AI

[Seravy]

star fires in white

Okay, I actually forgot about this, as swordsmen are usually not fantastic. Maybe sleeping and replying the next morning would be more effective than doing it in the middle of the night
The example was Paladins specifically, I assumed a white wizard.

I'm actually not sure what the probability of an AI having an answer to flight/invisibility is nowadays, there has been so many AI upgrades. And while those common nukes exist, they have a pretty low casting priority compared to something like invulnerability....wait, having a single unit on the human side raises that considerably...hey cool, I already fixed this part of the problem.
Invisibility though needs very specific counters. True Sight, Summon Zombie, and...I think that's all in the common/uncommons. Plenty of rares or higher but not very likely the AI has those and they are expensive.
Or there is Magic Immunity, if you have that on a fast enough unit, it works vs any wizard. Expensive but reusable so overall cheaper than losing a swordsmen every battle.
Attacking a city that has walls and only melee units also works as long as the wizard has no direct damage : at the very least Life wizards are vulnerable to that if a normal unit is the attacker.
I suppose it's harder to manipulate the AI this way than it used to be but I'm afraid it's still very much possible. I just never tried because, well, they can't run out of mana before the capital anymore and that's what I usually did it for. So I don't any have recent experience on the problem. In fact with all the AI units swarming everywhere, I'm usually too busy to think about such finer details, I'm happy if I can somehow stay alive against 6-8 stacks incoming per turn. Anyway, I'm sure it WAS a widespread, easy to abuse problem. If it's not anymore, that's very good news, but I'd need to actually try it first.

The exploit allowed me to turn the game around, but felt cheap, hollow, and unsatisfying.

 
While I of course prefer to fix every problem if I can, you always have to option to not use the strategy if you don't like it. (which is one reason why this is fortunately a secondary problem. Similiar to how you were able to summon in unreachable areas and win that way. For a very long time that existed but I didn't use it and still managed to have fun. Eventually I found a way to fix it though.)

there are 5 units it can target.

I think we are missing a few, there are more summons in the game.
Let's see...
Paladins - these can be targeted and it's worth it.
Demons - These are fortunately immune, we can ignore them.
Phantom Warriors - can be targeted and probably should not.
Phantom Beast - can be targeted and most likely should.

why you could easily say "exclude single summons from the WoD decision" but not "exclude single single summons, except for AE and EE, from the WoD decision."

They are about the same but the latter is obviously more space, by about, 6 bytes per unit type? And an extra 30 to access the appropriate data field.
You want "exclude single summons and at least one immune unit" though. Fortunately that isn't all that much worse either.

the bad news is, the existing WoD decision code is like, 120 bytes in a 126 byte space. Not exactly easy to add anything to it, a large enough space has to be made somewhere else somehow. And I don't have any, I already used it all so it has to be new. New, 200+ bytes of space is not that easy to get, especially as I already exhausted most options for other combat improvements. I might be overestimating the difficulty but I think it won't be as easy as it used to be.

Still, excluding all 5 is still a vastly superior solution than excluding none of them, because "being slightly less likely to win some battles" is a better outcome then "quickly become a completely helpless lame duck both in that battle as well as all future battles and then getting banished."

Did you actually test what happens if the AI does not have Wave of Despair? Does it spend less? I mean, a Reaper Slash or Syphon Life is also expensive and unlike WoD, those two won't even kill the centaur in one shot. Unless the AI is highly likely to use Black Sleep, disabling WoD won't help any, the skill will be wasted on the other nukes.
If you still have some saves from that game, use the tweaker to take away their WoD and see how they spend instead. If it does help, it might be worth considering... but it isn't easy to do and might be a bit excessive for such a specific problem (it needs a mono-color death wizard with no annihilate while the player can produce undead swordsme...wait a second, how do you even make those? You can only make living swordsmen in cities and blood lust is not cheap. I guess zombie mastery? Anyway, I think this is an extremely specific and rare situation, am I wrong? If it was zombie mastery I have to say I never lost a game when I had that spell active. Even without the mana milking, the zombies just eat up the enemy resources in the nonstop battles and keep coming back for free. It's an amazingly powerful spell despite looking like not much.)

[Nelphine]

If it matters, I make undead swordsmen by sending in very strong life steal units, and then, regardless of what they manage to raise, sending them off to fight little tiny battles like that, for just this reason. My first impossible death game I won, was won with 1 demon lord, who swamped the enemy with its own units. (I ended up with 1 more demon lord and several death knights, just to guard towers, but only the original demon lord ever went on offense - but he got absolutely HUGE amounts of raised units to support him. Like, probably more than 60 stacks if I'd ever combined them.)

Which is really the biggest problem with this 'exploit' - you conquer something, get more than a full stack due to raised units, send out all the raised units to steal mana, then use your main life steal guy (or zombie mastery) to conquer the strongest thing you can. Repeat.

Now I did this before WoD existed, but given how well the strategy worked then, I can only imagine how much worse it would be if my opponent had been using WoD.

[GermanJoey]

I think we are missing a few, there are more summons in the game. 
Let's see...
Paladins - these can be targeted and it's worth it.
Demons - These are fortunately immune, we can ignore them.
Phantom Warriors - can be targeted and probably should not.
Phantom Beast - can be targeted and most likely should.

I thought Paladins, Phantom Warriors, and Phantom Beasts are all Death-Immune? I don't have a game up where I'm able to check, but the docs at least say those all are that seems to match my memory.

And yeah, typically, as a player, most of the game is about struggling to stay alive under constant onslaught of vast hordes of AI units. That's a big reason why I don't think its very easy to mana milk in regular scenarios, even if theoretically possible - if the player is able to do this sorta crap and get away with it, why not just go ahead and win instead of goofing around?

why you could easily say "exclude single summons from the WoD decision" but not "exclude single single summons, except for AE and EE, from the WoD decision."

They are about the same but the latter is obviously more space, by about, 6 bytes per unit type? And an extra 30 to access the appropriate data field.
You want "exclude single summons and at least one immune unit" though. Fortunately that isn't all that much worse either. 

the bad news is, the existing WoD decision code is like, 120 bytes in a 126 byte space. Not exactly easy to add anything to it, a large enough space has to be made somewhere else somehow. And I don't have any, I already used it all so it has to be new. New, 200+ bytes of space is not that easy to get, especially as I already exhausted most options for other combat improvements. I might be overestimating the difficulty but I think it won't be as easy as it used to be.

It sounds like it's very difficult for you to mod the decision process for WoD. I guess that type of solution is impossible.

Still, excluding all 5 is still a vastly superior solution than excluding none of them, because "being slightly less likely to win some battles" is a better outcome then "quickly become a completely helpless lame duck both in that battle as well as all future battles and then getting banished."

Did you actually test what happens if the AI does not have Wave of Despair? Does it spend less? I mean, a Reaper Slash or Syphon Life is also expensive and unlike WoD, those two won't even kill the centaur in one shot. Unless the AI is highly likely to use Black Sleep, disabling WoD won't help any, the skill will be wasted on the other nukes.

If you still have some saves from that game, use the tweaker to take away their WoD and see how they spend instead. If it does help, it might be worth considering... but it isn't easy to do and might be a bit excessive for such a specific problem (it needs a mono-color death wizard with no annihilate while the player can produce undead swordsme...wait a second, how do you even make those? You can only make living swordsmen in cities and blood lust is not cheap. I guess zombie mastery? Anyway, I think this is an extremely specific and rare situation, am I wrong? If it was zombie mastery I have to say I never lost a game when I had that spell active. Even without the mana milking, the zombies just eat up the enemy resources in the nonstop battles and keep coming back for free. It's an amazingly powerful spell despite looking like not much.)

Well, yes, of course. I mean, they didn't have Wave of Despair for the previous section of the game, and since Green is probably the realm I use the most I've played Green vs Black many times. Typically, pre-WoD AI will respond to a centaur with Black Sleep and then gib it with the next attack by a ghoul or another ranged unit. If Black Sleep fails that time, they'll get it with the next. Trading mana and skill one-for-one with the AI like this is typically a losing proposition, so typically against death I find that it's better to try to either focus on the non-undead units, or buy time with RE, E2M and/or Web, saving a bit of mana for a few centaurs once the AI has run outta steam. Nature is kinda bad against black in-battle. I'm not sure if I've seen them use Reaper Slash or Syphon Life on a Summoned Centaur - I was giving those as examples to say that the AI would still be allowed to cast strong spells on these centaurs even if WoD was not allowed.

Undead swordsmen (or undead whatevers) are very easy to make via Drain Life, Syphon Life, and zombies. Between your wizard and your heroes, you can usually get a full stack of undead junk when taking out a low-resistance town. The first are often better than free against low-resistance enemies (e.g. barbs, lizards, etc) because the drained hp will convert into skill points at a high rate. Anyways, I had big piles of undead left over from my last conquest and was just using them to try to whittle down the invading AI stacks. I also used some blood-lusted (beastmen-built) centaurs but couldn't spare the casting skill to make very many.

[Seravy]

I thought Paladins, Phantom Warriors, and Phantom Beasts are all Death-Immune?

They are but WoD is blocked by Cold Immunity and Magic Immunity only.

[GermanJoey]

Ah, right... my mistake, sorry. Well, that was the last idea that I had...

[Nelphine]

Wish the AI could do some sort of efficiency check. If x is 80% likely to win, but costs 50% of the mana, and y is 100% on both, do x. However that's I think exactly what is truly missing. Humans can make that efficiency check, and also know when efficiency doesn't matter.

[Seravy]

Wish the AI could do some sort of efficiency check. If x is 80% likely to win, but costs 50% of the mana, and y is 100% on both, do x. However that's I think exactly what is truly missing. Humans can make that efficiency check, and also know when efficiency doesn't matter.

Well yeah if we could solve "chance to win =?" we'd have a perfect AI. Just look for whichever option says 100% and do that

Chess AI kinda works like this (the simplest ones at least) except they only look ahead like 5-7 moves as that's about as far as you can get in a reasonable time. In MoM though you have like, a million different choices per turn so it'd be able to compute, like, less than a single turn.

[Seravy]

I think the AI might be misusing alchemy.

When casting the spell of return :
Convert all gold into mana. No checks if it's necessary or not.
Not optimal but fine, better safe than sorry on the mana upkeep of spells. Besides the wizard is already losing anyway.

When they are not alchemist :


If gold>24000, or mana<gold/16, then convert half of the gold into MP
Afterwards, 10% chance to do the following :
if Mana<gold/4, convert 50% of gold to mana otherwise
if Gold<mana/8, convert 50% of mana to gold.

these are fine.


When they are alchemist :
If turn<50, set a ratio of 70% gold, 30% mana.
If turn>=50, set a ratio of 50% gold, 50% mana.


This looks fine at first sight, right?
It's not!
The AI will rush buy stuff when they have unspent gold. How much makes them buy varies, but it's safe to say if they have over 3k gold they spend like there is no tomorrow because gold sitting around is a wasted resource.
On the other hand, if they have less than 32* their casting skill in mana, they will produce MP from all power they have.
Which means this, they aim to have a lot of mana (say, 200 casting skill wants 6.4k mana.) . In order to reach it, due to the 50% alchemy ratio they also need that much gold. But if they have that much gold, they'll spend it to rush buy stuff. Ultimately, this means an alchemist wizard will neglect casting skill and research, draining most of their power base into buying buildings and units. Which is... a questionable strategy. If the massive onslaught of units does manage to conquer things, it might even be good...but if not, they potentially miss out on spell of mastery chance and enough casting skill to compete.

Currently, the AI "BUY" decision is this :
If Sawmill, always buy.
If Wizard's Guild, always buy.
If Nightmares, always buy (why? Those ain't even that good...)
If gold>20*cost of thing to buy, always buy.

so 4k gold will buy 200 cost units, etc.
Not sure if the alchemy, or the buying, or both needs to be changed but one of them most likely does.


EDIT :
I came up with these new rules -
-AI now prefers to buy Sage's Guild, Alchemist's Guild, Fighter's Guild, Amplifying Tower, Wizards's Guild, if they can pay 2x their buying cost, Sawmill always, everything else if they can afford 50x the buy cost OR have over 10k gold.

-AI with the Alchemist retort now prefers to maintain a 20% gold 80% mana ratio instead of 50-50 in mid and late game. This should allow them to allocate more on casting skill and research, and only the minimum needed on mana production. If the 80% would exceed 24000 mana, the ratio is adjusted to make sure no resource is wasted by the 30k caps.

[Tlaloc]

Does the AI know where the player's capital is without having discovered it?

[Seravy]

Does the AI know where the player's capital is without having discovered it?

Spells don't, the AI has to find the towns first to target them or tiles belonging to it by curses. Units know the entire map however. Spells that target units (Fire Storm etc) are exceptions and see the entire map.
The AI only stores which cities are scouted, and stores no scouting information on the map itself, so anything that involves units can use the whole map.