Improving tactical AI and other AI considerations

klokwerk

Prince
Joined
Sep 3, 2006
Messages
433
Location
Paris
I love the expansion. I think overall the AI is good because it's very hard to make an AI play a game like that.

Still, there are several things I see that could improve :

1) On city attacks :

The AI has a lot of trouble taking cities because :
- they don't build enough siege weapons
- they don't see how to protect their units

Too often, they send tons of melee units in front of our cities where they get instantly slaughtered. They sometimes have ranged units, but clearly not enough to take player cities, which they could have calculated before deciding to attack (depending on enemy city defense). They need to build more siege weapons.

Solutions I see :

*) build a lot more siege weapons : they can also be used defensively anyway, especially in city defense, so it's always a good idea.
*) city attack strategy :

a - don't attack any city if you don't have enough siege / range units to begin with. Check how much siege / range you'll need to take the city approximately BEFORE attacking the city and even during the attack. Siege weapons techs are top priority for city attacks (beeline). We should never see tons of melee units coming at a city then just do nothing because duh, I forgot I couldn't take the city so let's wait instead. Take the city or retreat. Minimize losses.
b - send melee in front but if possible outside the city perimeter (get around the city but outside 2 hex perimeter, so 3 hex). They're just here to clean the opponent's units and making harder to kill the incoming siege units. On the same turn, prepare siege weapons so that they're all basically at 3 or 4 hex distance (I mean 1 turn from being in city perimeter, ready to break the city defence).
c - then (1 or 2 turns after) try to get every siege (and possibly ranged) weapon involved in the attack enter the city perimeter (2 hex) on the SAME turn. Retreat if city taking chances become too low.
d - start city bombing from everywhere until city is near zero defense and finish with a melee.

It seems to me this part could be better executed, although an AI is really difficult to make with 1 unit per tile...

*) run a threat check on every unit hex : a unit should never finish on a hex where it will likely be killed unless it means the AI will kill more of them. Retreat / replace units when possible, especially siege units (plan for backup units if you can't make them enter the city perimeter all around the city simultaneously).

*) a land unit should never finish its turn at sea if the enemy controls the sea (AI still sacrifices tons of military units by sending them at sea just around my navy, that's insane, maybe have a check that detects AI units were recently killed at sea around the same place, and if so, if the player "rules the sea" according to this check, don't send any land units at sea anymore).

2) Defenseless cities are sometimes back. That should never happen, it's too tempting (if it happens, such a city should ALWAYS be attacked by another AI or by the player, since it's probably the best move).

3) Detect when a player / AI is about to win earlier and fight back better. We need to see more AI buying city states to stop diplo win, cutting enemy tourism or stopping space win (atomic bomb / capital attack ...).

4) Fight more for delegates.

5) Maybe give back a bit more gold to the AI at the start, it seems to have a lot more trouble early, probably because advantages were reduced (at least on emperor level).

6) Early warmongers (with ancient UUs) should try to use this advantage more by going to war very early.

7) Ideologies : you can anticipate revolutions / civil unrest better, no need to go to an ideology you'll have to abandon in a few turns anyway... Especially for low culture civs.

8) Evaluate defense pacts better, it just doesn't work. Japan / Morocco / me (Indonesia) were good friends and I had a pact with Morocco, making tons of money in trade. Now Japan decides to attack... his friend Morocco, which means I automatically attack him. It was really a poor decision. On a more general way, use friend blocks better like humans do, you don't want to be enemy with everybody, friends matter.

9) Make DoW bribes a lot more difficult. I shouldn't be able to manipulate AI strategic decisions that much since they can't do so with me. Make it 2 or 3 times more expansive. It should be nearly impossible to force an AI to go to war.

10) Make the AI less passive. If they're too low on score, maybe they can be successful on culture, or at least be part of a bigger friend block. An AI should never just look a the player / AI winning and just do nothing.

Feel free to comment or add other things.
 
Yeah, i agree with what you say.

AI is not only not aggressive enough (towards human player), but when it is, it doesn't use Siege units almost at all and the rest it uses them inefficiently. Such a huge difference in AI quality vs CiV IV it's such a shame, when many other features are really interesting and fun.

I think the developers' problem is the one unit per tile thing, they have just failed to mix it with efficient AI.

If they haven't been able within 3 years to fix the AI, i don't see it happening anytime soon, i hope i am wrong.
 
To be fair, it's a lot harder with 1 unit per tile to make a good tactical AI.

Still, some of the things Civ 4 Blake's AI (after BetterAI mod which later become officially implemented) are still not done today by Civ 5 AI, which is a shame because it could definitely be done.

For example, threat checks were a lot better in Civ 4 BetterAI, the AI didn't sacrifice half of the units it sacrifices now. It was also able to make massive sea invasions. It could also make "fake" attacks where the real target was another city elsewhere.

First step in my opinion would be better hex / unit threat checks (no more land units slaughtered at sea every turn by enemy navy, no more 10s of zulu impis wandering and getting slaughtered by enemy city defense...) and building more siege units. Then making those siege units pop around enemy city all at the same turn.

The way they attack cities now doesn't make sense. Half of the time they don't even bring siege weapons, so there's zero chance they don't get run over by the player.

Then 5 turns later they ask for peace giving you their second best city without me even threatening them in any way. And they were the ones who declared war, I just killed their small melee / ranged waves. Talk about suicide, and it's on emperor level...

Bring big numbers or don't come, bring siege weapons or don't come at all... Check every unit isn't alone under fire at the end of turn. Or retreat them.
 
I personally would be happy if the AI used Archers instead of siege (at least until/if that get's changed). At the high level the majority of cities taken by players are taken with Composite Bows (or camel archers or keshiks). If the AI is waiting for catapults (which are slow and useless for much else) they are just going to lose. That's a separate issue though (how useless siege is). Hell in my last game I took down a 50 strength city with (ridiculously upgraded) crossbows. Let's at least be honest here, the AI doesn't really understand how OP ranged is quite yet and cannot use it as effectively as we can (they have no clue how to use a chariot archer or their ilk).
 
Still, some of the things Civ 4 Blake's AI (after BetterAI mod which later become officially implemented) are still not done today by Civ 5 AI, which is a shame because it could definitely be done.

I totally agree. The BetterAI mod made the AI far, far more "humanlike" than the AI behaves now in CIV 5. The main reason why i still play CIV IV. ;) There had been an interesting thread about combining units. (http://forums.civfanatics.com/showthread.php?t=503278) If the AI would be able to combine units, let's say a melee unit and a siege unit, the gameplay would become far more challenging. Also, the AI would be finally able to protect their armies on the ocean. :rolleyes:
 
Top Bottom