My extreme life game

[Nelphine]

True. I've always played that style. I dislike the huge grind wars - I've always felt they favor the AI, so I've never tried to win that way.

I think the overland cast bonus is the single biggest factor in the AI toolkit, with the gold/production being a distant next (although both overland and production also rely on AI not being as worried about maintanence, so that's also a big one).




So what I'm trying to do, is describe the AI in terms of human player decisions; and then when it comes to decisions the AI is bad at, noting what kind of bonus it requires to maintain the same decisions as the human would.

So, assuming we like my definitions for difficulty, I'm using extreme as my basis. I should actually be using hard I think, but whatever - I think Seravy has actually taught the AI to play at an expert level in most aspects of the game, not at an experienced level.

So for extreme, if we imagine there is 1 AI that currently matters, and the human is at war with them, and everything is equal (number of cities, pick selections, spells chosen to be in the spellbook so research will open up the same choices, same minerals and city tiles available, etc, so the only difference is what the chooses to do vs what the human chooses to do), and assuming the player is an expert, then, in theory, there should be a roughly 50% chance of the human winning.

So the AI cheating bonuses should make it possible for the AI to make up for its limited decision making ability, but otherwise shouldn't really get the AI ahead.

Now, correct me if I'm wrong, but I believe we have taught the AI most things, with the major exceptions being: stack creation and ability to react to incoming stacks, as well as the fact that strategic combat is quite different to tactical combat and therefore fighting targets other than human controlled is different?

Assuming those two are the only major exceptions, we have tried to create a strategic combat system that will, in most cases, have results similar to tactical*, to the point that (for AI decision making) we aren't worrying about it - and for stack creation, we have the cheating bonuses so that quantity will make up for quality.

*My current question about ranged bias is being ignored for the purpose of this topic, under the assumption we'll fix it.

[Seravy]

The AI lacks
-the ability to quickly an effectively build stacks
-the ability to select the correct types of units for stacks instead of whichever is more powerful
-the ability to split or merge stacks freely
-the ability to attack with a stack based on their actual chance of success (considering ranged, abilities, etc) instead of strategic value
-the ability to consider their own and enemy spellcasting ability before starting a battle

-the ability to avoid moving stacks near to enemy stacks that could defeat them
-the connection between spellcasting and stack movement to
 -rebuff dispelled stacks
 -summon units specifically for the stack
 -target enemy cities with direct damage spells before attacking (albeit there is a good chance it happens, it's not intentionally done)
-the ability to move a unit/stack out of the way, or replace a stack with another in a city/node/etc
-the ability to guard towers
-the ability to use flying ships on land
-the ability to abandon a city/sell buildings when it cannot be defended
-the ability to pick overland spells for a coherent tactic of any kind instead of completely random
-the ability to make their own diplomacy decisions - they're forced to play by rules instead
-the ability to stop producing/building/casting stuff if maintenance is too much burden, or consider maintenance costs in any way
-the ability to disband ships if having maintenance problems
-the ability to cast Conjure Roads and Nature's Cures
-the ability to escort settlers or engineers to protect them
-the ability to always keep heroes in "safe" stacks where they can't get killed easily.

I think that's about all the things a human player can but the AI cannot do at the moment

[Nelphine]

I'm just going to mention one of those:

the ability to pick overland spells for a coherent tactic of any kind instead of completely random

I think most of the rest of the points boil down to overland stack building (and control, which o didn't mention). This point however is a good one, as you have used a similar technique with toverland spell casting to ensure that random overland casts are used in enough quantity that the actual desired spell (if it were a human) is cast. However, unlike unit production, I think this is a mistake. Specifically, you have taught the AI to use any spell it does cast, in a very effective manner; and you have rebalanced the spells so that (aside from units which have tiers) all spells are worth casting given their cost, throughout the game.

This results in the AI not being able to choose a strategy for its overland spells. In order to simulate a strategy, it throws many spells, such that the spell it wants will be cast.. But I think this backfires, because all of the other spells it casts are NOT wasted - so instead, you have successfully given the AI the ability to actually use multiple complete strategies at the same time, in order to use one.

I think this is the source of a lot of my frustration with the spell priority, and specifically why I decided to have this conversation before tackling the spell priority decision

[Seravy]

I'm just going to mention one of those:

the ability to pick overland spells for a coherent tactic of any kind instead of completely random

I think most of the rest of the points boil down to overland stack building (and control, which o didn't mention). This point however is a good one, as you have used a similar technique with toverland spell casting to ensure that random overland casts are used in enough quantity that the actual desired spell (if it were a human) is cast. However, unlike unit production, I think this is a mistake. Specifically, you have taught the AI to use any spell it does cast, in a very effective manner; and you have rebalanced the spells so that (aside from units which have tiers) all spells are worth casting given their cost, throughout the game.

This results in the AI not being able to choose a strategy for its overland spells. In order to simulate a strategy, it throws many spells, such that the spell it wants will be cast.. But I think this backfires, because all of the other spells it casts are NOT wasted - so instead, you have successfully given the AI the ability to actually use multiple complete strategies at the same time, in order to use one.

I think this is the source of a lot of my frustration with the spell priority, and specifically why I decided to have this conversation before tackling the spell priority decision

There are two types of the spell that AI doesn't use efficiently. Coincidentally these two make up 90% of the repeatable spells in the game which are
-Summons - yes the AI might be using the better summoning spells and might summon to "smart" locations but ultimately the fate of the creature is in the hands of the overland unit movement system which we agree on is weak. Basically there are the same as building units.
-Unit buffs - I still think the AI is inherently bad at using these because they aren't aware of the actual effect of the buff and cannot use them effectively because, again, buffed units are units and are moved by the overland stack/attack system.

[Nelphine]

Correct. But they can spam globals, dispels, disjunctions, fire storms, curses, volcanoes, and corruptions, city buffs, item creation, extremely effectively. So if a given AI is in a situation where the human player would cast say, a city buff, then the AI has to cast enough spells to get the city buff. But let's say 90% of his spells are summons/enchantments. And only 2% are city buffs. To be sure of casting a city buff, the AI has to cast 50 spells. And 45 spells will be summons. And 4 spells will be something else. So instead of just getting the city buff, the AI gets these other 4 spells, that it has been taught how to use as competently as an expert player.

So instead of just getting summons (which it needs due to stack problems) and the one strategy it was aiming for, it also gets 4 other fully effective strategies.

That's what I mean by the overland spells are in s frustrating state right now. The AI does what it should. And it gets the summons it needs. And both of those are extremely amazing, and the fact it works is great. But it also by the nature of the system gets things that it wasn't aiming at. But because its been taught how to use those other things, it ends up feeling (to the human player) that its actually getting to do multiple strategies at a time PLUS all its summons. Whereas the human player has to pick between all the strategies.

[Seravy]

fire storms and disjunctions yes, volcanoes yes.
curses are nonrepeatable - you can have only one on each city.
global enchantments are as well, you can only have it out once.
City buffs, same, only one per city, you can't cast it anymore after all cities have one.

Your numbers are just nonsense. The AI isn't getting a casting bonus of 400% or 5000% whichever that example is trying to suggest (to me it sounds like 5000%).

It gets +150% on impossible so using your example, it will spend 20 full turns to cast that city buff it wants, and will cast one instance of 4 other spells belonging to 4 different strategies each during those 20 turns. That cannot be perceived as pursuing 4 strategies actively, far from it.
Meanwhile the human player can cast corruption 100 times or summon 9 Stone Giants in those 20 turns which is a real strategy and probably works better than 2.5 times as much random spells.

Assuming the AI is as smart with spells as a human player and never casts anything wasteful (which it does, some creatures just aren't effective, some buffs are just crap and some city enchantments just aren't needed if the AI is not under attack etc), they are able to execute twice as many strategies on extreme and 2.5 times on impossible - which is roughly as much as the amount of extra summoning they need or less - they summon to their capital, all their cities distributed randomly, and their "frontier city" each at a 1/3 chance, out of which only the last one will turn into offensive stacks, and the first only after the capital is already filled up. But this only guarantees they have the same number of these creatures in offensive stacks, they are worse at using them so they also need more.)

Anyway, I'll play a few more games before I declare extreme too difficult, as I wasn't playing particularly strong wizards. (This last one should have been winning though, especially with that start.)

I want to try at least these 3 :
3 L/ 4 S / 3 N, Alchemy Archmage - buffing/hero/economy strategy with potential late game Sorcery based win, not sure which race. This is a very versatile setup that can win by many different ways depending on found spells. (tho I wish I had at least 1 more N book - maybe drop archmage?)
5 S / 5 C , Chaneller - slow but unstoppable if it survives the early game, should probably be lizardmen or nomads to ensure that
10 D,  Specialist, Archmage - I played this many times in the past when we tested Death and know it is powerful and works.

[Nelphine]

Globals and curses are cast once.. And then each time they are dispelled. You can't survive if famine is on every city, and the AI can cast it far more often than you can dispel it. It won't, but it can. If the AI has the power production advantage (it will if it matters), and a spell power advantage, it will be a lot more in practice than 2.5 times as much. Summoning also (in my experience) isn't 90%, its more like 50%. So the problem does appear to happen, and perception matters for things like this.

The chaos in my impossible game just cast 17 volcanoes in 4 turns. That's not a rare occurrence. (They only have about 170 casting skill)

[zitro1987]

I'm still 3rd place in the charts after defeating a wizard, having 40+ paladins, extremely good heroes/champions, and a series of summoning units up to storm giants (soon to have focus-magic behemoths).

I'm playing merely on hard, and I'm surprised how the early game appears not as challenging as the late game. Horus is casting about 1 storm giant a turn, 1 uranus blessing another turn on dozens of cities. Freya has v.rares and often casts gorgons. The death wizard seems beatable though, but still seems tough. It packs almost as much challenge on 'hard' as it used to be in earlier versions of CoM in 'extreme', and this is after you toned down the AI modifiers.

I am aware of some excellent players in this forum, but it seems like my comfort level in challenge is now 'hard' while a family member of mine who plays is 'normal' (that person I think beat original MoM in 'impossible', so did I).

My take is as follows:

% bonuses tend to be too high still, but the extra books on higher difficulties is a great choice and should stay.
*Power bonus should match gold bonus
*Production bonus seems too high
*Maintenace bonus gets absurd by higher difficulties.

Easy: With modifiers in the 75% range (essentially an AI with disadvantages)
*the game should be fairly easy by most beginners
*the game should be laughable by veteran players

Normal: With modifiers similar to player with the exception of a bit overland casting, magic power and maintenance help (current 'normal' is kind of difficult, should just have about 110% prod/gold, 130% power)
*Forgives many player mistakes, but provides some kind of challenge - not a stretch to lose some troops and towns
*Ideal for experienced players looking to try a new strategy or beginners who won 1-2 on 'easy'
*Still on the easy side for the experts who know the best strategies and tactical gaming

Hard: with modifiers similar to current 'hard' with maybe a bit less on power/gold, especially production
*Forgives several mistakes, but clearly not for beginners
*Ideal for experts looking to try a new strategy and going down in difficulty
*When the experienced but not quite expert players are looking for a challenge.
*For MoM experts trying CoM for the first time.
*Moderately easy for experts, should generally win with only some challenge.

Extreme: modifiers clearly favor AI, but with a bit less % bonuses than current, especially production and power
*Forgives only some mistakes. Experts-only.
*meant generally for experts to have fun and win most of the time. Meant for Seravy, myself, etc to win most of the time (I mean really, if the creator of this mod sometimes loses 'extreme', then it's too hard!)

Impossible: have modifiers somewhere between 'extreme' and 'impossible' with production maybe at the 'extreme level'
*meant for experts to have a clear challenge, but they should beat it at least 40% of the time with good strategies and tactical gaming.
*Currently, it is borderline impossible except for a couple extreme exploits - this is not a fun option and hardly playable.

[Seravy]

(I mean really, if the creator of this mod sometimes loses 'extreme', then it's too hard!)

I wish, it's not "sometimes" more like, "usually" nowadays. It might be the fault of playing too many test purpose games though.

Agreed with most of your descriptions and if these last 3 games I'm playing confirm it, I'm going to adjust difficulty in 4.01 roughly the way you describe.

[Nelphine]

I'd really like to gear what kind of thing allowed the impossible victory in the main thread.