Making AIs feel unique

[Anskiy]

As has been noted in other discussions, the AI currently tends to be rather predictable in what they pick, based on their schools. Sometimes the AI can genuinely pull some surprises, usually when doing triple realm plays. However I find that they tend to collapse onto certain predictable plays, like the Sorcery wizard massing water elementals, the Chaos wizard massing Gargoyles, the Nature wizard usually massing Spiders, and so on. I picked the No Trading option to help diversify them some, and while it helps to a degree, unfortunately it still seems like they go for the "standard" setups most of the time. Is there a good way to make them more unpredictable? I wouldn't care if it made them play somewhat worse either, I prefer an AI that surprises me and sometimes fails over an AI that plays it safe and succeeds more often.

[Seravy]

Not at the moment but I'm considering to add modding for the AI's choices on their starting and guaranteed spell picks later.
We might also need modding for research priorities to go along with that though.

[Suppanut]

I think this thing become problematic with "no trade" option as it is much harder for ai to get spell to perform excellent combo. I think ai should have option priority that with more variety so that they would have more options to do in case first priority stretegy could not perform. It appear as that one of Anskiy's game with no trade option have chaos wizards continue using hell hounds again and again in year 1412 or life-death wizard still using ghouls as main force in 1415 which suppose to be time which they should use uncommon spells more often. It is likely that ai could not form variety or secondary strategy when prefer effective CoM combo strategy could not perform. I think main strategy could be set priority as 85% of the time while other secondary strategies could be divide with last 15% like classic setting of Paradox game when trigger events (AI always pick first (canonical/historical) option at rate 85% of the time, while the rest of 15% is alternative option).

And if possible, perhaps ai personality should effect on unit types they produced and summoned too.

[zitro1987]

Generally does not bother me too much as i often see reasonable variety on their spell usage, though the way they protect town with magicians or ranged units instead of sturdy tank units is not that fun (and if you have the right counters, it is kind of inefficient strategy)

I mean, AI even prefers sprites over bears when bears behind walls can take a beating

A mix of ranged and slow melee is the ideal garrison

[JustOneMoreTurn]

Regarding the "AI bad garrison choices" issue, perhaps start a thread about bad choices you encounter in games. I just saw a Troll Wizard wiped out because 9 sprites was not a good defence ... and he had troll magicians and halbardiers outside his city. I'm sure other people can post more examples of what the AI should avoid.

I also suggest that Just Cause have a 'wait until you have x towns' delay for AIs. They waste their time and resources by casting it too early.

[Seravy]

This "9 sprites" issue is recurring but no better solution was found.
Every unit is countered by something, of course if you have the right counter to sprites, they are super easy but that's true for other units.
In particular, the above mentioned bears are a joke if you bring a ranged army.

Sprites have a garrison priority of 20. Troll Magicians have 55. The AI would never pick Sprites over Troll Magicians when they have that choice, but that choice ONLY comes up if a 10th unit is produced or summoned in the city.
Magicians produced elsewhere will never replace the sprites unless some of them are killed and empty space demands a new unit to be pulled into the city.

Mixing garrisons up by making units that have too many copies in the city lose priority is easy to do and possible. I think it was already discussed years ago why that would make the AI weaker but I don't remember. Probably because instead of a well defined strength and weak point, a mixed garrison is just weak to everything. Use your spells to kill what your units are bad against and the units beat the rest.

Also, the matrix of results is rather straightforward here :
Defender Melee Loses to Ranged (can't stay behind wall losing defender advantage, wall gets in way of charging towards enemy)
Defender Melee Ties with attacker Melee (don't engage and let spells do the talking, as the human player you can do that better than the AI, especially having the luxury of picking units for your army that are most resistant to the enemy magic available)
Defender Ranged beats attacker Ranged (both glass cannons but defender has wall and fires first)
Defender Ranged ties with attacker Melee (either the ranged units can kill fast enough or the melee units reach and slaughter them, walls don't help)

So picking the ranged unit of equal strength is always better.
Mixed units work best if there are 1-2 melee units for the gate and 7-8 ranged. As AIs typically put 1-3 strong (usually high tier summoned) creatures in their garrison and generic units for the rest, this usually happens without any additional changes to the code although the strong unit isn't guaranteed to guard the gate or even be melee, some of them like a Great Wyrm will go on the offensive, others like Colossus might be ranged themselves but that's still good enough.