Total War: WARHAMMER - Official anounced

[Easo]

So I guess we should just "Ok!" to CA and have forever shitty sieges? Alright then. No one needs perfect AI, it just needs to be able to do something. It cannot even stick to one decision, which is probably the biggest issue (climb up walls and shoot everyone down there, while they randomly run back and forth until dead for the best example).

[KrayZ33]

So I guess we should just "Ok!" to CA and have forever shitty sieges? Alright then. No one needs perfect AI, it just needs to be able to do something. It cannot even stick to one decision, which is probably the biggest issue (climb up walls and shoot everyone down there, while they randomly run back and forth until dead for the best example).

I never said anything about that, but I asked you why you make it sound so easy and why you expected anything different when there, apparently, aren't any games out there that do better.
So where do your expectations come from?

I even agreed with you saying that the AI is shit. But that's the case because the combat in TW:W is complex and if all they do is A-move you get even worse battles. See how efficient corner camping is for example.
Because the A move is what the AI does when it can't properly flank you.

It doesn't blast you away with spells, it doesn't use artillery until all the ammo is used up.
It's hard to program that I'd say.

I'm pretty certain that a proper TW AI would need proper machine learning.

[Easo]

I never said anything about that, but I asked you why you make it sound so easy and why you expected anything different when there, apparently, aren't any games out there that do better.
So where do your expectations come from?

I even agreed with you saying that the AI is shit. But that's the case because the combat in TW:W is complex and if all they do is A-move you get even worse battles. See how efficient corner camping is for example.
Because the A move is what the AI does when it can't properly flank you.

It doesn't blast you away with spells, it doesn't use artillery until all the ammo is used up.
It's hard to program that I'd say.

I'm pretty certain that a proper TW AI would need proper machine learning.

Your "easy" is debatable, but I expect more from a company with good enough cash flow and years and years of experience, especially on the same engine.

[Jastall]

Na na na, we are not going to go with "but but others do not have iiiitttt". Not to mention those titles are way older than TW:WH series. IIRC the current engine and pretty much the AI comes from the Rome 2/Shogun 2 times. Yes, I know AI is hard, as someone who has spent way too much time in Stellaris and Paradox attempts to make it less infuriatingly dumb.

This also does not excuse Chaos ponies and other recolored troops.
P.S.
Purely technically SC2 has absolutely superior AI capable of crushing almost all humans, if you remember Deepmind.
Still being actively developed https://github.com/deepmind/pysc2

Deepmind is a specially-developed AI made by people outside of Blizzard, designed for 1v1 Melee games of SC2. Put it in any other situation (even just a custom map) and it more than likely shits itself silly, given that what we know as AI in games is more like a whole bunch of more or less complex scripting.

The more variables you add to a situation, the more chances you have of the AI being unable to handle them all. That's basically what hamstrings the Total War AI, a conflicting set of priorities combined with over-reactiveness. Layer on top of that an absolute shit ton of variables that occur over a Total War campaign (including wildly different unit types, maps, battle types, magics/abilities to use, list goes on forever) and you have the result we see. Paradoxally, the more tools you give the player, the more the AI will shit itself trying to choose which one to use. For example take Skaven garrisons in game 2; the Warlock Engineer that is inevitably there is feared for his Warp Lightning casts. Why? Well first because Warp Lightning + close quarters = ouchie, but also because he has no other spell, so he always uses that one. As soon as you engage a leveled up Lore of Ruin caster he'll sprinkle some far less effective Crack's Call or Wither casts or whatever which are generally far less effective. More tools for the player, more ways for the AI to misuse them.

There's no cure for this, not one within the budget of a video game studio. AI is extremely difficult and expensive to implement, and having one that is even mildly competent at using what will soon be Warhammer's 22 races (23 with the inevitable Chaos Dorf, at least), hundreds of different units, on over a hundred maps spanning half a dozen battle types... not gonna happen, ever, even if they had billions to throw at the project.

All that said, I do think they should have been able to plug the worst AI goofs, such as wasting ammo on fast single entities, suiciding their flying units, leaving their artillery with 0 support, being headless chicken when outnumbered in sieges, and staying on walls to be obliterated by arrow fire. Speaking of which, I really do hope walls not come with significant (60%+) missile resistance, what are those battlements for if not?

[Dezerte]

I think you may be overselling it a bit on how hard it is to design AI. I would say, yes it is hard to design an AI that can act like a player, but in the case of several spells and utilities being made available to the player it's not hard to design a priority system, or a condition system for the AI (e.g. a priority action first on the list: if X enemies are grouped cast Y etc).

However, another limiting factor is that it can often be a performance issue. Having a complex AI, can be very taxing on the system. This is something that is obvious in grand strategy games, where the game can almost slow to a crawl in the end stages of the game.

[KrayZ33]

I think you may be overselling it a bit on how hard it is to design AI. I would say, yes it is hard to design an AI that can act like a player, but in the case of several spells and utilities being made available to the player it's not hard to design a priority system, or a condition system for the AI (e.g. a priority action first on the list: if X enemies are grouped cast Y etc).

However, another limiting factor is that it can often be a performance issue. Having a complex AI, can be very taxing on the system. This is something that is obvious in grand strategy games, where the game can almost slow to a crawl in the end stages of the game.

I really think you are underselling it.
It's not easy for the AI to even realize how the "worth" of a unit is stacked up on a battlefield like this.
Especially if you add priorities, range, cost, reinforcements, travel time and and and and.


I'm certain it can be done better than the way it currently is with enough investment. But every single thing you will mention and add, will have consequences on how the AI decide the next steps.

The funny excel programming of "if X then Y" is not only underselling the work that has to go into it, it's also extremely vulnerable to player abuse.
Just like how SC or AoE AI is absolutely weak to kiting units.
They probably attached some HP/DPS values to these units and the AI will now send spearmen towards your horse archers because it has the statistical "upper hand" in straight up combat.

Too bad however that my units can shoot on the move and I simply shoot and move 5 yards backwards into the range of my castles again.
The AI sees that it will lose the battle and retreats.
I attack follow with my horse archers (still shooting at them)
The AI thinks it has the "upper hand" and attacks again
I send my units back into castle range again.
The AI sees that it will lose the battle and retreats.

and repeat this for the next 2 years and the AI is still going to do it until it runs out of ressources, units, buildings or whatever you can come up with.

The same thing happens in TW:W when the AI shoots your lords/heroes on flying mounts and you micro them until every range unit is dry.
How would you go on and design it to ignore your shenanigans and go straight towards your army... and if you do, what are the consequences... it's not difficult to make use of that new behavior as well.

[Jastall]

I think you may be overselling it a bit on how hard it is to design AI. I would say, yes it is hard to design an AI that can act like a player, but in the case of several spells and utilities being made available to the player it's not hard to design a priority system, or a condition system for the AI (e.g. a priority action first on the list: if X enemies are grouped cast Y etc).

However, another limiting factor is that it can often be a performance issue. Having a complex AI, can be very taxing on the system. This is something that is obvious in grand strategy games, where the game can almost slow to a crawl in the end stages of the game.

The thing, those priorities have to be very carefully balanced, lest several of them pile up on top of each other and the AI cascades into purely reactive failure as it tries its best to adjust its priorities all the time. This is very obvious in sieges when outnumbered, as it'll shuffle units around nearly mindlessly to react to what the player does. And if you enforce priorities above all else (for example, force it to only use damaging spells, or to hold back its flying SEMs, or to spend significant forces protecting its artillery...), then the player can exploit those new parameters as well. It's a neverending cycle.

Plus as you said, performance in TW games is already not the best. Running many times more scripts in the background to hopefully make the AI smarter won't help at all. I'm already a bit worried that even official trailers for WH3 seem to run at 20 FPS sometimes, and it's not like there are huge armies on-screen. The game is still 4 months away but still, I hope it releases better optimized than WH2 considering 3K and Troy run far, far better than it.

[Makabreska]

[Jastall]

Loads to like here.

-Healing is now % of life based, and fire damage debuffs it by half for a few seconds. This both makes healing more consistent across the board and introduces a counter to it. Think you can heal that SEM being hammered by a Bloodthirster or shot at by Pink Horrors? Not so fast bucko.

-An army's initial formation will be saved between field battles. Fuck yes, saves literally hours of busywork per campaign.

-Flying toggle was already known but very welcome still.

-Better formation controls, on top of a bracing icon.

-Slow-mo when using spells and abilities is also neat.

Plus a bunch of other UI changes and such. The unit selection screen for MP battles alone is far better. This is the stuff I want to see. Fundamental mechanical improvements that makes every campaign, every battle better. Hopefully the campaign side of the equation sees just as many improvements when it comes to supply lines, diplomacy, endgame crisis and such.

[KrayZ33]

I don't quite understand what the healing change will change other than the fact that it buffs high HP units and nerfs low HP units that work with %-dmg reduction?

Or is that just it?

It will hopefully still heal entities in a unit by a big amount and not "0,8%" for each entity per tick?

[Easo]

Well, at least I can stop giving myself nightmares with creating checkerboards for my gunners every single time.

[Jastall]

Hot damn that was a sweet trailer. I love Nurgle's look possibly even more than I imagined, Ku'gath looks positively glorious atop his palanquin and he gets both Plague Toads and Beasts of Nurgle on launch, with Nurglings seemingly being a swarm unit? Maybe it's trailer-only effects but I saw them grapple and overwhelm Slaaneshi marauders in sync-kill like animations, and they seem even more numerous than Skavenslaves. Fingers crossed that it's in the game.

Slaanesh looks real good as well, the animations are just fantastic. Only thing I'm not a huge fan of is N'kari's face, he/she/it looks the wrong kind of weird. I know it's based on official artwork but still, I really hope Slaanesh gets the Forgeworld model for the SEM Keeper of Secrets as it's IMO by far the best interpretation of it.

[Easo]

Hot damn that was a sweet trailer. I love Nurgle's look possibly even more than I imagined, Ku'gath looks positively glorious atop his palanquin and he gets both Plague Toads and Beasts of Nurgle on launch, with Nurglings seemingly being a swarm unit? Maybe it's trailer-only effects but I saw them grapple and overwhelm Slaaneshi marauders in sync-kill like animations, and they seem even more numerous than Skavenslaves. Fingers crossed that it's in the game.

Slaanesh looks real good as well, the animations are just fantastic. Only thing I'm not a huge fan of is N'kari's face, he/she/it looks the wrong kind of weird. I know it's based on official artwork but still, I really hope Slaanesh gets the Forgeworld model for the SEM Keeper of Secrets as it's IMO by far the best interpretation of it.

Too many tiddies on that model, you can forget about it.

[Corvus]

Oh yeah, Papa Nurgle is here.

[Jastall]

Too many tiddies on that model, you can forget about it.

They have tiddies on models already, just cover em up if CA fears the American/Chinese censors so much. N'kari, the generic KoS (maybe Exalted?) and Daemonettes all have just enough to cover the nipples which is all that's required.

[KrayZ33]

Best trailer yet imho.

[ParanoiD84]

Looks abolutely amazing and it's already one of my all time favorite games and it's just going to get better.

[Corvus]

Papa Nurgle's mechanics are out - and they nailed it

https://www.totalwar.com/blog/plague...e-in-campaign/

I've been on the fence really so far about WH3 but now I'm going to get it. Nurgle sounds like so much fun.

[Jastall]

Nurgle's my first playthrough. Well the first one I'll start, if it ends up like WH2 I'll have two or three campaigns happening more or less at all once in different parts of the map. Probably Nurgle, Cathay, and then either Khorne or Kislev. The only thing I don't love about him so far is the lack of mortal units. I know they're not a focus on launch but Khorne had Warriors, Skullcrushers and chariots, Tzeentch had Knights, and Slaanesh seems to have a couple varieties of bespoke Marauders. I'd be disappointed if Nurgle has literally only a Cultist hero to represent the mortal side on launch. And we've already seen much of his building tree with no mortals in sight so I'm not optimistic so far.

[Belize]

Done, I've finally conquered an entire map, with all the rogue armies dead too.

Whew.