AI Thoughts

[RedRover]

This is the last of my holiday posts. Just a few thoughts about tweaking the Dom2 AI to make things a bit more challenging at higher than Normal difficulty:

1) The AI usually goes after independents before the player.

Give the AI a 50% chance to always target the weakest province first, regardless of who owns it.


2) Reduce the 10-PD exploit.

If the AI counts province defense when determining province strength, then give it a 50% chance to ignore the PD for purposes of targeting, but count the PD for purposes of assigning forces to the attack.


3) The AI seems to count province defense as troops when deciding whether to declare war.

Have the AI ignore province defense 50% of the time when deciding to declare war.

4) Limit the duration of AI wars against AI positions.

a) If the AI is attacking the AI, give the war a 5%/turn (cumulative) chance to end.

When a war ends, have the AI ignore the old enemy for 2d6+oe turns. Maybe create a "#peacewith" code that will prevent the AI from attacking the designated nation until it is cancelled. Such a peace binds both nations equally.

b) Or, give the AI a chance to end a war after it has taken some number of provinces

(example: Two + 1-4 provinces).


5) Fix the declaration of war so the AI will only declare war if it has forces in a position to directly attack.

Once the AI declares war, it should funnel troops toward the front normally.

Anybody else have ideas on cheap ways to tweak the current AI?

[PvK]

* Never send pretenders into the deathmatch arena.

* Build some supply-generating items and use them on units sent to accompany large food-eating armies. (Scouts with retreat orders, even?)

[Wick]

PvK said:
* Never send pretenders into the deathmatch arena.

Can we make that "never send uniques into the deathmatch arena." The worst contender I ever saw was a irreplacable, non-combatant, out of dominion immortal with a valuable special ability. She won her firt round too -- the SG Acolyte paralyzed her and kept touching and touching until the kinkiness freaked him out and he fled.

[Cainehill]

*chuckle* You're saying the AI sent the immortal Tien Chi healer to the deathmatches? LOL!

[RedRover]

PvK: I remain ambivalant about Pretenders in the Arena. It sometimes provides useful intelligence about the other Pretenders out there.

I do know that I would never send a combatant to an Arena combat if I had the slightest expectation that a Pretender would compete.

That said, I would like to see some more Arena options, such as:

Allow the PLAYER the option to host the Arena fight. For example, a "Host Arena Fight" event might go into your event mix as soon as you hold an Arena site. (Or maybe, the event is randomly assigned to an appropriate player, based on the Arenas in play.)

When the event occurs, the player sets the Arena match parameters:

1. Pretender only
2. Prophet only
3. Legendary Hero, must have the mark.
3. Hero (ie: at least one stat 18+, possibly by item boost)
4. Commanders only (no adjusted stat higher than, say, 16).
5. Open Arena (anything goes)

Additional parameters might include altering the prize, or requiring an entry fee that goes into the prize pool.

I would hate to lose the Arena completely, but would like to see it more flexible than it is in Dom2.

It might also be fun to have a "Monster Fight" variant, in which the fighter gets a prize if a monster is defeated--the prize (Item Type) determined by monster toughness.

Perhaps an individual could win only once at each level. Maybe an individual winning at every level gets an XP bonus sufficient to get into the Hall of Fame, and an heroic ability if he/she doesn't have one.

[Olive]

6 - AI blood nations never seems to blood hunt and summon blood creatures.

Well, don't know if blood hunting is easy to code. Here's an easy way of how it could work : at any multiple of N provinces owned by the nation, AI should choose a province (the nearest of 5000 pop ? ), build a lab, put taxes to 0, recruit and send 3 bloodhunters. N would depend of the nation and theme.

Well, that's pretty basic, certainly not always 100% efficient, but it should do the job correctly without lot of subtlety in the code optimisation.

[Endoperez]

The AI does hunt blood. It isn't very effective, but it does do something. I think the biggest problem is that Mictlan doesn't use the slaves effectively (summoning cheap summons/random summons that don't work together/ blood sacrifice) and that other nations don't have that many blood hunters.

Maybe one variable in the AI algorithm would be the gold cost of the most effective blood hunter? E.g. Jotunheim would have only few Skrattis hunting slaves, because they are expensive.
Also, if AI had lots of blood mages, it should know to start bloodhunting (Vaetti Hags have random Sorcery, so 1/4 has Blood, and they are very cheap researchers).

[thejeff]

But because the AI doesn't build castles, it can only build one national leader a turn, which means it has to choose between a cheap blood hunter and other more useful mages.

I'm not very good with the blood nations, but I know I don't really get the blood economy pumping until I can start dedicating castles to churning out cheap hunters, while still building some actual casters.

[NTJedi]

I know some AI opponents like ERMOR don't research any blood magic!

One of the biggest game improvements would be allowing us gamers to disable/enable specific spells and items. It's sad seeing the AI waste tons of earth gems on 20+ cave drakes during late game.

[Endoperez]

I happen to prefer enemies like Ermor not researching any blood magic...

Does the AI cast ritual spells that are meant to build castles?

If, say, Three Red Seconds was made a national spell for all the AI nations, removed from the research tree, and made Astral 1 ritual costing 5 gems... Could the AI use it?