Community Feedback Needed: Biggest AI issues

[KingOfLands]

AI needs to be more realistic in its early game peace treaty acceptance - when it's lost units at 5:1 ratio and failed to capture any cities it should not still be trying to demand a city for peace, in fact it should realise the war is not in its own best interests and offer a ceasefire. I believe this problem is caused by the early game AI power rating being artificially inflated thanks to its starting bonuses.

AI should not be declaring war on player for failing to adopt its favourite civic... of pacifism.

The AI severely underrates spellcasting units for power rankings among other things. Had a jolly time having absurd wars declared on me in my last game due to Auric / Keelyn / Rhoanna's firm belief that my gigantic stack of air / fire wizards and archmages was not actually that dangerous.

 

[feydras]

AI in .32e seems to be much better at founding religions. First game i've played in a long time that the AI's scooped me to all religions. I wasn't really gunning fast for them but i was only playing on Monarch. Nice change.

Minor AI inbalance: if Loki is adjacent he can keep an enemy stack chasing him (attacking him every turn). This seems to distract them from any other targets nearby. Counterpoint, roleplaywise this actually fits nicely but a Human player would never do it.

 

[GePap]

The AI needs to understand their World Spells a little bit better. The most glaring example I saw recently was for the Elohim. Fellow AI Doviello decided to go on the warpath, with Lucius and several beastmen and axemen. At that point the Elohim had three cities. Soo enough the Doviello took two of them, and when the Elohim had one single City left, they had the misfortune of being next to Orthus. That being said, the Elohim never used their sanctuary spell, which is designed exactly for this kind of situation. Most human players would have al least survived for 30 more turns in that case.

As for AI problems, soon after that the Doviello also overan Bannor, which neighbored the Elohim, taking their two or three cities. And that is where Doviello ended, because they never came after me (having a vast jungle in between helped, but they could have still done some damage). More importantly, after this start, Doviello became an economic backwater, and they did not even create any more cities until long after I had come to utterly dominate the continent. That to me was a glaring AI failure.

 

[smjjames]

@Feydras, yea, that is annoying. I once had a lizardman scout chase after me repeatedly. This was while exploring Clan lands, as a result, I had to duck into nearby Calabim territory to get it to stop chasing Loki.

 

[Pyr0mancer]

The AI needs to understand their World Spells a little bit better. The most glaring example I saw recently was for the Elohim. Fellow AI Doviello decided to go on the warpath, with Lucius and several beastmen and axemen. At that point the Elohim had three cities. Soo enough the Doviello took two of them, and when the Elohim had one single City left, they had the misfortune of being next to Orthus. That being said, the Elohim never used their sanctuary spell, which is designed exactly for this kind of situation. Most human players would have al least survived for 30 more turns in that case.

IIRC, Kael stated that the Elohim AI is specifically written to use Sanctuary when 2 or more civs are simultaneously at war with them.

As for AI problems, soon after that the Doviello also overan Bannor, which neighbored the Elohim, taking their two or three cities. And that is where Doviello ended, because they never came after me (having a vast jungle in between helped, but they could have still done some damage). More importantly, after this start, Doviello became an economic backwater, and they did not even create any more cities until long after I had come to utterly dominate the continent. That to me was a glaring AI failure.

Doviello tend to be an economic backwater anyway due to the research penalty, and their conquest-oriented early strategy backfires horribly if they can't keep up the momentum and keep their economy propped up with pillaged gold. 8-9 cities that early is a very good way to cripple an economy with maintenance costs.

@smjjames: Why not just charm it?

 

[GePap]

IIRC, Kael stated that the Elohim AI is specifically written to use Sanctuary when 2 or more civs are simultaneously at war with them.


Doviello tend to be an economic backwater anyway due to the research penalty, and their conquest-oriented early strategy backfires horribly if they can't keep up the momentum and keep their economy propped up with pillaged gold. 8-9 cities that early is a very good way to cripple an economy with maintenance costs.

On the first point I think Kael and the group should think about changing that. At the minimum, if the Elohim are down to one or two cities and still at war with any number of civs, they should most certainly use that spell, to give them time to build defense or manuever diplomatically. I mean, it makes the human's life easier, does it not, if the Elohim don't use their spell if the single human player decides to go after them?

As for the Doviello, it would make sense to rewrite the AI to have them raze cities instead of keeping them on any early blitz. It makes thematic sense and it spares them the economic collapse, as long as they make a few more cities here or there.

 

[Kael]

On the first point I think Kael and the group should think about changing that. At the minimum, if the Elohim are down to one or two cities and still at war with any number of civs, they should most certainly use that spell, to give them time to build defense or manuever diplomatically. I mean, it makes the human's life easier, does it not, if the Elohim don't use their spell if the single human player decides to go after them?

Just keep in mind we need the highest priority issues. With limited time Im most interested in the heavy hitters, these little things are nice (and you aren't wrong) but there are more important issues.

But specific to this point. If we did make it so the elohim use their world spell if they are down to 1 or 2 cities and at war players will begin to just declare war on them early on just to get them to use their spell. In my mind, although it could be better, Im okay with the elohim retaining the spell in 1 on 1 battles. If they cant survive the conflict, then they should probably be leaving the game. A 30 turn pause isnt going to save them. If they do survive then its best they still have the spell at their disposal. If they are dogpiled, thats when they should use the spell and turtle up, against multiple opponents 30 turns gives them time to breath and time for something else to change.

Its not perfect. We could write code that takes all of these conditions into account. I could spend an hour or 2 and get a more intelligent use of the sanctuary spell. Nut is that where we want the time to go?

As for the Doviello, it would make sense to rewrite the AI to have them raze cities instead of keeping them on any early blitz. It makes thematic sense and it spares them the economic collapse, as long as they make a few more cities here or there.

That actually a really easy fix. There is a iRazeCityProb on each leader. The Doviello are set to 10% (like most leaders). You are right that those numbers need to be reviewed because a lot has changed since they were set.

 

[JanusTalaiini]

I haven't read through the entire thread, so I apologize if these have already been mentioned:

I don't think I've ever seen the AI approach me for peace. Even if I'm on the verge of destroying them, unless I contact them and bring up the subject, wars last forever, literally.

The Khazad never have any money. At all. Which, of course, is terrible for their dwarven vault bonus.

 

[Bad Player]

Try giving civs a "preferred religion" like in the Total Realism mod (where Arabians will convert to Islam, Indians to Hinduism, Persians to Zorastrianism etc).
That is, Civ X will ALWAYS convert to Religion Y if they get even a single city with that religion.

Appropriate preferences would probably be:
AV: Sheaim
Order: Bannor
Empyrean: Malakim
FOL: Ljiolsofar
Runes: Any dwarf
OO: Lanun, I guess? Or Balseraphs?
Esus: None, really. Or dark elves, I guess.

Possibly do it by leader (Hannah seems more likely to be OO than Falamar).

The problem with that IMO is that it locks those civs into a certain playstyle. Additionally if you do this absolutely preferred religion, might it be better to do it by leaders not civs (since traits might work better with some religions)?

I guess Svalvartfar could work with FoL (elven workers?) or CoE (hidden nat). Also it might be in their interests to join the most powerful 'gang' even if they lose a bit of productivity by choosing a less than ideal religion.

 

[Bad Player]

The AI severely underrates spellcasting units for power rankings among other things. Had a jolly time having absurd wars declared on me in my last game due to Auric / Keelyn / Rhoanna's firm belief that my gigantic stack of air / fire wizards and archmages was not actually that dangerous.

To make this work for a power rating the problem is that some spheres become really powerful at tiers 2 and 3 but others are good at tiers 1 and 2. Therefore could I suggest that the power rating of magical units be based on what and how many sphere-tiers that player has. E.g. Sun-tier1 = 1 point; Death-tier1 = 2 points; etc.

The only problem I see with this is that players may not promote their units fully and so the AI won't realise how powerful the magic units are.

 

[Bad Player]

Just keep in mind we need the highest priority issues. With limited time Im most interested in the heavy hitters, these little things are nice (and you aren't wrong) but there are more important issues.
...
Its not perfect. We could write code that takes all of these conditions into account. I could spend an hour or 2 and get a more intelligent use of the sanctuary spell. Nut is that where we want the time to go?

Could you outsource these small code changes by listing what you want in a public thread and then if anyone has the time, ability, and inclination, they could attempt to write good code for it. If it's good enough then all you would have to do is copy-paste it into FFH2. If it doesn't get done by the public then nothing is lost.

 

[JanusTalaiini]

The AI does not launch seaborne invasions (I've never seen any, at least). Because of this, if you're on a separate continent or island from your enemy, you're completely safe aside from their ships attacking your ships or pillaging your fishing boats.

For the same reason, recruiting other civs to help in an inter-continental war is nearly useless because they won't actually contribute any combat troops - the best you can hope for is to have them cut off trade with the enemy and send in some naval harassment.

 

[uberfish]

The issue isn't really to do with spells.

On high difficulty levels the AI starts out with several free techs and units (originally balanced for the base civ4 game), however the cost to research the techs and build these units is significantly higher in FFH2, which essentially means the AI has a 70-80 turn head start in tech on normal speed as well as 30-40 turns of production (extra warriors and worker on immortal.) Now it's fine if the AI uses this head start to build economy, what is bad is the AI abusing it to rush lucian/axemen at you on turn 40-50 immortal when it's not even possible to have anything better than warriors out, or to make a quick 2nd city then spam warriors/scouts out of both cities non stop.

in 0.2x you spent the early game fighting off barbarians, in 0.32 it seems the barbarian attacks have largely been displaced by AI rushing and I don't think this is a change for the better.

 

[Kael]

Could you outsource these small code changes by listing what you want in a public thread and then if anyone has the time, ability, and inclination, they could attempt to write good code for it. If it's good enough then all you would have to do is copy-paste it into FFH2. If it doesn't get done by the public then nothing is lost.

Probably not. Its not like pedia entries, the AI is amazingly complex and interwoven with everything else that is happening. Even outside creations that are good on their own would be problematic combined with everything else.

The biggest risk to a total conversion mod is the unfixable crash. That will ruin a project so a lot of what we do is based around risk avoidance. Cutting in SDK code from different programmers would be a huge risk to create an issue down the road within their code, or in the interaction of their code and other pieces that I wouldn't be able to isolate. Which is why I rewrite everything that goes into FfH to try to simplify and put it into a format that is easy for me to understand (because at the end of the day its my job to find and fix bugs).

For example Sto came up with the brilliant mechanic to implment random alignment leaders. It was great work, but I wouldn't cut in Sto's work. Instead I rewrote it into a form that I understood and could work with. Thats not to say my form is better than Sto's (Im sure its not, Sto's a much better programmer than I am), and it does limit the mod by my programming skill. But at the end of the day we probably have the best chance of being able to quickly find and fix bugs.

 

[GePap]

The AI does not launch seaborne invasions (I've never seen any, at least). Because of this, if you're on a separate continent or island from your enemy, you're completely safe aside from their ships attacking your ships or pillaging your fishing boats.

For the same reason, recruiting other civs to help in an inter-continental war is nearly useless because they won't actually contribute any combat troops - the best you can hope for is to have them cut off trade with the enemy and send in some naval harassment.

Back in either .31 it was late in the game, only on Noble, but the Calabim put together a significant overseas invasion force, and for several turns I had to contend with multiple attacks against multiple cities, as I shuttled forces from one counterattack to the other. IN gave me several turns of problems before my superior economy allowed me to win the Naval game and cut off the invasions.

So I don't think this is that serious a problem as some others.

 

[KingOfLands]

I played a few games of .32e last night, and I must say that patch improves the general threat posed by the AI drastically; by the second "Holy crap, it's axe-rushing me with HOW many units?" give-up I decided that I can in fact get away with leaving AI No Building Reqs off now.

And then, naturally, the next game I started I got rushed by Charadon and Sheelba almost simultaneously with an early wave of barbarian orcs and skeletons... as well as Orthus appearing next to my second town while I was waiting for Sheelba's stack of "doom" (I'm not sure a couple orcs and 4 goblins is all that scary under any circumstance) to hit my capital.

Long story short, I'm a much bigger fan of Faeryl now than I used to be; I managed to hold that off and keep expanding over the next several couple hundred turns. My research went in the tank, however, as I was so busy dodging blocked tiles and researching defense enabling stuff that I didn't get around to education for quite a while.

Anyhow, I'm posting this here because I noticed one glaring issue and one minor one. The big one was that Lucian spent several turns wandering around near my borders at all of .9 health before getting killed by a spider; something funny is going on with how he's managed. If Charadon had bothered to heal him I might have been totally boned, as opposed to just slowed up by waves of beastmen and orcs. The minor issue is, as others have pointed out, there is just no quit in Charadon's playbook; despite the fact that I took a barbarian city, one of his, and built a third while making a total mess of everything he sent my way, he refused to make peace for over a hundred turns on Standard. If he'd, say, taken the first offer I'd given him for a treaty and reloaded with Axemen, he'd have done a good deal better; even if he'd just called the war off and gotten back to expanding it would have been an improvement. That's inconsistent with his personality, I suppose, but I'd much rather see him regroup and come back strong than keep feeding my hunters and warriors experience with stacks of 4-6 beastmen and axemen.

Oh, and one other thing: He sure did summon the wolves at a bad moment for me. Whatever you did about his world spell code has given him excellent timing; kudos.

 

[smjjames]

Sort of sounds like he went into 'no surrender' mode.

 

[Bitwise]

Also noticing that the AI rarely makes naval assaults. Isolated island civs remain isolated through the entire game.

 

[Nimbus]

The Khazad never have any money. At all. Which, of course, is terrible for their dwarven vault bonus.

comments of this nature have been made to one degree or another in this thread or another for while now. Could a solution for the AI be to rework the bonuses of the dwarven vaults so that the lowest setting gives no penalties or bonuses and then works more slowly upwards from there?

 

[Maniac]

The Khazad never have any money. At all. Which, of course, is terrible for their dwarven vault bonus.

The money shown in the F4 screen means the money that leader is willing to give away in trade deals, not their full treasury. The Khazad are willing to give away zero percent of their money reserves.

Btw, if I try to download the mod, it stops midway, so can't get the mod.
Edit: Woops, thought this was the bug thread.