn't quite follow that last post. What do you think Industrious matches up to?

[alexman]

After everyone's comments, I agree that a better approach to creating a killer AI would be one with smaller changes, at least initially.

Here's a revised table of build priorities that might work:

Civilization  Off Def Arty Setl Work Nav Air Grth Prod Hap Sci Wlth Trad Expl Cult
----------------------------------------------------------------------------------
Rome (4)       X   -                          -    X                 +         +
Egypt (3)                                     X    X    +       +              -
Greece (3)                            -            +        X        X         X
Babylon (4)                                        +    +   X                  X
Germany (5)    X                                   +        X        +         -
Russia (4)                                -   X    +        X        +
China (2)          +                          -    X        -        +         +
America (3)                               X   X    X    +                      -
Japan (4)      X                      -            +    X            +
France (1)                                         +    X            X         X
India (1)                                     -    +    +       X    X         -
Persia (4)     X                                   +        +   X    -
Aztecs (4)     X                                   +    X            +
Zulu (5)       X                                   +                 +    +
Iroquois (2)                                  X    +    X                 X
England (3)                           X            +            -    X         X

X = original
- = removed
+ = added

In modifying the table, I followed these rules:[list=1][*]Maximum of four build-often items per civ.[*]Production a build-often for everyone[*]Each civ needs at least one marketplace category (trade, wealth, or happiness)[*]Take advantage of civ traits: Scientific->Science, Religious->Happiness, Militaristic->Offense (or Defense) and trade (cheap harbors etc), Expansionist->Growth or Explore.[*]Don't change anything unless it violates one of the above rules.[/list=1]

Theseus, what do you think? Could we set up the next AU course with this, or a similarly modified AI build list? It would be interesting to compare notes on how the AI is affected by these changes.

Edit:
India: wealth instead of growth
Rome: culture instead of growth
China: culture instead of growth
Persia: wealth instead of trade

Edit:
Egypt: wealth instead of culture

[FrustratedPoet]

Originally posted by alexman

Theseus, what do you think? Could we set up the next AU course with this, or a similarly modified AI build list? It would be interesting to compare notes on how the AI is affected by these changes.

It would be interesting to repeat AU 101 using these modified values. That way we would having a direct comparison.

[alexman]

Hmm... good idea, but I don't think it's possible. How do you edit the rules in a save? I don't think you can import its map into another bic file either.

[Theseus]

Awesome!

Some questions:

* Why add Def to China?
* I thought Soren indicated that Wealth was good for some builds too... are you assuming that having both Wealth and Trade is overkill?
* As I pointed out in an earlier post, Culture might be one of the key prefs... how about alternating Prod and Cult?

I definitely envision using this for AU... I'd love to get a minimalist approach to unit / building mods, maybe from Vel and those guys, and merge it with this.

Good work!!

[Theseus]

Re FPs point, we could generate a new map with all the same settings as AU 101... although starting environment would differ for each of the civs, which is probably the biggest determinant of who goes killer.

[alexman]

Originally posted by Theseus
* Why add Def to China?

China is militaristic. I thought flagging defensive units would at least take some advantage of their trait (and add variety). Perhaps they will build more barracks. I didn't add the offensive units flag because China is the only militaristic civ with low aggression. But maybe it's better to flag offensive units, as the AI makes enough defensive units as it is. I don't know.

* I thought Soren indicated that Wealth was good for some builds too... are you assuming that having both Wealth and Trade is overkill?

Initially I was proposing to put wealth in the build-never list, which is clearly a very bad idea. I started out with the intention of keeping Wealth flagged for some civs, but that would make them have more than four items in their list. Whenever there was a choice between a trade and wealth, I chose trade, because the latter clearly has more valuable builds, especially for militaristic civs.

* As I pointed out in an earlier post, Culture might be one of the key prefs... how about alternating Prod and Cult?

Culture and Production have very different builds. You clearly need both. On the other hand, happiness and science produce culture, so there is some overlap there. How about ensuring that each civ gets at least one culture-producing build-often category (culture, science, happiness)?

I'd love to get a minimalist approach to unit / building mods, maybe from Vel and those guys, and merge it with this.

I suggest player 1's conservative patch suggestion mod. He has made some brilliant tweaks that nonetheless keep the original Civ3 flavor.

Re FPs point

If we start from a bic file instead of a save, we can use the same map to create two initial saves with different AI strategies. We could post both saves as part of AU102, and people could choose which one to play, if not both.

[Theseus]

China: Get it.

Wealth: At least one civ should have this... change one of the Comm?

Culture: Good solution. BTW, I think Zulus have none.

I'll check out player1's mod.

Re AU 102, let me think about which one might best highlight the differences in the emergence of killer AI civs.

[Chieftess]

Originally posted by Arrian
Hmm,

In my experience:

Stong: France, Greece, Egypt and China (note 3 industrious civs in there)

Medium: Aztecs, India, Babylon, Persia, Japan, America, Russia & Iroquois (hmm, lots of religious civs)

Weak: Rome, Germany, Zulu, England

I think it depends on the eras and starting locations, but here's what I've noticed:

Strong: Russia, Japan, Persia

Moderate: England, Aztecs (if they get a good stronghold), China, Germany (depends on their starting posistion too..), India, Iroquois

Weak: Greece, France, Egypt, Babylon (they can expand fast, but in my games, take out their capital and surronding cities, and they're toast), America, Zulu, Rome

While Greece and France have some of the best defensive units, I have yet to see the French USE their musketeers! I've used them with great success, since their stats are like 2 swordman in one.

[Arrian]

Chieftess,

Well, that's about the opposite of my observations. Clearly, all of this is heavily dependent on geography. A nice starting spot will mask terrible traits/build preferences.

Alexman,

Perhaps the scientific civs ought to have "happiness" instead of "culture." They are all flagged to build scientific things anyway. Happiness will encourage temples, cathedrals, colosseums and marketplaces.

I'd say add wealth and drop culture from Egypt. The addition of happiness keep culture production solid and get some more markets in there... and wealth will top that off. Actually, when you think about it, Growth+Production+Happiness+Wealth = Civ in a nutshell.

-Arrian