Post 0.41h AI feedback needed

i've noticed that the AI seems to calculate and do its moves, THEN do any casting after it has moved.
I think it should do some of the casting (like any area of effect spells, right of fire, etc.), and then after that calculate its moves. It might help them out a little.

phizuol said:

I'm going to echo earlier statements about summoned units, which is made much worse when using illusions since they heal after combat. I wiped out several stacks of units last night by sending a mirror image of Alazkan into enemy territory and letting them suicide on him repeatedly.

I think that's going to be a tough behavior to correct, especially with illusions since sheer numbers aren't enough to destroy them.

Click to expand...

Yes, I just completely destroyed a Ljos army that VASTLY outnumbered me by having Gibbeon twincast summon some illusion earth elementals. This should really be addressed. As much as I hate to say it, the easiest way to fix this is to stop illusions from healing completely after each defensive combat. Something else could be added to them to balance them out (immune to magic?).

The AI Lanun are building pirate coves. Very nice.

Medicine_Man_55 said:

The AI Lanun are building pirate coves. Very nice.

Click to expand...

But they aren't building any ships other than workboats. Not so nice.

Izmir Stinger said:

But they aren't building any ships other than workboats. Not so nice.

Click to expand...

Patch J is obviously a part of the end product, not the entire end product. Personally, I'm pleased with some of the improvements in the AI.

Medicine_Man_55 said:

Personally, I'm pleased with some of the improvements in the AI.

Click to expand...

Some of the improvements are nice. However, J is pretty much unplayable for me, since I can now beat Deity without even trying(I was struggling in Immortal in patch G). I wish that the AI changes were released as an optional beta patch, with the bugfixes in the normal ones, so I could use the old AI but still have the bugfixes.

Here comes an example of one of the quirks of the new AI which is almost qualifying as a bug. It really bothers me to see how many units the AI is building, even in peace times! It slows the mid and late game down a lot and I can't believe, that it is a good thing from an economical standpoint either. Most of them are UNITAI_COUNTER, and UNITAI_BARBSMASHER.

The attached screenshot, taken from a noble game on a standard sized tectonics map shows a stack of 84! malakim barbsmasher units. Keep in mind, that the Malakim own only two cities and the one and only barb city around is the dragons city.

and UNITAI_BARBSMASHER.

Click to expand...

I was complaining about huge BARBSMASHER stacks before. I wonder are all these huge stacks are BARBSMASHER?

Jean Elcard said:

It really bothers me to see how many units the AI is building, even in peace times!

Click to expand...

If you start building units when you already got a war declaration, it is far too late . I hope Kael fixes the bug next patch, which prevents Malakim to go into Conquest mode.

Emptiness said:

• The AI's promotion choices are haphazard and mismanaged. The worst example of this is casters without actual spell promotions, either because of unspent promotions or because they have purchased all non-spell promotions. A caster that can't cast any spells isn't worthless (any unit can be fodder) but is not worth the production investment made in choosing to produce it. The AI should be made to always spend at least the first promotion it buys for a caster to gain a spell. This problem isn't limited to casters, however. I've seen many instances of AI combat units with both Shock I and Formation I (for example). This is disadvantageous for the AI. Taking two levels of Combat would also provide +40% vs Melee and +40% vs Mounted, but would also provide +40% vs every other type of unit, as well as potentially unlocking other useful promotions (March, Magic Resistance, Guardsman, et c.) Only on a unit with Combat IV+ would mixing Shock and Formation potentially be viable, and even then it would probaby be better to have one unit with Shock I & II and one unit with Formation I & II than to have two units each with Shock I and Formation I. If the pre-H AI had these problems I had not witnessed them. Another, more specific, instance of this problem is Barnaxus having been promoted with City Raider I - III, instead of Combat promotions. I fail to grasp any situation in which Barnaxus should choose any promotion other than Combat promotions until he has all five. Although there is certainly room for different promotion philosophies, and I can accept that the AI won't be using my philosophy, this seems like a situation where common sense should produce the same decision in every viable philosophy.

Click to expand...

AI will give combat promotions to casters only if
1.) unit has enough experience to be upgraded to mage/archmage
2.) none of the available spells makes sense (like fair winds for a city mage)
sure there is still some optimization possible but not really worth the time IMO. The promotion selection for arcane units pre patch h was total garbage.

Shock and the other counter promotion work different from what you think. They don't increase strength by 40%, they reduce enemy strength by 40%, that's why shock I is often better than two Combat promotions.

Emptiness said:

Turn 417: In Evermore can be seen an injured Longbowman. Many turns before I had noticed this unit injured, to the same degree. Watching the unit over the next several turns it does not heal. I can see no cause for this, so I conclude that it must be moving out of the city and then back in each turn. This behavior makes no sense for an injured defender in a threatened city, or even for any injured unit with access to a friendly city with other defenders (ie safe place to heal quickly).

Click to expand...

a good observation. The problem is that if you call the BTS function AI_retreatToCity with a unit already in a city, it can often happen that the unit rather than skipping the turn and healing instead wants to retreat to another city. But once the unit has left the city it will decide to retreat to the city it just left.

easy to fix by putting a
if (!plot()->isCity())
before calling
AI_retreatToCity
in the move functions that were added in CvUnitAI

While the new a.i. needs some improving, the fact that Sephi (and Kael, although less so in regards to this particular facet) have taken the time, energy, and effort to make our experience more enjoyable isn't lost on me.

Thanks, guys. The system needs work, but I'm glad you guys are trying to make it better.

I just finished a game on a continent map. I started on my own little island as the Amurites, and decided to go for a Tower of Mastery win (I had to extend my empire to another island nearby for enough mana nodes to do it). My problem, though, is that in a game of 450 turns, I never once saw any other player. Two barbarian ships attacked me the entire game, in which I quickly crushed them with Drown. The computer circumnavigated the globe around turn 300, so they had to have at least one ship. But the fact that they never made it to my empire made me realize how poor the the ship AI really is. I guess it is back to pangea-like maps for me.

Also, for the first time ever I started getting games with turns that would never end. I had to reload, jump into world builder, and delete a bunch of stacks of enemy units to continue. The AI really is building too many units.

Was playing with my friend in a LAN the other day against some AI's. He ended up settled fairly near the Kahzad, so he went for early conquest. Built 7 warriors and headed down. He was somewhat surprised to see the AI also with about 7 and combat promos from aggressive (His first time playing since before H).
He proceeded to siege the dwarven city, pillaging improvements and camping out on good resources while we teched towards catapults (among other things). The AI would occasionally attack, but only with up to three units or so. One major problem the AI had was that the sieging units were across a river. This effectively negated the slight advantage from aggressive. If I was playing the dwarves, I would have made a sizeable stack and moved them around to attack from the other side. This would mean splitting my forces, but it would give me much better shot at winning some combats and generating shock promoted troops. The AI did send out small groups of two or so warriors to randomly patrol, but there really wasn't anything they could have accomplished.
Eventually, catapults came around and the dwarves were crushed without incident. It's good that the AI was able to field enough units to defend against an early warrior rush, but it needs a better idea of what to do with those warriors.
Hopefully that's what's being worked on right now.

The new AI seems to be massively improved, I've learned rust can be a pretty darn good spell after the AIs used it against me several times. The new barbs are much smarter than the old ones and make for some early game strategy instead of just sitting behind walls getting XP.

The issue i've seen is directly related to Fafnirs, AI city defence. The barbs now do the 'warrior' rush his freind did every game, they stockpile units till they think they can attack, in my games this has pretty consistently wiped out at 1\3 or more of the AI players before Orthus even appears. Barbs seem to be spawning and attacking earlier too, as I've seen AI wiped out by turn 10 several times. Its fitting that Orthus should have a good chance of wiping out a civ or more, but it is irritating to see the level of competition drop so sharply so early on a regular basis.

Also on city defence the AI doesn't seem to have a dogpile instinct to wipe out strong units. The most noteable case of this is when I sent in two axemen and a tribuchet as a diversion against a city of 15 units, chariots and archers. I thought they would kill the diversion and move on, but they just huddled in the city and let my three units whittle them down until they alone captured the city, with re-enforcements this meant that one unit killed almost 30. One axeman had STR of 13.8 with weapons, orthus's axe, combat, empower, and enchanted blade promotions and 3-7 first strikes, so he was a beast, but he was still taking damage most turns and needed Sinoras Touch to keep at full health. Realistically he couldn't have survived more than 5 chariots dogpiling him in one turn. After this he teamed up with Valin and they became a two man city killing machine, they could walk in, pick a city, and whittle it down without being attacked even in the middle of enemy territory. I suppose the main SoD would have sacrificed a few units to take down a hero or other strong unit, but other than that they seem to be unwilling to touch units they have a >1% chance of killing first hit.

Its been said before but the AI scouting for a good start loc is great, but they settle too close sometimes and that puts both the AI's way behind in the long run.

Thanks Sephi, the patch is great, I'm very impressed with how much the AI has been taught to do.

Game settings: Standard Pangea, Monarch difficulty, low sea level, random climate, 4 AI, marathon speed, allow permanent alliances, no tech brokering, all victories enabled.

-------------

Sephi said:

Shock and the other counter promotion work different from what you think. They don't increase strength by 40%, they reduce enemy strength by 40%, that's why shock I is often better than two Combat promotions.

Click to expand...

Ah, yes, yet another thing I've learned on these forums and then promptly forgotten. Stacking Shock and Formation isn't so bad then.

This is an interesting AI change, but the two biggest problems I'm seeing is that the AI builds way too many units, and is lackluster, even completely handicapped, when it comes to the sea.

Is it really necessary for the AI to build so many units? It's really gamebreaking. Unless I go into world builder and delete their monstrous stacks now and then, my game will crash.

Is there any way to start a game with only AI's in it? I mean, no human player, no me. I just want to test some AI changes I am making and wish to see the AI play against itself without my involvement in any way (and the game will go much faster without me too)?

you can turn require complete kills on, open world builder, delete your starting units, and give yourself a shadow in the corner.

With the cheatcode set in the ini file press shift ~ to open the python console. Then type CyGame().setAIAutoPlay(X) where X is the number of turns you wish the AI to play. The AI will take over your civilization.

If you press ctrl + z you enter debug mode and can easily see what each AI is up to. (zooming into cities, checking their advisors, etc.) Press it again to turn it off. Problably best to have it off while the AI is auto playing, might speed things up a little (less graphics to redraw all the time).

perfect!
thank you