AI nation skills
 

I've played that game many games already and what I noticed is that AI handles some nations much better than others. Did you observe that too? If it's common I think it should be addressed in a patch, I don't like getting same nations getting strong every time and some other disappearing even faster.

I have played mainly MA against difficult AI, so if you have any observations write that age and difficulty.

MA:

Nations that are extremely strong, always:
Ermor [huge armies, huge numbers of castles]
Pangea [rapid expansion, huge armies]
Oceania [unless it meets R'lyeh]


Above average nations [that most often are stronger than average with some unlucky moments]:
Mictlan
Pythium
Jotunheim

Very weak nations:
Bandar Log [huge research, dies very fast, huge problems with fighting indies]
Caelum [dies fast]


I play TC most of the time so I don't know about their performance...

Re: AI nation skills
 

It has to do with the amount of rubbish units that the AI has available to buy/obtain, I've found. The AI is also rather solid with MA and LA agartha, oddly enough- if it has enough resources. I think that this might be because these nations have no real chaff recruitables (similarly to pythium...)

Re: AI nation skills
 

In the EA the AI always does well with the glamour nations (Vanheim, Helheim, Tir na n'Og), Ermor and Ulm. It also gets decent mileage out of Yomi and sometimes Pangaea. Depends a lot on what scales it picks, misfortune 3 tends to slow it down a lot.

Re: AI nation skills
 

AI should really be able to adjust pretenders and scales to the map size.

Re: AI nation skills
 

in so far as it picks randomly, that would be nice. Anything is better than random.

Re: AI nation skills
 

In terms of vanilla balanced mod nations the AI is surprisingly strong with Arga Dis. It gave me fits just by cramming huge numbers of Argan Hoplites down my throat early on.

Re: AI nation skills
 

One of the greatest benefits an AI can receive is an order_3 and productivity_3. It's gold income increases per difficulty level and an order_3 can really make it powerful. I've seen and felt the difference.

To better understand which AI nations are the strongest each would have to be given identical scales with identical pretenders of small size on the most balanced map which can be found. The human player can provide himself two separate provinces with no neighbors, high astral mage pretender at start to cast eyes of god. Rotate starting positions and record results after 100 turns.
**Water nations should be tested separately since most AI land nations lack the knowledge for attacking water provinces.**

Chaos_3 and Unluck_3 really make a weak AI, Death_3 weakens the AI long term.

Re: AI nation skills
 

Its probably not worth having EA or LA Mictlan in a SP game. It usually dies of dominion death because the AI does not use the perform blood sacrifice order.

MA Mictlan does not have this problem as the Lawgiver banned blood magic in MA.

When you can defeat impossible level there are various things you can do to increase the challenge even more, but MP is where the real challenge is.

Re: AI nation skills
 

Meglobob?

I don't have time for MP. How can one increase the challenge beyond impossible AI in a sensible way ie. still fun to play? Thank in advance.

Re: AI nation skills
 

There are 2 ways to do this that I know:

1. Create pretenders for the AI and start game with all factions human. One thing to watch in this is bless pretenders, because the AI does not use bless well. Better to use good scales and rainbow pretenders. Immobile AI pretenders might be the best choice, but that is subject to debate.
2. Edit a map file. This takes longer, but has definite advantages. You would place each AI in a province, and make AI pretenders and scales that are much more positive for the AI than the game allows. Make sure you than select "Cheat detection off". You can even give the AI fully outfitted SC pretenders. Also set all AI nations as allies so they will only attack you. If you do this and play on "impossible", you will have to be a top notch superior player to win. (I have found that I am not.)

3. Set indy strength to 3 or less. This will let the AI expand very quickly with the obvious benefits.

Re: AI nation skills
 

Thanks Edratman. Presumable on 1. all other human fractions resign on turn 1 and the AI take over. Is that correct?

Re: AI nation skills
 

That is correct.
http://forum.shrapnelgames.com/images/smilies/laugh.gif
That is the only way I know how to select the pretender and scales for the AI without editing.

Re: AI nation skills
 

To the very weak nations I would add Abysia, because seeing fully fire resistant armies without any fire mages in support is really funny, I usually kill them wihout having to wistand any fireballs(or fire clouds).
The same thing is probably true to all mage reliant factions.

Re: AI nation skills
 

I just had a thought. Would it be possible to have on the menu screen (the one where you set indy strength) an option to make all AI allies in the next patch? I have no idea if the coding would be managable or not. The default should be non-allied if it is doable.

Re: AI nation skills
 

Another option, very easy to implement, is to “stale” your first three or four turns. Just hit end turn and do not give any orders for the first few turns. While it does not change how the AI performs, it gives the human a big disadvantage. As the growth tends to be exponential, those lost turns at the start really hurt in the long run.

I regret that I do not remember who first posted this idea, as they deserve credit for such a simple fix.

Re: AI nation skills
 

Play on NI (no independent maps) helps alot.

Set resources to 300 helps the impossible AI as well.

Re: AI nation skills
 

AI strength is not my biggest problem. I would like to see all AIs equal. I do not want to have always strong Ermor against me. I would like to meet strong Bandar Log or Caelum. Now if you play on a big map with most nations in you may expect that the same nations will be your biggest threat.

Re: AI nation skills
 

This can only happen in MP play Zeldor or via extensive modding of those nations which would of course change them into a completely different nation.

AI for individual nations would be great but I can't see it ever happening.

Re: AI nation skills
 

I play SP when I want many turns a day, I won't get that in MP.

Some nations are weaker in SP not because they are made weak but because AI cannot handle them. I don't think mods can change it, unless you can rewrite a lot of AI mechanics.

Re: AI nation skills
 

If you'd like to see some huge armies - which the AI is better with than mages, for sure - set the Supply and Resources multipliers to 300. Setting Gold to 300 will give you so much you'll be able to build a couple castles every turn by the second year, which changes the game drastically. Between this and the No Indies mods, you usually have quite a bit to deal with pretty quickly.

Re: AI nation skills
 

It's possible to handicap things by adjusting nations' starting troops or sites, or doing things like giving them modded immobile monsters which serve as gem generators or so forth.

Modding an AI's nation-specific commanders to all have a supply bonus would go a long way to preventing mass starvation without freeing the human player from the same constraints. Giving a research bonus to its mages might also help.

You can't block it from going Turmoil/Death/Misfortune on a large map without actually setting up the sides, but the Misfortune could be ameliorated by fortune-teller abilities.

Re: AI nation skills
 

One downside is the AI opponents don't march beyond a castle province until the province has been captured.

Re: AI nation skills
 

I am sure there are simple solutions to improve AI skill and it wouldn't be so much work to do, for example:

- make AI choose scales and pretenders somewhat wisely
- make AI recognize geographically strategic provinces
- stop AI from using extremely stupid ideas like hordes of indies with C'tis miasma

Re: AI nation skills
 

the first item we can already do ourselves with maps and mods

the second thing needs more defining. Some people would say "build a castle in a choke point then dont take the neighboring provinces" but others would say "stupid AI built a castle on a chokepoint that only had 3 neighbors"

last item is out completely. There is only one AI and it builds its armies the same for all nations. The thinking must be something like "build x infantry, build x archers, build x mounted, build 1 commander or mage". And now you want "unless you are ctis with miasma then build". But then we end up with lots of "unless you are" all through the code. That would be a lot of work. And it would really slow the game down for solo playing

Re: AI nation skills
 

Wikd Thots:

Point 1 would be really easy to implement. At least scales.

Pint 3 is indeed very complicated, but it should be like that http://forum.shrapnelgames.com/images/smilies/happy.gif Human plays every nation in different way and so should AI. I should have gone to cognitive science so I could help with that...

Re: AI nation skills
 

Zeldor, since you have an intimite understanding of how strong the different nations are when played by the AI why not just adjust difficult for the nations accordingly. Perhaps increase difficulty for Bandar Log and Caelum. Or decrease difficulty for Ermor and Pangaea.

If you want different nations to get strong in each game without before hand knowledge you may put three AIs to higher difficult and all with random nations.

Re: AI nation skills
 

Folket:

That's what I plan to do next time, but I would prefer not to use tricks like that. And choosing higher difficulty does not make AI make wiser choices with scales and so on.

Re: AI nation skills
 

Well you can start them as a human player, choose appropriate scales and then turn them over to AI. But I'm pretty sure the impossible AI is always stronger, because it gets a huge bonus to pretty much everything.

Re: AI nation skills
 

That's why I use map edit commands to provide the AI with strong scales, strong blessings, Mighty/Impossible difficutly setting and pretenders it won't blindly send into the death match arena.
Much more powerful results than starting a nation as human then switching to AI.

Re: AI nation skills
 

Well, I find that it is not as fun when you know what you are up against.

Re: AI nation skills
 

I find it's more fun knowing I'm up against a thriving difficult opponent compared with spending 40mins to discover nothing but withering weaklings.

Re: AI nation skills
 

That's true of point one also. It would still end up having to be a list of every nation and the scales it would take. Plus I can come up with 2 or 3 standard scales for every nation depending on how its played. If you think that there is a set of scales that will work without doing an "IF" for each nation then let us know what is it.

But if they did do an if for each nation then it wouldn't matter much since the lag of thinking would only be at the creation of the game. Not as bad as trying to do it for in-game thinking like buying reasonable troops.

Re: AI nation skills
 

Not necessarily a list of scales for every nation. I want the randomness in there. I don't want the same pretender design every time, nor a list of 2-3 choices for each nation.

But a couple of things ruled in/out would help a lot.
Some general rules like:
Don't take 3 misforture and 3 turmoil.
and some specific ones:
Neifelhiem: take cold 2 or 3
EA/LA Mictlan: high dominion and/or awake pretender
High production with MA Ulm

Don't think of it as optimizing the AI's design, but avoiding stupid mistakes.

On another note:
If the code is well-designed, there should be little to no lag even for in-game nation specific thinking. You don't have a series of checks at every decision point, you have separate, preferably inherited, functions for races that want special handling.
(Of course, it's probably not well-designed. Most code, including mine, isn't. That's not a criticism, just an observation. Large codebases tend to grow haphazardly and need to be rethought from the ground up, but there is rarely time or interest.)

Re: AI nation skills
 

thejeff:

There are much less choices for scales on very big maps and bad choices are really not firgiving.

Re: AI nation skills
 

Granted, that adds more complexity to optimizing pretender design, but there is still merit in ruling out particularly bad choices.
(Some of which is already done, right? I don't see cold 3 Abyssia very often. A race's temperature preference is taken into account. Why not preferences for other scales?)

Or are you saying there's less complexity? One set of scales that is optimal for all nations in a large enough game? Even if that were true, (which it is not, there being some obvious exceptions) it wouldn't hold in smaller games.

Re: AI nation skills
 

thejeff:

I am just saying that choosing bad scales for a smaller map has a smaller influence than taking them on huge map. You can survive 1:1 game with Death 3 if you use it for example to have super awake pretender or nice bless. But try using it on 500 province map against 20 opponents. And AI should take it into consideration.

Re: AI nation skills
 

Or since it can't handle the SC pretender or a bless strategy it should avoid taking the bad scales even on small maps...

Re: AI nation skills
 

In addition to the NI maps, someone ought to create mods for the various nations that eliminate units the AI shouldn't be creating, within it's own nation.

Re: AI nation skills
 

Not a lot gets done waiting for this 'someone' person to do stuff.

Re: AI nation skills
 

Well, I'd elect myself, but I've got no working computer to use. Won't have, for a year plus. That and I'm not a very good player anyway, so the AI doesn't need that much help squashing me.

Re: AI nation skills
 

Well go ahead and write it all out. Post it so we can see it (I already have some problems with your examples).

And, the code wasn't designed at all. It was written on an Atari many MANY years ago and has grown in pasted-in code ever since. That is part of what Johan has been saying the whole time. Changes like some people want would involve rewriting the whole thing from scratch. Not fun (and then won't happen)

Re: AI nation skills
 

Don't bother. Somber has gotten that argument many times.
A better AI mod needs to be written by one of these people who feels that the AI sucks and they know exactly what it should do to play better. Various threads have tried to pull out the suggestions. I think there was one for a better Mictlan AI that Somber and someone else did finally make mods from.

ALSO there is the SemiRandom project. It lets us all design gods, scales, equipment choices for the AI to start with. And it will randomly pick from those files when it starts a game. If more people did AI design files for it then it would have a bigger pool to choose from but still all be sensible choices. That is probably easier to go with than trying to get the game code rewritten

Re: AI nation skills
 

*points at sig*

Though I need to update that mod to the one I'm using right now. The currently public one is screwed up. I also need to rename it to "Better Independents"

Re: AI nation skills
 

The "if the code is well-designed" part was somewhat tongue-in-cheek. As I said, code is rarely well-designed if designed at all and I knew enough of Dominions history wo assume it wasn't. Had forgotten the Atari beginnings, though.

Still, I suspect that kind of logic wouldn't increase thinking time significantly if implemented well. Especially since that's already a tiny part of the lag between turns. Impossible to know without looking into the code, though.

I'm not sure what you mean by "write it all out." It's not going to be implemented, not is it moddable, so I'm not sure what the point is in coming up with a complete detailed list. Nor am I qualified. I haven't even played most nations.

All I was saying, is that a few common sense rules could prevent the AI from making broken pretender design choices. Some of which, on further reflection, it probably already does. I think the races are biased towards their preferred temperatures, at the very least.

The rules need not be complete to be useful. Rules that prevented a particular bad design would still be an improvement, even if other problems were still possible.

I am curious what problems you have with my examples? I was going for obvious and non-controversial.



Hmmm. catgod reads pretender files, right? Does that mean we know enough to write them? That would let us play with tweaking random pretender design. Couldn't get the AI bonuses though.

Re: AI nation skills
 

I will help get this ball rolling. I will have a minimum of 10 AI pretenders by Nov. 1st. If everyone chips in 5 to 10, there should be a substantial number of pretenders available.

Now I need some guidelines:
1. Should they be generic or nation specific pretenders?
2. Do you stay within the game points limitations or is a lttle cheating acceptable? (IE, if another 10 points or so are required to give an additional level 4 bless, will that be considered acceptable.
3. What format is required for the submissions? I would think a text format that works with map editting would be good, because it allows flexability. But I have 3 different editors that work on map edit and while all 3 are compatable with the game, I do not if they would be mutually compatable with some central compiler.
4. Where do I send them? (I am not computer knwledgeable enough to serve that role.)
5. Any other guidelines or suggestions?

Re: AI nation skills
 

Wikd Thots:

I've never heard about that Semi Random project but it sounds great. I wish it was posted as a big sticky so people would know about it. Any more info on it? It should have pretenders divided by map size too [at least small and big map].

Re: AI nation skills
 

For those wondering about the NI (or actually BI, as of today) mod, it's been updated.

Re: AI nation skills
 

Edi, does this mean that I should not start making pretenders for AI or is that a seperate issue?

Re: AI nation skills
 

That's a separate issue. These mods only prevent the AI (and players) from recruiting hordes of indie archers, light infantry, militia, light cavalry and other crap that the AI generally likes to use and forces one to rely much more on national troops. The nice indies like crossbowmen, the good heavy infantry, heavy cav etc are still available, as are specials like amazons etc and you can mass archers if you get woodsman provinces, but it makes the AI nations far more interesting since they don't all use the same horde tactics with the same units under different flags. The mods do NOTHING about the AI still designing crap pretenders on auto.

Re: AI nation skills
 

It started as a way to change Gandolphs random maps to something more sensable. Instead of just wild things thrown everywhere people are supposed to design logical provinces. Where the pop, sites, guards, extra equipment on commanders made sense. Such as a province where a magic site allows you to recruit sorcerors should have sorcerors in the province defense. And then a map is randomized from all of those province files to have really cool surprises all over it.

But recently they added sensable AI gods also.

You can check it out in this thread
RanGen program