Proof of stupid AI spellcasting

[geo981010]

Originally posted by Wendigo:
Then you should rename the thread PDF, because your example proves nothing.

This was a mistake of the strategic AI sending to the arena a mage that knew no combat spells (ie, had a little thau research, but no evo or alt research). Mages default to cast, if the mage just knows crap he will cast crap regardless of how good or bad the spell AI be.

Despite the nay-sayers, the spell AI is one of the best parts of the game. You want some proof? get the same mage, research evo1 & thau1 and send him to the arena...he will cast star fires instead of com master/slave. But you cannot plant melons & expect apples to grow.

Hmmm, I could go so far as to say the AI is ok, but that due to the nature of the spell complexity that this is a really complicated process that needs some improving. You really need special logic for a lot of special cases like this - it should prevent mages from casting Communion slave without having any mages to benefit from this, or communion master if there are no mages to be slaves possible.

And the mage would have been able to at least cast Twist Fate, so he did have one more good spell to cast. If nothing else, then I guess he should have entered melee after that. but let's face it, he was dead.

Besides specific spell quirks like the Communions, the AI does have other problems that cause some problems:
a) no concept of it's chance of successfully hitting a target. Like a Warlock trying to Nether Darts an lone final unit, while he should be Mind Burning.

b) no concept of it's chance of successfully beating MR. The AI loves to repeatedly attack a SC pretender with MR check spells in his own dominion with 20+ MR, while he could instead by taking out the calvary units riding down on him and destroying his line that are the more immediate threat. I'd like to factor in the closeness of the unit to the mage and to friendly troops, how long before it could be in range (ie flyers should have a higher rating here), and some threat level (HP, skills, etc) with some random factors so it doesn't just always strike a particular unit.

c) no concept of friendly fire losses. Same case as a, but this time your warlock nukes a good portion of your troops. Even worse if that warlock has an eye loss or blinded. There really should be some logic for checking whether area/multiple effect spells should be used or the 100 accuracy single target ones based on the size of the force facing you, and how likely your troops are to get hurt which takes into account the mages precision.

d) Does the AI still pick spells that are obviously lesser Versions of another? Like Blessing instead of Divine Blessing, or FirebLast instead of Fireball. Unless this was fixed recently, this was a concern in previous Versions.

e) Casting area effect protections spells in crowds. Breath of Winter, Soul Drain, Fire Shield, etc from mages that fry your nearby leaders/bodyguards is an annoying flaw. You learn very quickly to place mages who don't play nice away from others, but early on you just get pissed and later on you only occasionally get pissed as you will forget one. The one I forget about the most is a random magic 1 Fire mage who casts fireshield - have to remember to sort them out and name them properly so I don't forget!

Sure there are more too. I am not alone in basically scripting out all my battle mages first 5 spells, and I know I definitely restrict my research to prevent my mages from casting spells I think will cause more damage on my side than not. That's a sign that the AI could be improved.