Suggestions for Improving the AI opponents

[NTJedi]

Another Topic for Improving the AI for the computer opponents.


1) Higher Percentage of purchased mages placed into research. By day_40 and beyond human players always seem to be leading on research against the AI opponents.

2) Have AI opponents become more aggressive... especially with human players. Very often I see Ermor or another AI remain peaceful with 400 troops on their border where I have only 20 troops... even after 20 turns.

3) STOP the AI opponents from sending their pretenders into the Arena Death Match. Curse, Horror Marks, Battle Afflictions, etc..., not worth the cheap prize.

4) Stop AI opponents from doing empowerment on the weak/cheap 90 gold mages or less. These gems should be used for summoning creatures or casting global/ritual spells.

5) Decrease the number of weak units purchased by AI opponents... especially for late game.

6) AI opponents don't know how to play with a death scale except for ERMOR. As a result any AI opponent with death scales usually end up with armies which are starving when going into battle... which causes low morale and diseases for its armies. Quickest programming fix is have most AI nations use the growth scale. I made this suggestion since the AI opponents are apparently not even aware of when units go starving and/or how to help them(items, nature mages, etc.) .

[ioticus]

Great ideas. I hope the developers are listening and can implement some of these. I simply don't have time to put into multiplayer games so would like to have a better singleplayer experience where I can play at my own pace and not worry about getting my next turn in on time.

By the way, this game reminds me of the best of the fantasy board wargames I used to read the rules for, although I never had the opponents to play with. Thank you Illwinter for creating this masterpiece of fantasy wargame design! I hope(nay, I pray) we will see a Dominions 3 with even more options and better AI!!!

[PvK]

Good suggestions.

I'd also like to see it spend at least a little more on provincial defense, especially on provinces with castles. Maybe 1% of the budget, with 10x the weight distributed to provinces with castles, and the rest distributed based on magic sites and gold income. Often the AI will have say, an enemy capitol, conquered for years of game time, with zero PD on it, so one Black Hawk can lay seige to the castle, etc.

PvK

[Arralen]

The AI still does not build castles, so why should it care about building PD in provinces with castles?

It it where building them, it wouldn't have such a hard time keeping it's hordes of cheap inf from starving... even without a growth scale.

I Agree, though, that the AI should never use a death scale unless it's using some of the themes which have pop-killing dominion build in.

It's way too easy to just fend the AI of at the border and watch it die from it's own dominon...

A.

[Boron]

i never saw the ai summon a unique summon like an earth king or something like this .
even at impossible with standard research .
i played ryhleh in a spgame and didn't bloodhunt but later wish for bloodslaves . still i was able to summon EVERY unique blood summon .

with globals the ai is better . but ctis always casts burden of time and this spell isn't so useful for ctis because it has hordes of cheap ctis troops and so it harms itself .

the ai isn't too bad but once you have done enough research it simply can't defend .
if you start with a good combat pretender you can easily defend + defeat one attacking computer opponent and once you can summon and equip more sc's the ai is finished .
especially with the immortal sc's the ai can't deal . you don't need to equip them very well some cheap items are enough to cope with almost every ai army .
the only "problem" can be the 50 combat rounds if the ai gets lucky and e.g. paralyzes you .

[Boron]

if you want a difficult game you should take a small map with 150-200 provinces with all ai opponents and
use some modded nations as opponents especially cohens crusaders because they have supplyproducing leaders and very strong units so the ai doesn't suffer as heavy under starving effects as with almost any other nation expect ermor and carrion woods .
another "solution" for a harder game is using no sc-pretender and playing a heavy magic nation with bad national troops like tien chi s&a under very difficult research .

a good and easy solution would be if for the ai castlebuilding costs are halfed and supplies are multiplied with at least x4 . perhaps the templebuilding script can be used for castles too since the ai builds temples quite often .
it would be a small cheat but i think help the ai a lot .

[NTJedi]

The temple building script could include a random variable where the AI builds a fort instead of a temple for 1 out of 5 temples. Then a new process doesn't even need to be setup.

[tinkthank]

Great ideas!

Some thoughts:

The AI seems to have problems in these fields:

Strategic Planning:
- The AI builds odd pretenders with no real "plan" to do anything with magic paths, e.g. an earth mother with 7 fire who then does nothing but cast fire shield. AI Pretenders get bizarare paths and often don't really know what to do with magic paths once they have them, either.
Since I bet good ideas are very hard to program, an oversimplified solution could be: Cap number of paths spent on new 40+-costing paths at 1, cap path investment at 4, give priority to dominion increase; give priority to positive scales. Might not be too interesting, but more effective.
- The AI often doesn't know which spells will be good for it to cast later: Maybe give a few "goals" to the themes (e.g.: flaming arrows for machaka, then summon XYZ, then magma eruption, etc.). To avoid over-predictability, three or four "schemes" would have to be developed per nation theme.
- The AI often doesn't know what items to forge; I have seen AI Astral-Earth Molochs playing for Caelum with crystal matrixes running about with no communication slaves in sight.
I think there is no easy answer to this, but perhaps one could make a "library" of helpful items matched to the possible Users (e.g. a fire3-earth-2 death-1 Black Sorc might want a PQR, while a fire-2-death-2-earth-2 might want a XYZ). This is probably very time consuming.

Tactical Management:
- The AI often doesn't know which spells could be good to script in battle. See point above for possible solution.
- The AI will often send out either overkill mobs of troops or too few into a battle. I don't know how the AI currently gauges its opponent's strength -- but there could be some kind of "army matchup algorithm" which attempts to reduce suicide attacks and/or shooting flies with cannons.

Strategic Implementation:
- Fort building and Troop building have been covered.
- The AI often seems to shuffle its troops about randomly, as if it had a multiple personality: I want to go *here*, no, I want to go *there*; I don't know what can help here.
- The AI's troops starve too often. Causes include death scales and lack of forts, but also, this doesn't seem to be something the computer checks for. Have it do so and try to remedy this by producing appropriate supply items.

[mlepinski]

Originally posted by tinkthank:
- The AI often seems to shuffle its troops about randomly, as if it had a multiple personality: I want to go *here*, no, I want to go *there*; I don't know what can help here.

I don't see anything wrong with the AI shuffling it's troops at random. I agree that it might be better if the AI "intelligently" determined where to move it's troops, but that type of "intelligence" is tough to implement and random shuffling does keep the Human player guessing.

As a great philosopher once said: "If you can't be smart, be random"

- Matt Lepinski :->

[tinkthank]

Originally posted by mlepinski:

quote:

---

Originally posted by tinkthank:
- The AI often seems to shuffle its troops about randomly, as if it had a multiple personality: I want to go *here*, no, I want to go *there*; I don't know what can help here.

I don't see anything wrong with the AI shuffling it's troops at random. I agree that it might be better if the AI "intelligently" determined where to move it's troops, but that type of "intelligence" is tough to implement and random shuffling does keep the Human player guessing.

As a great philosopher once said: "If you can't be smart, be random"

- Matt Lepinski :->

---

Heheh who said that??
But you are right, I am making a big deal about nothing.
Say, Matt, didnt you say you werent playing dom2 anymore because of time constraints?