AI for Mages Using Gems

[Zeldor]

The bad thing is that they get gems left they can cast some really ugly spells, like mass protection against abysia So better script them some gem-using spell after the one that you want to cast. So if they have gems cast they waste it on that, not something dangerous. And if not - at least they will pick instead another spell that does not require gems.

[Tuidjy]

I have always been amazed how the AI has no concept of harming its own men.
I have had an Arco human mage in a huge human army cast rain of stones against
enemies that were absolutely no match for my guys, killing some mages and about
a hundred of my archers. Yes, he also wiped out the Maverni opposition, but
who cares? They would have been dead anyway.

[vfb]

Must have gone something like this: Snakebite leader, Bravo Six, for the record, it's my call. Dump everything you got left ON MY POS. I say again, I want all you're holding INSIDE the perimeter. It's a lovely war. Bravo Six Actual and Out.

[sansanjuan]

Tuidjy said:
I have always been amazed how the AI has no concept of harming its own men.
I have had an Arco human mage in a huge human army cast rain of stones against
enemies that were absolutely no match for my guys, killing some mages and about
a hundred of my archers. Yes, he also wiped out the Maverni opposition, but
who cares? They would have been dead anyway.

Would be interesting if commanders/mages had an intelligence attribute (with all the potential implications). Low value IQ commanders would be prone to instigate "friendly fire" and general gem abuse. Smart ones would take advantage of knowing likely battle outcomes (each battle might be run internally with multiple passes, first pass with no magic either side, second as scripted, etc....?). These outcomes would be weighed by each side's mages per each IQ attribute with smart mages pulling up a lawn chair and watching if they deem there is no need to risk unleashing death from above on the entire battlefield. Of course there are way too many permutations and randomness for perfection but might be useful for obvious ones like you've cited. Dom4?

-SSJ

[Tuidjy]

> Dom4?

Dom4 on a Cray supercomputer, maybe... Naah, no way.

There are so many possible actions a mage can take that if you were to have even
one percent of the possible 50 round battles generated, you will not host the
turn on Cray in a day.

[sansanjuan]

Tuidjy said:
> Dom4?

Dom4 on a Cray supercomputer, maybe... Naah, no way.

There are so many possible actions a mage can take that if you were to have even
one percent of the possible 50 round battles generated, you will not host the
turn on Cray in a day.

So we all chip in....

I was wondering what % of a turn's cycles are spent on animation creation. If that was surpressed... For your specific scenario I wonder if even a single iteration might be useful. Just a thought.

-SSJ

[fantasma]

I think the unpredictable and sometimes stupid mage is perfectly normal.

Imagine yourself in a situation where the sky is torn apart, the earth shivers, lightnings flash, etc, etc, a few hundred/thousand monsters are crying and milling about and slaughtering each other and you in the middle of this have exactly the fraction of a second to ponder all the implications of possibly several hundred options.

Sure, you make the right decision 10% of the time, if you are lucky!

Of course it's annoying when you watch the replay with a cold beer in your hand and scratch your head and think oh my god!

My mages cast mass protection always against Abysia! And Aby had more pyromages on the field than troops...

[capnq]

sansanjuan said:I was wondering what % of a turn's cycles are spent on animation creation. If that was surpressed...

Animations are only relevant for watching the replay; there's no reason to calculate the visuals while the battle is being resolved.

[Tuidjy]

While the animations take a great deal of processing power, and skipping them will
reduce the run time considerably, the move selection still takes quite a bit of
time. Not convinced? Replay a major battle, and hit 'n' a few times. This will
skip the animation, and just do the number crunching. It still takes a few
seconds. With hundreds of choices per mage, ten mages on each side, and 50 rounds,
the total number of possible battles dwarfs the number of atoms in the universe.

Yes, some branches may get curtailed, some spells may be ignored as obviously bad,
etc... But it is still a matter of writing better AI, not running exhaustive searches.

[sansanjuan]

capnq said:

Animations are only relevant for watching the replay; there's no reason to calculate the visuals while the battle is being resolved.

ok. I'm a bit intrigued...

So when the .trn file is being generated with the battle sequences, every unit ends up with a "this is what unit-x did this turn", etc. etc.? As the battle engine cycles through commanders and units each unit reacts to every appropriate unit on the battle field for both friends and foes (morale check, can't move there, can swing at bog beast, etc.)? The computational heavy lifting results from all these (each unit's) array checks. If so I can see why it's a bugger to finesse the AI into "deep and clever" code. Thematically I do agree that some mages would nuke their own friends due to panic. I also still am convinced a single attribute (IQ) that would leverage a single with/without magic iteration might be doable. I may be full of hooey... wouldn't be the first time.

-SSJ

Edit: We'd call it the "mage watches battle in lawnchair" iteration. Perhaps only useful if they are the only mage in play for that battle.