An AI question for a Dev

[Particle]

Originally posted by von_Schmidt:
I do agree that the AI issues and suggestions seem to be quite valid. But cut the devs some slack; that's all I'm saying.

Indeed. I agree with this.

[licker]

Children children...

Thanks for the post Kris, I hope that you are able to figure out some way to externalize some parameters without having to do too much new work.

Anyway, the seed has now been planted, and maybe taken root in your mind, so for the next project perhaps you'll look in this direction. I think its a good direction to move in as it allows you to pull in the resources of the fan base for getting better settings. And the best thing is that at some point you can use different settings to actually make Easy, Normal, Difficult, Impossible, actually have more differences than just more or fewer resources for the computer players

[Gandalf Parker]

I have taken a hand in externalizing AI. In a MUD game (which is a freely available option Id highly recomend to anyone interested) but the particulars would be much the same. I love games that provide it and do believe it would do much toward increasing the lifespan of this game. Im not sure how likely it would be to add it though.

The first thing needed would be recognizable tags for everything the AI would "see" and do. If those are not already in the game (which I dont know) then there would be ALOT of programming to put them in. If we pick one of the AI suggestions we could discuss it further as far as what info would have to be made avaiable, what actions would have to be made available, etc.

[licker]

Ok, lets pick 'army composition' as the element to externalize. Lets ignore commanders and summons for now, focus just on balances with recruitable units.

First lets assume that the AI has a composite picture of its current nation every turn, that would be tresury, income, upkeep, resources (gotta break this one up a bit to capital, forts, and other) and existing army composition.

Now the AI would have to have some default setting (this is the externalized part) that defines what ratio of different unit types it should strive for. This is further complicated by how many armies the AI will have, but lets just look at the national picture and worry about creating the armies later.

So you would have a weighted list of unit types (could be all units in the game, but that seems like overkill).

So to do this in a simple way...

Militia
LI
LI-ranged
MI
MI-ranged
HI
HI-ranged
Archers (short, long, comp, xbow)
LC
LC-ranged
HC
HC-ranged (does that even exist?)
Flyers (perhaps light and heavy? how many differnet flyers are there outside of caelum)

Aquatics would have a different list, but lets stick to the land for the time being.

So from the above list of units types a check is made to see if the nation even has the posibility to recruit the unit type, followd by the amount of available resources to spend on each unit type. Then some algorythem dependant on other settings (national preferences, AI personalities, counter needs, ...) is applied and the existing %s are modified to new values (or left unchanged as needed). Finally, units are ordered to fit the new %s.

That's how I'd probably approach army compostion anyway. Realize this is only part of the overall package, the location of the units recruitment and the recruitment of commanders would also have to be factored in somehow, but that should happen in another matrix I think, though the matricies should be tied together through the controling algorythem(s) that affect them.

[Gandalf Parker]

Not bad. Not bad at all. You went a different direction than I thought you would. I must admit that I expected to do alot of "AH HA! but look at this" Posts but you cut me off.

So rather than a full scripting language with if-this then-this triggers and responses, you are looking more for external variables? Allowing the game to do the recognition and math?

OK for troops that can work. And although I can see alot of "the AI did a stupid thing" it might also provide some intelligent moves. So how would you set the parameters for Ulm?

[licker]

Originally posted by Gandalf Parker:
Not bad. Not bad at all. You went a different direction than I thought you would. I must admit that I expected to do alot of "AH HA! but look at this" Posts but you cut me off.

So rather than a full scripting language with if-this then-this triggers and responses, you are looking more for external variables? Allowing the game to do the recognition and math?

OK for troops that can work. And although I can see alot of "the AI did a stupid thing" it might also provide some intelligent moves. So how would you set the parameters for Ulm?

Heh, glad you see the simplicity of what I'm suggesting. Its alot of matricies, but once you have them set up, its more about making the appropriate algorythem to use them than a butt load of scripting

I think that if the game is keeping track of all the information anyway, the AI might as well use it with some rules (the algorythems I speak of) to make 'better' decisions as to how to spend its resources. I admit this doesn't make the AI better strategically necessarilly, but it makes the AI have a better rounded (or focused) nation to use.

I'm loathe to actually try and do this for any nation, as I am really quite new to Dominions, I was hoping that some of you vets, who know what works for you, would take a crack at it, then I can try to fill in the cracks or elaborate on the driving algorythems.

But since you asked...

First you have to determine what kind of pretender Ulm will have, and you have to decide on the theme. If we are using the 'personalities' that has to be taken into consideration as well. Anyway, Ulms strength would either be in its HI or blessed Knights (depending on pretneder). So I'd set up a generic table with 20% xbow, 20% LI/Militia 40% HI, and 20% HC. The numbers would then be effected on a per turn basis (not drasticly) as Ulm expands and the gold/resource ratio changes. The specific units within a class should probably be an equal mixture (for Ulm since the differnces arn't great). Those specific units (flail v. axe) can have their weights adjusted later on when (and or if) enemy forces are also considered.

That's crude, and I'm no expert (especially not with Ulm), but does it give a decent example of what I'm after? I think you'd do better to set one up yourself and let me ask questions and comment on it

[georgevtucker]

My $.02:

I don't know anything about the programming end of the AIs. Having said that -- I think it would be a good idea to enable a set of likes/dislikes, or attack priorities, for the AI players.

For instance, Marignon should always attack Ermor in preference to any other neighbor. Abysia and Jotunheim would seem natural enemies, as well. Alliances between "enemy" nations would be disallowed.

To extend this thought further -- nations that have similar requirements, Abysia/Machaka and Caelum/Jotunheim, would be least likely to attack one another. If their scales are going in the same direction, they're working toward the same end. This kind of grudging friendship would both help and hinder the human player, and increase the depth and personality of the separate nations.

At the highest level of difficulty, all AI players would attack human players first, then neutral, and one another Last.

I'm thinking of old Master of Orion, where all the races had separate personalities unless you played on the highest difficulty, when they all became Aggressive Xenophobes/John Ashcrofts.

Having said all this -- thanks guys for the outstanding game.