AI

[Seravy]

Finished implementing it.
This is the latest version of the table, do let me know if something doesn't make sense or could be improved in it :
   

[Seravy]

The AI is currently casting Floating island without checking if the tile has enemy units on it, so we probably need to remake the whole procedure.

Any suggestions on what targeting rules to use? Currently it casts it as close as possible to one of the AI's towns chosen at random. I ideally would prefer targeting the registered "ship called here" locations, but that information is unavailable, it's not yet calculated when spellcasting happens.

If nothing better, we can also keep it as is, only adding the "must not contain a unit" condition.

[Nelphine]

I'm assuming this falls under standard conjuration rules, including bonuses for summoning costs etc.

That means the AI should still be summoning 2-3 islands in the same situation a human would be summoning one. That means I wouldn't worry about getting an ideal targetting algorithm. So just making it not targetting a unit is probably fine.

[Seravy]

No, it's an "intercontinental travel enabling" spell, and the category is only used if the AI has too few transport units. Generally it will summon 1 island per level of difficulty while it has no ships, and none once it has more ships than that. So yes, you are right they will summon several, unless difficulty is very low.

[Seravy]

For the "AI is too weak early" issue, I think a fairly good solution would be to give them a difficulty based advantage on initial casting skill. Probably a 2x ultiplier for Lunatic, 1x for Fair and below, which would mean 1.25 for Advanced, 1.5 for Expert, 1.75 for Master.

-The higher skill allows the AI to summon more units. So the player has more things to intercept before they can attack, and are at a greater danger of retaliation if they do not intercept them. A lot of these will be intercontinental - Sorcery will have more Nagas, Nature will have more water walking bears and sprites, Chaos will have... nothing useful unfortunately, Death will have more wraithformed or normal ghouls, and Life again won't have anything really useful. However chaos will have more corruptions if the player picks a fight with them which is something. Life, well, they'll have more buffed units. Probably not intercontinental but they might be difficult to overcome for at least some strategies.
-The higher skill also allows more attrition damage - so makes the fight most costly for the human.
-It also increases minimal human stack size - if the AI casts 1 more confusion or 2 more firebolts each battle, you need that much more forces for each attack to succeed.
-Higher skill does require more mana crystals to fuel, but the AI has them - that extra ~20 skill at most they get out of this is paid for by one and a half magic markets on Lunatic, so successfully not losing as much as 1 city thanks to this, which is probably a conservative estimate, is enough. The AI has a fairly decent pool of mana and gold on turn 1 as well so they are unlikely to run out.
-The maintenance cost on the extra units summoned will make the AI spend more on making MP - so less on SP, causing the skill advantage to slowly erase itself - skill being nonlinear making this bonus obsolete by the time uncommons happen also helps with that.
-One possible downside I see is the AI also spending less on RP - but considering the price of research this isn't likely to be relevant. The ~300 cost of a common spell pays maintenance on 20 Nagas for 20 turns for the AI so they'll be like one common spell behind - but again, successfully not losing a single city that has a library covers for most of this loss, defending 2 is already an advantage. The maintenance cost kicks in over time anyway, so probably won't be enough to use up the starting MP before the AI finds some in treasure or builds enough buildings to no longer care.
-While the AI does get more creatures, it won't allow them to snowball more - if they had a target nearby that is beatable by a 9 stack of their best common creature, they can and will already conquer it. Doing it 4 turns earlier makes no real difference, especially as the AI is not very likely to invest that into snowballing their economy. Since the stacks will go to the main action continent, even having multiple stacks won't allow the AI to clear more continents so their overall efficiency won't increase much in this regard.
-The only real worry is those extra creatures attacking the human player, and extra casting skill causing problems in battles where the human is defending, but that's what the 40 turns of guaranteed peace is for. By then the effect of the bonus should lose some of its relevancy, and seriously, if you can't handle large stacks of common creatures after having 40 turns to prepare for it, don't play Lunatic. Meanwhile, if the player does not take advantage of these 40 turns and attacks the AI, the retaliation being stronger is exactly the effect we are trying to achieve. (also in reality this will be more like 45-50 turns : even if the AI is allowed to attack they might not do it immediately, and need to bring their units into the human's territory.)

So overall I think this might work way better than it sounds at first - the side effects are kinda minimal while the impact on what we want to achieve is very high - deter players from playing pure offense without garrisons.

[Nelphine]

I'm not sure it's a good idea. We've already very carefully measured things like how much treasure the lunatic AI get early (and I already think ai particularly sorcery and death get WAY too much treasure too fast. Having 1+ nodes in 1401 absolutely does snowball the AI like crazy.)

If the AI ever lost fights it wouldn't be so bad - in the same way that the human can misjudge things and lose - but the AI doesn't. They only attack with superior forces, and we've shown with how strategic combat works that equates to always winning. So those extra units absolutely will allow the AI to snowball.

This style of solution specifically aims at making it so the human cannot gain strategic parity and gets attacked for being too weak. In turn this makes games even more luck based dependant on personalities. Get peaceful and lawful and it still seems easy. Get maniacal or ruthless and you're dead.

Giving the AI more snowballing (and more units absolutely translates to more snowballing) doesn't solve the problem.


The ai already has MASSIVE numerical superiority. Look at my halfling game where I had 40 cities and my opponent had 11 and he still had 30% more strategic strength than me.  I was 4 times the size, and some obnoxioud amount like 5 times the power production and he had 30% more military than I did.

Giving the AI MORE units won't solve this. It will make games where the AI splat the human with 9 focus magic naga on the first turn war is allowed more common - but it won't change anything if you survive that. And I still won't build garrisons because they still won't be able to fight off 9 focus magic naga on turn 30-40 anyway. It will drive me further toward using brutally fast rush tactics.


The ai needs to have LESS units in order to encourage better and more diverse gameplay. And if you want to encourage garrisons more we need to make the AI smarter. The two goals cannot be treated the same or the game will devolve to one where only certain tactics have any hope of winning.

[Seravy]

I don't see how this would give the AI more snowballing, see my previous post.

The AI will have the exact same stack of nagas - 4 turns earlier. That only really gives the AI 4 turns worth of node power as an extra, even if they did get the node that much faster. That much won't even cover the increased maintenance costs. Sure the AIs, all of them, will have more of these stacks. Doesn't matter. No snowballing from it. The number of targets a stack of 9 common creatures can beat is limited, the AI would finish most of them before turn 40 (on Lunatic) anyway.
In fact, the AI grabbing the weak lairs and nodes in their OWN territory earlier, is a very good thing and cuts down on various treasure based abuse.

The ai needs to have LESS units in order to encourage better and more diverse gameplay. And if you want to encourage garrisons more we need to make the AI smarter.

That's impossible. This is pretty much the limit, considering the limitations (AIs are NOT allowed to play like a human, so even if we had access to a neural network AI from the next century to achieve that, it would still go against the game design), the complexity of the game and the fact the AI is playing against an unpredictable opponent who can react to its moves and counter them anyway.

At this point teaching the AI to see into the future and let the AI decide based on the player's future actions feels a more realistic goal than making the AI smarter.

Anyway...

Giving the AI MORE units won't solve this.

Then what will? The AI is limited in their actions and by definition is dumber than a human. So it plays with a major handicap. I really hope you aren't suggesting to put the AI's summons into a buffer and have them spawn 9 at a time next to the human's least defended city. That sure works but...that's the level of cheating even I won't tolerate from a 4X game.
Alternately I could put a message at the beginning that says "please don't kill the AI's units and let them reach your cities so you can lose them if you left them undefended" and make everyone laugh at it.

If the players keep killing everything the AI produces, the AI has to produce more. There is no other solution. As much as I'd love to, I can't possibly teach the AI to sneak around taking the longer path just to always be out of sight range of the human's units until they reach a city to attack (even if it was possible the human can just build more scouts to see every inch of the map all the time-spearmen are cheap enough) - and as I previously said and you agreed, there is no stack the human can't wear down and kill once they found it if they really try. Which is good - if an unbeatable stack existed and the AI had it, the player would need to immediately surrender since it is unbeatable and building garrisons still wouldn't stop it.

btw the AI having more units, while counterintuitive, actually reduces the chance for war. Militarist war that looks for weaker players we disabled for the early game, and this only really has an effect there. Equal army war won't trigger, the player is too weak. Relation based war and hostility are unaffected. Of course it definitely makes it harder to make treaties with those AIs - but we can drop some of the difficulty modifier on the same formula to compensate.

Aside from these I do see one more thing we can do, but I think that would be really bad for the game. We can force garrisons if we make rampaging monster stacks stronger and more frequent. Problem with that is, they'll also kill the AI's fresh cities so the AI will fail to expand. The other problem is, it's not fun. I've seen this thing in Civ 4's realism mod - there were so many barbarians spawning I was kinda losing the game with no AI player ever contacted. It also affects late game strategies and probably hurts them the most while we want the opposite - we want peaceful players to be able to build less garrisons while warmongers should need more.

If you want, I can try running the AI on the same map with and without the bonus and compare the results, then we'll see if there is a snowballing potential or not.

[Nelphine]

If you run it with say.. 5-10 maps on expert and 5-10 maps on lunatic, if be ok with it. (You'd need to make sure at least 2 games with each mono realm, and I'd prefer at least 3 games with focus magic sorcery although these could be dual realm or mono sorcery).

But getting things a few turns earlier is exactly how human rush builds anowball. It really is. So I just don't see how it wouldn't snowball.

[Seravy]

From what I've seen human rush builds are faster by dozens of turns and reinvest 100% into offense.
That's exactly what's causing the problem. The AI doesn't do that so it won't snowball as much.
That many maps will take a while but it's doable. We only need to run it for the first ~100 turns fortunately.

[Seravy]

   

3 maps on lunatic done.
The 3 big numbers which are probably the myrran wizards for each map went up consistently - don't like that.

By the way, is this even still a problem? We've been nerfing the problem early strategies, are there really any left that can intercept all the units a high difficulty AI throws at the player?


Would changing the "40 turns" rules so that it no longer applies if the player starts an early war help? Or maybe changing the rampaging monster spawns to be more relevant earlier?

Ultimately the goal is to force the player to have garrisons, not to make the Myrran wizard 10-20% stronger, nor to have an unbeatable amount of nagas on the map. But we need a solution that does not involve making the AI smarter - it's just not possible to outsmart a human doomstack.

We already have an overexpansion diplomacy penalty system which ensures getting into fights with everyone and getting attacked from all directions - how could all the units from 3 AIs be not enough? (ok, assuming the first one is already beaten, two more AIs who attack from two directions different from the one your doomstack went to beat the first wizard.)