AI Discussion Thread

[unmerged(42223)]

Ah, how quickly to proliferate...

When it comes to the AI, if one move will do it, skip the long train of events. A set of 15 to 25 variables can be far more than enough.

How many different ways do you as a player in game think about proceeding? It does not take many options to look intelligent.

Yes, expose as many options as possible to modding, just because I hate coming back and asking for more later.

However, in practice, use the fewest possible to accomplish the purpose looking intelligent. Keep all routines straight forward and simple. Be slow to complicate.

Gosh, a few variables can mathematically expound to hundreds with a wink of an eye.

Look at this way.

Keep all things for instance to nine factors, naming them A, B, C, D, E, F, G, H, I.

Now A relates to B, C, D, E, F, G, H, I
And B relates to C, D, E, F, G, H, I
etc.

And further, now compound in threes...

ABC, ABD, ABE, ABF, ABG, ABH, ABI
BCD, BCE, BCF, BCG, BCH, BCI...
etc...

and four, and five

well then relate a three to a four and two to a five etc.
Not very long I will be in the hundreds.

Give me nine distinct characterists of units, of terrain, of movement, of combat etc.

And you get this esculating on and on...



Rule one. Keep it simple
Rule two. Keep straight forward
Rule three. Keep it consistent or logical
Rule four. Keep it explained and clear to understand

Oh mathematics what is your depth? You proliferate to the heavens and leave not hell untouched !!

 

[unmerged(9145)]

A certain amount of AI cheating is pretty much inevitable because AI cheats are just about the only way to avoid sys requirements that would make even a Blue Gene look underpowered. The trick is to make the cheats as "un-noticable" as possible, and as inobtrusive as possible.

GalCiv has a competent AI that doesn't cheat; more than competent, compared to most games. And GalCiv is a far sight more complicated than EU 2 ever was, or EU 3 is likely to be.

Max

 

[unmerged(42223)]

I believe there are two valid points in comparing with GalCiv or with a Battle Mod.

A. It is not a choreograph of history.
B. Paradox games, I am convinced via HOI2 1.3 are simply yet to plumed to their true potential, again because of so much fixation with 'repeating circumstances of history'.

 

[Rotten Venetic]

Rome: Total War's AIs were also almost competent.

 

[unmerged(9145)]

I believe there are two valid points in comparing with GalCiv or with a Battle Mod.

A. It is not a choreograph of history.

Neither are any of the games that Paradox has released so far. They start historically and quickly diverge from that point into something very alternate - whether Paradox intended them to do so or not.

B. Paradox games, I am convinced via HOI2 1.3 are simply yet to plumed to their true potential, again because of so much fixation with 'repeating circumstances of history'.

EU 2 had a bad AI, but the game was simple enough that it could present some challenge if you chose to play a minor nation (at least until you grew to major status). I expect that even with the increased complexity of EU 3 the game will still be fairly simple when compared to the likes of the HOI series, both games of which were distinguished by an appalling AI that just couldn't cope with what was required of it. EU 3 will require far, far less, but should benefit from the fact that it comes after two tries with EU, two with HOI, one with Vicky, and one with Crusader Kings.

Or so I hope. If not, perhaps Paradox can contract out AI development to the GalCiv boys the next time around.

Max

 

[unmerged(42223)]

Neither are any of the games that Paradox has released so far. They start historically and quickly diverge from that point into something very alternate - whether Paradox intended them to do so or not.
Max

Yes, but those responsible for using AI assets never got over the 'historical fixation' and I simply maintain it colors what is possible and demonstratable with the HOI2 AI.

 

[unmerged(37697)]

Some good points raised throughout this thread, by the way - good stuff

The thing is - and I know it's a lot to ask, but I'm not a MPer really - I want the AI to be effective. Let's say I'm England, which I prolly' will be most of the time. If I'm at war with France and Scotland takes the initiative and attacks me from the north, I want to think: "shit, I've got to defend Northumberland from the Scots, which means diverting troops which are needed in Normandy."

Likewise, if I'm little backwater England in 1588, and Spain (Spain, the most powerful nation on earth!) declares war on me because they intend to annihilate my proud little fleet and blockade Portsmouth and Dover, and finally besiege London and force a crippling piece on me; or even vassalise Ye Olde Albion, I want to actually be scared - I want to have to rally my men and have to bring all my ships and troops back home to defend England's shores.

Although a totally different ball game, HoI2 (and DD after it) had a pretty good AI; certainly much better than EU2's.

The thing I hated about EU2 was the fact that the AI never made any serious demands. I am aware of the fact that the ferocity of an AI could be modified - but in terms of the default AI (I hope different nations have drastically different AIs, naturally...), if Spain's captured my south coast and decimated my only major fleet - why should they settle for 300 gold and a slap on the wrist? They'd carry on, they'd march to London, take it, and make damned sure that I pay for abandoning the true faith.

You see what I mean?

I want single player games to be fun, unpredictable, dangerous, adventerous - I want them to be... enthralling!

 

[unmerged(42223)]

Ideally we still look for an AI that will well beat us. Then we can hobble it down for a fun game. Not have the AI lame and then try conjole it to be challengable so we can have fun.

I am not beating on the same drum to create conflict, yet AI modding has a lot philosophy running through it so seeming from those engaged in it.

My objective is to make the clear point to Paradox that there is evidence that they created a much finer game engine with HOI2 AI wise than is generally known. For reasons I have mentioned. Many a time I have read something 'blamed' on the 'hard-coded' game engine like it was some errant boy determined to misbehave, only to find myself these last 18 months discovering some 'solution' with the coding exposed to modders. Admitedly some things needing coded would be easier with less hard-coding, yet again, we are still behind the learning curve when it comes to utilizing what has already been provided by Paradox.

So, concluding be of good cheer and keep it clear to understand the methodogy while exposing more calls to us modders, we will 'get it' by experimenting eventually.

Also, a special note to all those wonderful historically prejudiced players and modders. It is their continued demands for more and more detail in order to more smartly well simulate history, that impels the very progress we all now enjoy.

I'll get off my high horse and move along now...

 

[smn_]

GalCiv has a competent AI that doesn't cheat; more than competent, compared to most games. And GalCiv is a far sight more complicated than EU 2 ever was, or EU 3 is likely to be.

Not disagreeing with the suckiness of the EU2 AI, I want just to register my opinion that EU2 is, contrary to your opinion, not at all a simple game

 

[Rotten Venetic]

It is not simple but it is easy

 

[Sindai]

GalCiv has a competent AI that doesn't cheat; more than competent, compared to most games. And GalCiv is a far sight more complicated than EU 2 ever was, or EU 3 is likely to be.

That's not really true at all. GalCiv is only about as complicated as EU2, and GalCiv2 only adds about as much complexity as EU3 is. The reason GC AI is better is that it's just about the #1 priority of the game's design. There's no multiplayer and the game, where possible, left out features that AI is weak with.

But most importantly, both GalCivs only need to run a handful of AIs that play in turns. The EU3 AI must run over a hundred instances of itself in real time. I'm not saying there's no room for improvement, but there's just no chance of seeing GalCiv level AI in a traditional Paradox-style game.

 

[Ayeshteni]

There could be pre-defined battle plans the AI could refer to, sorted by phases of conflict.

For example, a strong nation at the start of war could choose between a broad front tactic where they storm all provinces they can effectively gain control of or a narrow front tactic where they penetrate deep into enemy territory with the goal of elliminating enemy reinforcementsa and occupying the capital before taking care of the rest.

If losing, a nation could resort to spamming peace offers, desperation tactics, last-ground tactics (where you group your armies in a few provinces and counter-attack when you can), guerilla (harassment; siege-and-run, etc).

A winning nation, depending on how it started, could go for the throat (take the capital) or for the hands (obliterate the enemy's armies before further progress).

.

Concepts are too modern.

Most advanced into enemy territory through areas that could provide good supply routes (and protection) through the advance.

Strategical manoeuver that may force peace through military pressure, militarily gaining a strong strateic position.

then the battle, hoping for the best tactical position. Hitting each other with large heavy bits of metal.

Winner then continues strategic manoeuver etc.

No mass front blitzkrieging across the border. No lightning strikes to an enemies capital (just gets you surrounded and cuts off your supply route). Unless relieved your army in the enemy capital surrenders.

but yes. It should be interesting to see what they do with the AI.

Ayeshteni

EDIT: oh yes, and I dont want to be spammed to death by AI peace offers thank you very much.

 

[unmerged(6777)]

GalCiv has a competent AI that doesn't cheat; more than competent, compared to most games. And GalCiv is a far sight more complicated than EU 2 ever was, or EU 3 is likely to be.

Max

Um....Stardock makes amazingly good oranges so Paradox needs to chop down all its apple trees hunh?

As Sindai correctly pointed out, comparing Stardock's AI to Paradox's AI is pointless because they are completely different types of games. Compare Gal Civ to MOO3, Civ IV, or any other turn-base game and maybe you have a case. But when you compare it to a continuous play game with about 50 x the number of AI opponents, about 1000 more provinces, and an overall level of game complexity that is roughly on par, you do not make anything remotely close to a meaningful case.


(Note: I've played a lot of Gal Civ 2 and Stardock did a really good job with the game's AI. So don't interpret me as bashing Stardock. I'm not. I'm just saying that you can't compare them because they are not the same type of game so the requirements on the AI and player's system are a lot different.)

 

[Assos]

Exactly right, EU and/or any Paradox games has ALOT of AI interaction. Everything the player does, the AIs have to ajust to the difference moves of each other. Extremely complex, I have never played anything as complex as EUs, Vicky, and CK.

 

[unmerged(8915)]

can you guys recommend any good reading on AI theory (game based AI theory that is)?

 

[hansmaa]

can you guys recommend any good reading on AI theory (game based AI theory that is)?

Try this one. I think it is one of the best AI websites around.

 

[unmerged(42223)]

but there's just no chance of seeing GalCiv level AI in a traditional Paradox-style game.

gosh, that seems just a trifle little rough... ,especially that 'no chance' part.

As far as good AI game theory book goes, in addition to that excellent web site, you could collect all the work done around here and start writing a book.

For the story is still evolving...

 

[unmerged(6777)]

gosh, that seems just a trifle little rough... ,especially that 'no chance' part.

It's actually quite true.

We are years away from users having the kind of computer systems available to be able to run a Stardock AI type of model on EU3's scale in real time. Processors would need to be well into the hundreds of GHz range, if not into the THz speeds. RAM requirements would be a lot higher, and the bus speeds would need to accomodate the data transfer rates.

 

[unmerged(42223)]

Double gosh and darn.

I guess I am both skeptic and idealist.

There has got to be a way. Let me take a look at that drawing board again... :wacko:

Anyway, I run HOI2 1.3 on x64 with four processor head to a total of 9+ GHz as well more ram that it can use. But the x64 platform has the single advantage of letting a program float in x32 and use all the resources it generally can handle. I run even multiple instances of HOI2 1.3 at once, just swap them back and forth down on the tray. I wish I could run them in the background though, I guess I could if went through the trouble of windowed mode. And I understand it is possible to dedicate a processor to single instance of a program. But it is that is yet another nasty learning curve for one doing everything oneself all the time.

Anyway, I use only one AI script posted in one place, for all 53 countries (a modded version) and it runs fast. I am sure this is not the same, yet there is always a way. (back to the stubborn idealistic mode ) And the AI never stops working or gets stumped as some complain about.

And as far comparing, I actually cannot as I have never played or looked under the hood of Stardock's AI.

But I am a mathematician knowing there are ways and then there are ways. One of which of course is to create a climate easy for the AI to navigate in.

All is well on the drawing board of an idealistic world.

 

[Rotten Venetic]

No mass front blitzkrieging across the border. No lightning strikes to an enemies capital (just gets you surrounded and cuts off your supply route). Unless relieved your army in the enemy capital surrenders.

EDIT: oh yes, and I dont want to be spammed to death by AI peace offers thank you very much.

A. The AI doesn't need to be irrational like adolf hitler and not relieve the army which has just sliced deep into enemy territory and its flanks are threatened; they will direct their efforts to ensure it does not get cut off.

B. That's exactly why it would be a good tactic

We are years away from users having the kind of computer systems available to be able to run a Stardock AI type of model on EU3's scale in real time. Processors would need to be well into the hundreds of GHz range, if not into the THz speeds. RAM requirements would be a lot higher, and the bus speeds would need to accomodate the data transfer rates.

This is somewhat discouraging. But, for such requirements, are said AI's up to scratch? Can they truly compete with humans?

On the EUIII side, I guess the AI is going to be better than the EUII one, at least it will use more features of the game. If such performance levels as complex strategy cannot be achieved, at least the AI should know the basics of the game well. One option would be to learn from the player: that way your AI gets better and better at countering you, making for a good challenge if quite artificial.

Would that the game would record the player's every move and filter his style into the AI

It would then be interesting to swap AI files between players (if allowed).