AI

[Nelphine]

When an AI unit in tactical combat has determined it is too weak to run away, it runs away intelligently, even if all of my units are invisible. ( A giant spider with 1 figure left, I have 1 night stalker; the giant spider runs away based on where the night stalker is.)

I think the 'run randomly due to invisible units' should take priority over this. OR run to a random corner. But it shouldn't run away based on where my invisible unit is.

Also, units in lairs always heal to full. I have a feeling this is on purpose, but it feels annoying given that that is what regeneration is supposed to do, but instead everything gets it.

[Seravy]

I think the 'run randomly due to invisible units' should take priority over this. OR run to a random corner. But it shouldn't run away based on where my invisible unit is.

Armies typically contain mixed units so this is a rare situation.
I mostly didn't add it because honestly I'm happy the base feature is working, it war pretty hard to do. Space is finite too, and invisibility checks would need to be added to at least 2 or 3 places AND they would also need to check if the AI has illusion immunity available.

On top of all that, it's questionable whether the AI should or should not know. When I run around against invisible units as the human player, I always know the exact location of said units. (yes it's somewhat of an exploit to check the movement cursor as it shows a blocked tile where invisible units are, but it is part of the game.)
Even if I did not, I could make a pretty good estimation based on where they turned invisible and which way they need to move to reach my unit. (or if I have no idea I can still avoid them 90% of the time by merely moving roughly at the same speed in circles as those units can follow, preferably ~1 faster in case of terrain modifiers)

Seeing the units is not a good solution but I think moving randomly would be worse because it's common sense not to move in the direction you last saw the enemy at the very least. (also if the AI used any AOE spells they saw the location of units, and this again the game cannot simulate. Flame Strike, Call Lightning, etc all tell you where units are and the AI has a high priority to use these if enemy units are invisible - to hit them)

Also, units in lairs always heal to full. I have a feeling this is on purpose, but it feels annoying given that that is what regeneration is supposed to do, but instead everything gets it.

Units in lair don't exist unless you attack the lair so they don't have a health bar outside combat. Storing 900 lair defenders with the unit cap of 2k (original 1k) is unrealistic and aside from health, serves no purpose.

[GermanJoey]

Rjak's Lizardmen capital is here: (see minimap)




He's been sending a nonstop stream of settlers here: (reference the minimap)




Not a terrible spot with Coal and Mithril, but a lotta coast and tundra too. Been pretty nice for me, as my waterwalking Beastmaster hero has an item (Pixies' Favor, +3 Res) that makes him immune to Black Sleep. I've basically been farming settlers for big XP every other turn. However, why does Rjak repeatedly try for this spot over and over instead of:




I was completely flabbergasted when I found this area completely empty despite it basically being adjacent to Rjak. You can even see in that last screenshot that he decided to walk two of his settlers straight through the area, like a boss. These screenshots were taken in an Extreme game, mid 1406. It actually took me a long time before I uncovered this area myself, as my early spirits got sniped and then found myself too busy casting other stuff to conjure more. And then Ariel broke through that tower you can see in the lower left of the 2nd screen shot in freaking Nov 1404 with a stack of Adamantium War Trolls and Halberdiers. Rjak did indeed eventually settle the area; after I finally beat Ariel back, I took him out around 1412 and at that time found that the whole sub-continent was filled with mid-sized towns. But why wouldn't he settle here with one of his very first cities instead of sending nonstop streams of settlers to me?

[Seravy]

The place on the first picture has 2 gold ores, 1 coal, 1 mithril, from the looks of it 25 max population...WAIT!
The spot you're pointing at is too close to your city and it's not possible to build there. Nor on the tile 1 to the north, west or northwest. Which leaves really crappy tundra tiles the AI will never target before the good spots.

This leaves three explanations
-When the settler started moving, they went somewhere else nearby and found it is no longer available. Then they decided to go for the crappy place because it was about 3 times closer than the good one when they had to reroute. (unlikely)
-They are trying to enter the tower visible there instead but you kill them before they reach it (much more likely, especially if the wizard is expansionist, or another type with high preference to colonize the other plane)
-The place at the second screenshot was already colonized either by him or someone else, but the city got razed before you found the place (most likely by monsters, if yes you should have found a ruin there containing a low amount of gold stolen from the wizard and nothing else)


I also see another tower open right next to their capital which, if opened before the one next to you, might explain where the rest of the settlers disappeared to.

If you can upload a save game for the first few turns of this game (unless you started new games, save1001.game should be your first turn, 1002 the second etc) I can watch the AI play and see what happened.

[GermanJoey]

Already started a new one. I have a save for you from that one, though.

The rightmost city in that screenshot (the one with the mithril etc) was neutral, by the way, so it woulda been there from T0.

[Nelphine]

That's a reasonable explanation behind running away from invisible units. Thanks.

[GermanJoey]

Why does the AI build granaries at so low priority? I very rarely see them in conquered cities, but they should be one of the first buildings built. (after sawmill, ofc)

[Nelphine]

I believe its because they have large bonuses to things like production, money, power, etc, so population isn't as important (and i think they also have bonuses to population). If you already build things in one turn, you don't need more workers. If you never run out of money, you don't need more taxes. Etc.

[GermanJoey]

Oh, I guess that makes sense!

[GermanJoey]

Continued from here:

http://www.realmsbeyond.net/forums/showt...#pid618484

You have a very good point... but then, a wizard (player or AI) only ever has so much mana to spend. Even an Impossible AI's mana is not infinite. For example, the whole reason I'm trying to drain this sucker's mana in these otherwise pointless fights is so that he won't have much left to use when I send in my real armies.

The other thing about Wave of Despair that makes it especially vulnerable to "mana milking" is that there's no other spell that has its combination of:

1.) Very expensive per-cast
2.) Affects all targets, and is re-castable
3.) Devastating amounts of damage against some targets, but deals no damage at all against others
4.) The immune targets are easily producible, and cheaply
5.) The wizard has no other spell that is easily able to kill the immune target

(actually, now that I think about it, Massacre also qualifies, and the wizard acquiring Annihilate stops this situation from becoming a problem)

The closest other spell I can think of is Flame Strike. However, a fire-immune Fire Giant is much more expensive to produce than a death-immune Skeleton; between wasting overland skill and the extra mana it costs to summon the FG, I'd be hurting myself as much as the enemy trying to milk them like that. Furthermore, a Chaos wizard also has numerous other ways to kill the Fire Giant on the first turn of battle (e.g. Lit Bolt, Doom Bolt, Warp Lit, Disintegrate) as well as other ways to kill the centaur (fireball, or any of the above), so it is extremely unlikely the AI would actually want to cast FS first.

That said, I can't think of an easy solution of how to differentiate between when to cast WoD or not, given what you said in your prior post.

1., doesn't matter. Unless the AI can't use up all its skill in 25 turn otherwise which is...extremely unlikely. a Cheap 20 mana spell times 25 already uses 500 skill.
2., Targets also doesn't matter, recastable does matter though. However, buffs can be used as many times as the AI has units per spell...and if there aren't many units, summons can be used many times instead and then those can be buffed anyway.
3., deals no damage is the key, as it allows you to force the AI to stay in combat and keep casting. Buffs can also achieve this and you don't even need an immune unit. Life wizards are the easiest targets.
4., this matters, in fact you need a unit immune to everything the AI can use against it for the strategy to work. So if they have too many different nukes, it fails. Undead units are immune to a whole lot of things though.
5., exactly.

IMHO, its really the combination of these 5 properties of WoD that make it possible to milk the AI through it.

1.) I think its high cost does matter here. There's three issues related to its casting cost:

a.) It uses skill too fast. For example, lets say both the player and the AI want to win the battle at all costs, and both have 200 casting skill. The player casts Summon Centaur 4x for 12 skill per cast, while the AI casts Wave of Despair 4x for 50 per cast. By turn 5, the AI will be completely spent, while the player has 152 skill left remaining. At that point, the player can just do whatever - keep casting centaurs, throw nukes, etc.

b.) The fact that it only takes 4 turns to drain the AI of, in the above example, 600 mana (remember the "milking" strategy wants the player is at 1X distance and the AI is at 3X), means that the player doesn't have to put any effort into surviving for 25 turns. In fact, if my undead swordsman actually even survives for 4 turns against the ghoul, using the first example I gave, I'd be completely thrilled and try to retreat it to use another time.

c.) The "milking" strategy doesn't actually bring in any income for the player; the player is basically hurting themselves to hurt the AI far worse. 150/12 = 14.5 ratio. However, once you figure that the AI essentially gets a 3X resource bonus, and the player is also throwing away a cheap unit, the average "milking ratio" might only be about ~4x or so. A situational change as simple as the player also being at 3X drops the ratio down to barely above even, rendering the strategy ineffective as

2.) Agreed, the recastableness is what makes it explotable. But, that's why I think targets matter. Thinking about it some more, there's really only 2 situations we'd wanna stop, and only at high-distance (at least 2X cost for the AI):

a.) a player sending in stack of spearmen one-at-a-time (B-movie ninja style) at an AI melee unit, to induce the AI to cast WoD 1x9 battles. I think spearmen are the only problem unit for this scenerio, because even swordsmen are expensive enough that the effort+risk of trying to milk with them probably isnt worth it.

b.) the player summons cheap units in battle, to cause the AI to cast WoD repeatedly. In this case, the only problematic units are Centaurs and Fire Elementals. Earth Elementals are also not death immune, but they're expensive and thus not something the player will be able to milk with.

So, I think limiting WoD to not be cast when all of these conditions hold:

a.) the battle is of average difficulty
b.) the distance is at least 2x
c.) the target is a summoned centaur, a summoned fire elemental, or a non-undead spearmen

would something like that be possible? Note that this only affects WoD (and probably Massacre, and I guess Wrack as well) - there's nothing that says they can't use Reaper Slash or Black Sleep or whatever on these lone threats. Not quite as much of a sure-kill as WoD, but still pretty close to it.

3.) As far as the general issue is concerned, of the AIs not really knowing how to ration its mana out, well, I don't know what to do about it. But, that's why the AI gets big bonuses at high difficulty levels, as a way to still be able to pressure a much more efficient player. I agree that Life tends to be the worst with this kinda stuff, but then they get even more bonuses than others with all their city enchantments so its not too bad of a problem I guess.

4.) Yup. That's why I think Wod, Massacre, and Wrack are the only problematic spells.

5.) Yes!

We can conclude : This is a more generic problem as any wizard that does not have direct damage spells is affected. If they have some and your units are immune that just results in the former scenario which is the root of the problem.

The mana saving tactic they have based on remaining mana crystals should be enough to at least avoid losing their capitals, but unfortunately that's as far as it goes unless you have a really good formula that tells how important a specific encounter is and how much mana should be spent on it. Right now the AI can only tell apart three cases, "Capital","Important" and "Average" battles. Also, the proposed formula should be compatible with the overall overland unit movement strategy AIs use, in other words it cannot result in the AI losing the majority of their forces just because the human player attacks them one unit a time while they are on their way to group up. There is no way around that, units will go 1-2 at a time towards assembly points.
Actually I think there are some serious implications based on how the game works.

Assuming the territory is invaded by enemy stacks - visible or not doesn't matter - that can strike at any nearby city or unit passing through.
New units are produced in cities, and a stack is limited to 9 units. The way the game works, a stack containing fewer than 9 units is weaker. When a new unit is produced, there are 10 units at the city, which has to be split to two groups. It's pretty obviously impossible to make two 9 stacks out of 10 units, so either one, or both stacks has to be quite suboptimal.
As we have proven the game mechanics force players, human and AI alike, to move their units in suboptimal stacks -most often leaving 6-9 in the garrison and sending 1-4 towards an intended destination - this means the majority of units will have to be moved in vulnerable stacks at the very least after they have been produced.
If the AI tries to defend their vulnerable stacks because it deems them important, it can be mana milked but the units will be safe.
If the AI abandons the units as unimportant battles, it'll not lose mana but it won't ever be able to form new armies anymore, as every unit will get killed before they can join up with something else to create an army that's big enough.
We can conclude that this is a core part of the game and it's unavaoidable : having enemy presence in your lands means you either need to waste lots of mana fighting them back (yes, even as the human player) or giving up the ability to form armies entirely, strictly keeping your units in your city garrisons of 9.
And btw this shows how the problem can be avoided by the AI. If there is NO enemy presence, the problem doesn't exist. The AI needs to wipe out every unit in their territory...to do that they need to fight them...and spend the MP on those battles to win. So...there is no way out. This is an unforgiving game system, you either spend resources to get rid of enemies, or you lose control over your own territory and units entering it just go there to die.

I think we should continue this in the AI thread though.

Well, yeah, that's the whole crux of the game, and what makes it intersting. The AIs are vastly bigger and more powerful than the player, and the player essentially has to figure out how to efficiently use their relatively meager resources to design a strategy that will win.  Part of that is clearing out threats, and part of that is sometimes using that 9-stack garrison to attack out, because that's the only thing strong enough to defeat the enemy army. The idea of the enemy army camping outside of a city and picking off units one at a time forever seems a bit silly to me - if this army is so strong, then why don't they just attack the city and win? (barring edge cases of strong city defensive enchantments, like Flying Fortress or Spell Ward).

Here's an idea - for these unimportant battles - and only in unimportant battles - would it be possible to give the AI a mana budget based on the roll of some dice? So, for example, sometimes they'll be able to use the full brunt of their casting prowess in an unimportant battle, but sometimes they'll limit themselves to only 20% skill. This would at least make mana milking far less reliable a strategy for the player. The AI, at higher difficulty levels, typically vastly outnumbers the human player in terms of units, and thus the human only has so much fodder to throw at them.