Caster of Magic 2 AI alliances and elimination tests

[Seravy]

Test 1.
I accidentally let this run for an unlimited duration instead of the intended 1518, but stopped it in 1525.
I think the picture speaks for itself, fixing the starting relation bug made AI alliances pretty broken again.
   

I will run more tests just in case...

Tests use Expert difficulty, 13 players, default settings on everything else.

[Seravy]

Second test :
   
   

Yeah, too much alliances.
It might be helping in eliminating the wizards who didn't have enough allies, but this isn't playable.

I believe the idea most worth trying is adding a -15 penalty to AI to AI diplomacy positive rolls for each alliance the AI making the roll has. That should automatically limit wizards with lots of positive modifiers to like 3-4 alliances at most and those without the modifiers to 1-2 without affecting any other systems.

But that's for tomorrow...

[Seravy]

3rd test, this time with a positive diplomacy roll penalty of 20 per alliance already in place.
   

Looks promising. Freya, Rjak, Ariel, and Sss'ra are Chaotics so their alliances are not "normal" alliances and we should ignore them which kinda makes this not very useful as a test unfortunately. The only "real" alliances are the white-blue and the violet-red.

The test ran until 1418 and only two wizards are dead.
   

Many of them have been reduced to insignificant levels of power but they are still in the game...

I'll run the next test without Chaotic wizards.

[Seravy]

4th test results (No chaotic wizards allowed)
   
   

Number of alliances looks about right. Wizard elimination is a bit slower than intended but it seems to be happening.

[namad]

BTW, you might want to add a distance component to alliances. If you limited alliances by making it so that if fortresses are something like 30 or greater squares apart (or some other large number) was a large minus penalty to alliance acceptence... you might get results more similar to what people expect in 4x games. Of course in those other games the lack of long range alliances is more often due to lack of scouting instead?

I know that aggressive force projection is based on some distance math (either tiles or same continent or both?) So, it might make sense for alliance considerations to use a similar bit of math. Now, logically, you wouldn't want to form alliances with the same people that you might consider taking hostile action against... however for the purposes of a game... I'd say actually... alliances are irrelevant unless there exists some chance you might have wanted to consider taking said hostile action?

Although your current solution is probably fine too. I was just thinking it might be nice if alliances groupings were formed into clusters by region. Although with how mobile and effective scouting can be in master of magic, maybe that's pointless. My thinking is routed in games like stellaris/civilization for what to expect in terms of alliances and regional groupings.

[Seravy]

Test 6
   
   
This doesn't look so good and reminded me of one problem, and that is, you can still have a lot of alliances on one AI if they are not the one making them but the ones on the receiving end. So the modifier should be the greater of the number of alliances on either AI wizard involved instead of just the one making the treaty.

Also, we had a runaway AI on Myrror this time, this is what you get when one person hogs all the lairs on a plane in a 13 player game :
   

Finally, the amount of damage monsters do with 3 players having Call the Wild in play is insane. This is exactly the kind of stuff why there should be fewer wizards still in the game this late. (tho at least this spell at least solves the problem it creates as the monster eventually do eliminate wizards, other stacking global enchantments don't.)

[massone]

Did the runaway AI happen because there was only 1 wizard on the plane even in a 13 wizard game? Is that randomized or set up to be like that?

I don't agree with alliance penalties for distance. You normally want allies who can help you attack a mutual neighbour, which means you want allies on their other side.

Also, I think a simpler fix is to apply the same 40-relations threshold for alliance. Why isn't the AI-to-AI alliance subject to the same rules as player anyway? I've mentioned before in my human played test games, that I found some particular alliances strange. It should be harder for non-shared realm wizards to ally each other, or even Pact.

As it stands, it's far too easy for AIs to pact with each other, at lower relations than the human player. This is probably a big part of the problem.

[Seravy]

Test 7

   
   

The two charismatic wizards were able to get more Alliances (4 and 5) but both have a high personality modifier in addition (Lawful or Peaceful) so I believe they are ok as is, or at most, the Charismatic modifier could be lowered from +40?

Wizard elimination definitely doesn't happen fast enough here either.

Did the runaway AI happen because there was only 1 wizard on the plane even in a 13 wizard game? Is that randomized or set up to be like that?

No, there were 3 of them but one (Freya) was already eliminated early. Her Historian bar barely went anywhere so it was probably an early war, possibly paired with a bad start.

The runaway AI had Dark Elves as far as I remember, which paired with Life+Nature sounds like a recipe for win, accelerating magic power with Stream of Life based growth and summoning strong Nature creatures buffed by Life magic to take lairs and nodes quickly.

Then we have Omniscient and she found a book in all 5 realms...

I don't agree with alliance penalties for distance. You normally want allies who can help you attack a mutual neighbour, which means you want allies on their other side.

Exactly. An ally who is near is just a pain who kill-steals your conquests. But an ally at a remote location where you don't want to conquer anything at all is great.

Why isn't the AI-to-AI alliance subject to the same rules as player anyway?

Mainly because they play by different rules in general as a core design in general. The AI is bound by formulas and rules in their decisions but the player is free to act.
Like, if an AI sends me a Wizard's Pact request then I can accept it even if our relation is -50.
But if I do the same to the AI, they must refuse because that's the rule, and the rule exists to provide a clear threshold of how much tributes the player must offer (or how much books and alignment they must share) before they can have a pact or upgrade that to an Alliance.
However AIs can't tribute to each other. If there is a threshold, they can't put effort into overcoming it. So that would hard lock the AIs into a system where they only make treaties with each other if their starting relation is high enough (or they cast Aura of Majesty) in other words, limits the game to alliance blocks where everyone shares the same realm. Which is bad because it makes the size of the alliance blocks entirely random and potentially very high so it doesn't fix the problem.
It's not like I don't want to fight against an alliance of 7 AI wizards who share a realm, I simply don't want to fight against a block of 7 allied AI wizards, period, because that's unwinnable. (and yes, we could possibly say I don't have to if I ally with all 7 of them but looking at it realistically, that's either not happening or if it does then we have the "human player has too much alliances" problem...)

If the problem is AIs pacting too easily, then we can change the Pact formula from (Formula>=102) to a higher number. Is it though? The AI actually needs to roll success on the pact before they end up in a war through rolling a success on hostility or war declaration instead. (and even then a war declaration roll can dissolve the pact although that grows more and more unlikely the longer it's in effect through raising relation)

[namad]

My thinking wasn't that you'd penalize someone (AI v AI) who was your neighbor's neighbor, or just behind your neighbor... but someone who was the neighbor of the neighbor of the neighbor of your neighbor.

Hence the 30+ tile distance, not 10-20. In my idea there'd be no distance penalty to forming an AI + AI alliance to dogpile someone in between themselves. But then there would be an alliance penalty that would reduce the odds of someone behind behind behind that first ally from joining in on the dog pile. Someone that far away might not be likely to gain territory from the war, but they would count as an ally and penalize the more useful dogpiling style alliances that you wanted.

The idea being that at some quite long distance you're simply too far away to matter and too far away to waste decision space on alliances. Although also hostile PvP spells cast on the overland map have unlimited range, and that may or may not be desirable to add to the dog pile.

In other words this idea isn't actually meant to reduce the number of allies, but instead to reduce the number of irrelevant alliances to increase the amount of relevant alliances that form, taking into account the fairly high per alliance modifier you're already using.

The hope I have here is that you can get successful dogpile style wars going on where two neighbors team up on an AI between them, while having fewer total overall alliances in the game. With the added benefit that the player might actually be able to comprehend the wars they see on the diplomacy screen. Having an interesting 2on1 AI v AI war go on near the player is an important facet of the game and having 5or6 irrelevant additional alliances belonging to that war, but being too far away to participate might just be confusing.

My hope behind the suggestion is that you'd end up with the ability to weirdly raise your per alliance penalty by adding this extra new penalty on top of it, and somehow, you'd end up with more functional alliances (as in alliances that led to war declarations against 3rd parties). Simply because your current penalty is stacking per concurrent alliance and if you can prune a couple unimportant alliances from the list then that penalty stacks fewer times.

Maybe the AI isn't smart enough for them to consider distance though, and fortress to fortress distance is actually kind of a terrible measure of distance as many AI empires will end up looking like snakes and two adjacent neighbors might have fortresses quite distant from each other. So, yeah, I can see how this idea wouldn't really work, because the concept of border and the concept of neighbor both don't really exist in master of magic. Essentially this is just an idea I lifted from stellaris, and it might not apply here at all

[Seravy]

Test result 8 :
   
   

This looks reasonable for alliances. No one is Charismatic this time so that seems to confirm the Charismatic modifier is too high.
Next run, I'll try tweaking the elimination combat modifier and the effect of Charismatic and personality modifier in the AI to AI positive diplomacy formula, because it feels excessive.