AI - The Artificial Idiot

[Gandalf Parker]

If I remember right, the 5-script was the reasonable amount of time into a combat that pre-combat suggestions would tend to reach. After all, Mages were considered to be commanders in their own right.

Altho, some sort of general "after that you should probably go offensive" does sound like something Kristoffer might have been willing to consider.

And like others I also do feel that buffs are imbalanced. I tend to avoid researching Alteration because of it. That tends to avoid many of the spells that tends to throw the AI off of script.

Id be willing to mention that to them but before I do Id want to do some testing to make sure there isnt a logic factor Im missing. Something that makes the mage feel nervous that I could have avoided. Ive tested formations, which does help in some spells but not in avoiding buffs. But I havent tested "our numbers vs their numbers" or "did my pretender bother to give me any protective equipment".

[Executor]

Originally Posted by Gandalf Parker

If I remember right, the 5-script was the reasonable amount of time into a combat that pre-combat suggestions would tend to reach. After all, Mages were considered to be commanders in their own right.

However they turned out to be either severely mentally challenged commanders, or just folks who bought a mages diploma and didn't know what to do actually.

IMO, the main problem with late stages of the game is the limited usage of mages in battles. That's why most games revolve around SC's who are just scripted to buff and kill in 90% of the time. And usually that's the main battle strategy in almost every single game as you just can't rely on the AI scripting.
How many of you actually use armies in the late stage of the game? That's a damn fast way to lose any game. The enemy just sets up a trap and most of the time you'll be dead precisely because of the 5 turn rule.
You get that very limited amount of time to hopefully penetrate MR with certain spells, or get lucky if the enemy get's in range, and after that the mages are as good as non-existing for most parts. They quite commonly even do more damage than good.

So, would it be possible to make a "repeat" action? As in you script your mage to cast Pillar of Fire for example, and he continues to do so for the remainder of the battle.

I think this could brink to a very large battle spells diversification. For example instead of having 5 mages scripted to cast say Soul Slay and Bane Fires, who will after the initial 5 turns change to casting something like paralyze (which is in my comparison putting down a fire arm to pick up a pebble to throw at your enemies ) and fire flies, you set one mage to cast Soul Slay, one to cast Bene Fires, one to cast Pillar of Fire etc.
So instead of just relying on the initial 5 spells from your mages, after which you pray they hit something useful every now and than, you get as many rounds of the stuff you've orders for the duration of the battle. So in stead of 5x5 you get 5x50 spells.

Would this idea be feasible or am I just talking out of my ars?

[Gandalf Parker]

Its one of the more feasable ideas Ive heard.
But not because it would make the battles more useful to the players. It would be more feasable because it can be made to fit Kristoffers concept of the game. Its not absolute, and not illogically over-controlling of a battle far far away. It could be pitched as just ... and at that point he is given a general statement of "and then continue doing xxxxx for the rest of the battle". It still would be able to be overridden in the usual manner, and doesnt seem to stretch the instructions beyond a simple battle plan.

[Kref]

Originally Posted by Gandalf Parker

The AI does extensive thinking, and any more would cause more complaints than now.
...

Gandalf, why do you think so? There is a large enough list of MP games on llamaserver. MP games have heavy drawbacks vs. SP games:
player can do no more then one turn per about 3 days, and even worse - he should do so; his waiting time for turn results - three days, much more even if AI would think much longer; player don't customize game as he wants in minutes but should come to agreement with others on forum (and others should also want a game somewhere of that type and in that time); player is not sure that the challenge of the game would randomly and dramatically decrease/decrease due to someone can make a chaotic (not intended to win) warfare or leave a game. And all that for a high chance of a good intellegence of opponents. I agree that turn time may be of importance for beginners, but veterans, especially with games on large maps, generally don't like (I am quite sure) that they make turn for a few hours, and AI makes stupid decisions, also in few minutes. Processor should work for me, not I for him.

[Kref]

2 Calahan:
How do you work on AI mod? I've read "modding.pdf" and didn't found anything that can improve AI, except the low-applyable methods like used in mod that raise resource costs of independents.

[Gandalf Parker]

@Kref:
I agree with what you say. There are many MP games on all 3 of the large servers. And more in games on peoples on machines. But Im one of those people that doesnt think the MP outnumbers the Soloers.

As one of the main server testers for the game I often requested that anything removed for speed of hosting be set aside to make available with a game switch on the command line. Because servers (mostly MP games) dont tend to care how long the game takes to host a turn. Unfortunately we didnt get that.

BUT part of the problem is in the AI itself. Its linear and general. If you turn on debug and read the log it takes up quite abit of its time. And most improvements tend to be in trying to make it less general. Once you start that it gets pretty involved. There are way too many nations, units, strategies in this game to start trying to take them into account for every action

As to modding the AI the only thing available to us is to try and remove things we feel the AI doesnt use well, and boost things that it does

[DeadlyShoe]

Yeah I was pondering just removing Holy Avenger for that reason, or making it a #onebattlespell for some units.

[Kref]

2 Gandalf Parker:
I've recently sent a suggestion of my participating in AI development for Kristoffer and Johan on e-mail (including prototype SK-equipping algorithm, also it doesn't work quickly) but didn't receive an answer yet.
Perhaps I'll put here one of my suggestions right now: is it possible to make a description of all savedgames files (ftherlnd, .trn, .2h) publickly available? This would allow any part of community to develop AI as they like. To use alternative AI anyone should just make a multiplayer game and then make 2h files with alrenative AI programs.
I believe its unlikely that the formats of savedgame files contain very valuable technologies.

P.S. "But Im one of those people that doesnt think the MP outnumbers the Soloers." - I completly agree. The described drawbacks are too heavy. So, the good AI is needed for SP. I don't mind if anyone would use a quick "meat-throwing" AI, I just think that large amount of regular community would like a more sofistickated, even if it would mean a much slower, algorithm to exist. As I understand, dominions take only one core of processor. So, its easy to press alt-tab and do something else while AI thinks.

[Gandalf Parker]

I cant seem to word a response for this forum. Contact me by email to gandalf at the dead.isp of Community Net to continue

[Fantomen]

Originally Posted by Kref

2 Calahan:
How do you work on AI mod? I've read "modding.pdf" and didn't found anything that can improve AI, except the low-applyable methods like used in mod that raise resource costs of independents.

Calahan is altering spells in a way that keeps their effects intact, but makes them more or less attractive to the AI. For example using nextspell commands to hide effects from the AI so it thinks the spells doesn't do anything.

This way it is possible to limit the amount of spells the AI chooses from to those that are the most generally useful. The result being a much more effective AI in practice.

Initial tests have been quite successful as I understood it, I believe we can expect this mod to become very popular when finished.