Imp. Question About Eldrazi Spawn before the next release

[lujo]

Has anyone tried Eldrazi Spawn decks with Shandlar AI?

The Manalink AI handles them horribly - it will often sacrifice them for no reason at all and prioritize sacrificing them over tapping lands. It will also hold on to an Overrun while having an army of them on the board. It won't even chump block with them!

It will literally sacrifice them for no reason during seemingly any time (it's turn, opponents turn, doesn't matter).

So can anyone confirm it's the same in Shandalar, so that we can report this in the bug thread and maybe there's still time for Korath to tweak it before the release? Because you could make a reasonably functional Eldrazi deck (or several different ones) for the AI if it didn't ever sacrifice them, sacrificed them for 6+cc spells only, (obv. prioritized lands before even tinking of sacrificing them), but when you see it sacrifice a bunch to cast Coat of Arms (or sacrifice them post Overrun but before attack and do NOTHING with the mana) it's very

UNIMPORTANT EDIT: The more I think about it the more it's mind-boggling in the pure "what has been seen can not be unssen" phenomenological or spectacular sense. The little bastards have a million applications and truly misusing them is quite difficult. A montage of what AI does with them in Manalink would be actually amazing footage of very inventive insane behavior.

Just try to run this thrown together thing past the Manalink AI 5-6 times, open up the opponents hand so you can see what it choses among everything it coudl choose, see what happens, if it's fixed you'll miss out on bizzare sights:

Code: Select all

.11001   4   Nest Invader
.10973   4   Kozilek's Predator
.10885   4   Broodwarden
.10942   4   Growth Spasm
.11069   4   Ulamog's Crusher
.11040   4   Skittering Invasion
.91   23   Forest
.409   4   Desert Twister
.3222   2   Overrun
.1513   1   Gaea's Cradle
.285   4   Wild Growth
.996   2   Coat of Arms

Like this, I just had to share this:



(open in new tab to enlarge) That broodwarden was cast by sacrificing 2 Eldrazi Spawn. Now pay attention to the 2 untapped forests. That's pedestrian, btw, the really mind blowing stuff I can't screenshot - because you only realize the level of insanity after it's happened - like the AI throwing down Coat of Arms... and sacrificing all it's tokens and casting nothing whatsoever!

[Korath]

It's mostly because they're flagged as activateable instead of just mana source. This is always an error, unless the card has a non-mana activated ability too, or perhaps if its mana ability has a side effect like Pristine Talisman.

[lujo]

Hmmm, but I didin't see this level of insanity with just about anything. Also, do you mean they're flagged separately as a mana source for the AI and for the player?

I made a separate thread because I made a report about prioritizing mana sources acting strangely, but these little dudes have to be taken into account. They make colorless but probably should be prioritized the last after any available option. Or something.

EDIT: It's really odd because the interface will go out of it's way to not autotap your Sol Ring to pay for colorless things, and will go out of it's way to pay whatever it's casting with these guys even before basic lands

[Korath]

None of that follows either from what I said, or what actually happens.

Autotapping and AI mana tapping, which uses the same algorithm, doesn't prioritize anything higher than basic lands. (It does give Gem Bazaar and cards that produce only artifact mana, like Mishra's Workshop, the same priority, and it'll sometimes prioritize mana sources lower if they've been enchanted by Blight or Psychic Venom or something similar, but that's not relevant here.)

Having the Activate flag makes the AI consider spontaneously activating them, rather than just when it's paying a mana cost. I'm reasonably certain that what's happening is the AI following a branch where it activates an Eldrazi Spawn, independently sees it has enough mana to cast a Broodwarden afterward, and does so; alternately, it follows a branch where it sees it can activate an Eldrazi Spawn and have enough mana to cast something, and then when it speculates a layer more deeply after the sacrifice, it decides it's better off not casting whatever it was.

I can't construe anything here to even vaguely imply that there's a difference between the AI and the player.

Please limit bug reports to actually-observed behavior in Shandalar. If you want to make guesses about why that happens, that's ok, so long it's separate from the actual behavior; but such guesses by non-developers are notoriously inaccurate, even in simple, sanely-designed, well-understood software, which the MicroProse MTG engine is none of. Further reading (the world would be a better place if everyone who used software got through at least the introduction section). And of course talking about how it works in Manalink or Forge or MODO or Apprentice is just about never helpful.

---

On the other hand, I'd appreciate your assistance with bug #735, which I don't get locally, but apparently very many other people do, including you in your screenshot above; and nobody's been terribly interested in tracking it down.

[lujo]

I got #735 all the time. How do I help?

(Also, the question the tread was about, before my rambling ruined it, was "does this sort of thing happen with Eldrazi in Shandalar?")

[stassy]

So far, after running a few times against an Aga Galneer with your eldrazi deck (wizard mode so 19 life, nearly like a Manalink duel), I only saw once a bad sac choice for nothing at cleanup phase :
- AI correctly sac if I cast Control Magic on it
- AI correctly sac for mana to ramp for others Nest Invader / Predator / Skittering Invasion / Ulamog (it even comboed Nest Invader to Skittering Invasion to Coat of Arms to Overrun, massive sac of token but got 2 of them alive on the bf with 8/8 trample stats and Broodwarden for the kill)

[lujo]

Thaks for trying it out!

That sounds like good news, except the more you play the more those will crop up. You'll see it sac 3 tokens to get a broodwarden onto the board with no tokens left, or cast a broodwarden and then sac all the tokens for Coat of Arms. Or cast Overrun and then sac tokens before the attack. I've seen it make fine plays and it beat me often enough, too, but the player probably wouldn't be saccing tokens for anything other than Ulamog, if you think about it. Not in that deck, not more than 1 per spell at most.

What Korath said explains some of it, possibly all of it, but not necessarily. Or rather, it doesn't necessarily make the AI play them well enough. If it, say, casts a skittering invasion for 5 tokens, and then registers that he has 5 mana sources and sacs them all for Coat of Arms on the same turn that's awful. If the AI gets them in that state, it'll be a deck that beats itself as often as it gets beaten by the player You can't even test that deck properly to see how strong it is, or edit it properly because the AI can make the results too random.

Now, whether this can be fixed or not is up for debate / giving a hoot, but it does figure into how you make an Eldrazi deck for the AI, and all that. EDIT: If the mana source / activated flag fixes the random sacrificing for no gain, they can be used as pure ramp, but making an Eldrazi Spawn deck (as in one which pumps tokens) that won't go haywire could still be problematic. Also, even if you use them for pure ramp, getting the AI to use them to ramp out something expensive instead of blowing them on just whatever is it's own issue (meaning that they would probably be the last sort of ramp you'd want to use in a deck that wants to ramp out a big Eldrazi). If it blows the token from one spawner to cast the next spawner, and then the tokens from that to cast the Broodwarden - wth is is going to cast Ulamog with? And, more importantly, at what turn and after how many other spells? Stuff like that.

[lujo]

I've a different, but somewhat related question:

How difficult would it be to assign a custom AI to an opponent? Code up a system for it along the lines of: The AI goes through a deck and assigns values to specific cards, "slots them in" for certain categories. Or the person making the deck assigns those in the deck file somehow, and the monster uses these values when rating/assessing cards in hand etc.

Just a thought experiment, mind you.

As an example: say we want to have a deck with dark rituals and sengir vampires and want to make sure the ritual gets used to cast the vampire at turn 3 if at all possible. The Vampire gets slotted into "finisher" slot, and dark ritual gets slotted into "ramp" slot, so to speak, and the AI is instructed to only use "ramp" if it leads to casting the "finisher" unless there's 5 mana on the board. Just a random example.

Or, and this is a very concrete problem I've been struggling with, the deck has an enchantment like "Fires of Yavimaya" flagged as "key" (or whatever). Then it prioritizes casting it above everything else if there's not one on the board, in which case it becomes reluctant to cast it over anything else.

A third very useful application of a mechanism such as this would be being able to flag creatures as "expendable", so that the AI attacks into trades agressiviely (like it does with Erg Riders which seems to attack every turn no matter what), which would seriously help with the AI not getting bogged down in ground stalls.

Naturally, with something like this in place it would be possible to have the AI play the Eldrazi Spawn properly, too. Because right now I'm tweaking/building decks around AI limitations probably more than any other factor including card availability and being able to modify timing/evaluation along a few general categories on a per-deck basis rather than having to work around a set global system all the time would be immense.

Oh, and it would help hugely with non-basic lands. Being able to set playing priorities for them on a per-deck basis would be huge.

[BlueTemplar]

This reminds me a lot of the AI insanity I've seen with Mishra's Factory and Zuran Orb :
tracker.php?p=4&t=793
tracker.php?p=4&t=795

[BlueTemplar]

I got #735 all the time. How do I help?

Are you talking about this #735 :
tracker.php?p=4&t=735
?

[Korath]

Yes.

[lujo]

I uploaded the files related to #735 just after I got a fresh install to get the bug in the bug #735 ticket, just in case it's easier to spot, I'm reporting it here, too.