Post 0.41h AI feedback needed

[Lone Wolf]

None of them had a single catapult.

That may be related to the bug I described in that thread and which will be fixed in the next patch.

 

[bonedog]

In my most recent game I spotted a goblin unit heading towards the Ljosalfar capital, and moved in to intercept with a warrior to earn some cheap xp. On turn 12 the goblin entered the Evermore's BFC one turn before I could attack, so I was going to back off (not wanting to declare war on the elves just yet) when I noticed the city was completely undefended. Needless to say, by turn 13 I had killed the goblin and seized Evermore.

Was it intended behavior for the AI to leave Evermore undefended like that? Is it because the Ljosalfar start off with two scouts, and so the AI sends off both to explore instead of leaving one behind to protect their capital, hoping to churn out a warrior quickly before someone waltzes in and destroys them?

 

[DioAurion]

In some ways, the AI has become smarter. The Fellowship-AIs beelined for Hidden path. Hannah founded the Overlords as she should.
The game seemed more peaceful than usual but that needn't be caued by the patch.

But the AI has bad problems with aggressive wars. Both Alexis and Cassiel sent a huge stack that contained hardly anything but archers, which they had both promoted with city defender. None of them had a single catapult.
By sheer mass, Cassiel managed to take one of my cities while I was healing from Alexis' attack. He didn't raze the city but instead moved every single unit on to my next city, leaving the first city undefended.

Later in the game the problem with them leaving cities undefended becomes less noticeable. You'll see the AI start to spam city defenders and rush them in to control their new cities. They also are not against turning their stack of doom around if you try to rush in and recapture your city. I haven't played pure ffh with the changes yet but this is how it has been working on the wild mana mod so I'm sure it isn't much different.

 

[Onionsoilder]

The AI can, in fact, still build catapults. I saw the Luchuruip with a stack at least 20 of them. They also had a lot of Fireball Golems, though they didn't understand sacrificial mechanics. They would summon 10-15 fireballs, then let them die the next turn because my stack was too big.

The barbarian AI is much, much more annoying, but it's not too hard to put up with. Then again, I only have to deal with skeletons and orcs, so my warriors could take care of them. I can see were lizardmen can be problematic.

 

[Medicine_Man_55]

Incidentally, I'm not complaining about the difficulty posed by barbarians in 2.41h. I'm just laughing at the irony. I have been planning for some time to set up a game of FFH2 with me alone in the middle of a "continental barbarian super highway". It is hysterical that I chose this patch to do it.

The barbarian AI is interesting so far. The early game barbs (mostly from lairs) cruise in with the AI_ATTACK routine and hang out. They attack units in the field freely. Once they reach critical mass in numbers, they assault your capital.

The occasional regular barbarians will spawn (not from lairs) that are using the AI_CITY_ATTACK. They will hit your palisades in small numbers. Easy kills. Other regular barbarians will spawn, later on, that will actually use the AI_PILLAGE routines. They suck. Very tough to deal with if 8-10 of their fellows are still in the neighborhood.

All in all, an interesting challenge. If the AI can be taught how to survive the molestation, I'm all for keeping the barbarians this way. The early game stages of FFH should be a terrifying fight for survival.

 

[Deon]

The AI can, in fact, still build catapults. I saw the Luchuruip with a stack at least 20 of them. They also had a lot of Fireball Golems, though they didn't understand sacrificial mechanics. They would summon 10-15 fireballs, then let them die the next turn because my stack was too big.

The barbarian AI is much, much more annoying, but it's not too hard to put up with. Then again, I only have to deal with skeletons and orcs, so my warriors could take care of them. I can see were lizardmen can be problematic.

I think this is because Luchuirp build a small amount of scouts/recon units. I saw Alexis building a huge stack of catapults, but later when I killed them and she trained some hunters I haven't seen a single catapult.

 

[Jenaelha]

One problem I had with the old AI was them sending wave after wave against Archeon (Red Dragon) without protection against his fear. In my last (41H), the Illians sent a couple of large stacks against hiim with at first seemed no protection, but in world biulder I spoted an assassin with Aeon's (spelling?) promotion (no fear) which I suppose gave the AI Illians some hope?
So in general, I've only noticed this one instance of the AI going after him, and it may have been legitimate.

 

[Morlark]

I agree with what Gigaz said about the game seeming too peaceful. In my current game, I've got Decius of the Bannor with two huge stacks of axemen traipsing up and down his territory. Even though he is worst enemies with Flauros, and the Calabim city (yes, singular) is very poorly defended, Decius still hasn't declared war on him.

(Of course it's entirely possible that I'm just misinterpreting the sitation. Maybe Decius was initially intending to use those axemen against me, but then he converted to RoK (which I'd founded) and now we're bestest buddies. Even if that had been the case though, you'd think that he'd put those axemen to use against Flauros, now that he has them.)

The AI also doesn't really seem to know what to do about Barbarians. More than once I've seen a lone barbarian warrior saunter into Decius's territory, pillaging as it goes, and it's not until a good while later that he splats it with one of his axemen stacks (which was within striking distance of the barbarian the whole time).

 

[mahazel]

Something is wrong with AI religious weight and religions spread speed- in my current game as Bannor someone found OO, in 3 turn Hanna, Einion, Flauros and Berri turned to OO. In next turn Necronomicon was created and in another turn OO appeared in 3 from 4 of my cities. I went for Honor, founded Empyrean and created Dies Dei (sp?), few turn later Flauros converted to Empyrean too. I checked in WB - he didn't have Honor tech, but Empy was in 5 of his 6 cities. Also Einion had OO in only one of his cities, and RoK in 3 or 4, but he didn't convert to RoK.

 

[Bill Bisco]

There's nothing wrong with the AI Religious weights. I don't know about spread.

 

[Turinturambar]

I just went through the AIchooseProduction function in cvgameutils.py, which handles AI production now in almost all cases if I understand the code correctly. Sometimes it appears that this function gets bypassed. It would be nice if anyone could explain why this happens.

There are some very huge flaws in that function, most notably:

A lot of stuff is missing, so the AI will never produce the following unless in the very rare cases that the function gets bypassed:
1)All rituals, so the AI will only extremely rarely build the Pact of the Nilhorn or any of the Illian related Auric progression path.

2) Several lategame units: All nationals (Knights, Beastmasters,Crossbowmen), Muskets, Horse archers.
This means the AI will build the lower tier units and upgrade them. Since they get a huge discount on upgrades at all levels (75% on noble, i. e. the AI pays only 1/4th of what a human player pays) this is kind of an exploity strategy.

3) All ships apart from workboats.

4) A few world wonders: Bone palace, the pillar of chains has a value of 0(meaning it never gets built), which is bizarre as it's one of the better wonders.

There is also some code at the beginning that aims to prevent the Sheaim from building Pyre zombies as primary city defenders, because they don't function as military police. This check should be modified to also include warriors if the city can build any other kind of troop, as the AI will simply upgrade all warriors to Pyre zombies anyway.

 

[Y T]

...

All in all, an interesting challenge. If the AI can be taught how to survive the molestation, I'm all for keeping the barbarians this way. The early game stages of FFH should be a terrifying fight for survival.

I think theres an (easy) solution to help the civ AI out. Instead of changing/tuning the barb AI, let the civ AI learn to fogbust!

In my current game Im helping a fellow good civ (Beeri) far away on the other side/edge of the map by positioning a hawk in his only city for reconflights, otherwise he'll be overrun pretty soon.

 

[Monkeyfinger]

2) Several lategame units: All nationals (Knights, Beastmasters,Crossbowmen), Muskets, Horse archers.
This means the AI will build the lower tier units and upgrade them. Since they get a huge discount on upgrades at all levels (75% on noble, i. e. the AI pays only 1/4th of what a human player pays) this is kind of an exploity strategy.

The real answer to this problem is to remove those upgrade bonuses. That strategy you described is the smart thing to do even with normal upgrade costs, so having the AI direct build nationals would hamstring it, but the reduced upgrade cost makes this behavior exploity, as you said.

The old AI didn't use tricks like this and used upgrading purely to prevent old troops from going obsolete, so it was a forgivable crutch, but now that they've been taught some strategic upgrading that bonus can go.

 

[Turinturambar]

I'm not aware of the AI getting any kind of strategic upgrading logic. Afaik it still upgrades everything and even saves the cash for it. I could easily be wrong though, as I haven't seen the source code yet.

 

[Monkeyfinger]

I'm just talking about the act of building cheap units and instantly upgrading them.

 

[Turinturambar]

Ah but that's clearly not intended as there is specific code in that function that ensures that the AI prefers to build champions to axemen, axemen to warriors, longbowmen to archers etc. So these missing units seem to be an oversight.

 

[Monkeyfinger]

Really? I remember discussion of having the AI use warrior upgrades as a strategy in the WM thread. Wonder who decided not to use that, Sephi or Kael?

 

[Onionsoilder]

4) A few world wonders: Bone palace, the pillar of chains has a value of 0(meaning it never gets built), which is bizarre as it's one of the better wonders.

I was wondering about that. I noticed the Bone Palace still hadn't been built long after I researched Guilds. I never started production because I always assumed the AI was going to finish it next turn.

 

[kumquatelvis]

The better AI shouldn't really apply to all units. Skeletons for example; they should just beeline the nearest living non-barbarian unit and try to kill it.

 

[Bill Bisco]

The better AI shouldn't really apply to all units. Skeletons for example; they should just beeline the nearest living non-barbarian unit and try to kill it.

Skeletons want to win.