Requests for AI improvements

[Nexhro]

AI doesn't pay mana to untap creatures when Mudslide is on the battlefield, even if it has lots of mana sources and no cards in hand.

[BAgate]

I've noticed the AI has been disadvantaging himself more recently. He targeted himself with a Cursed Rack and chose a color that he played and I didn't for a Curse of Wizardry.

[HarlequinCasts]

I've noticed the AI has been disadvantaging himself more recently. He targeted himself with a Cursed Rack and chose a color that he played and I didn't for a Curse of Wizardry.

I feel like the AI is making random choices upon spell resolution. I've seen it do wildly suboptimal things. A very common one is it will correctly choose to cast Tinker, but then it will Tinker for a Mox Sapphire or something completely ineffectual even though its deck has Platinum Angel or Blightsteel Colossus or something much much better.

[BAgate]

Will of the Council AI sucks. Was just testing Council's Judgment, and the only logical thing to do is for the AI to pick whatever I picked, but it sometimes doesn't (almost always will pick an artifact if one is available after I pick a creature). Don't know if this is card specific or global, but the AI's choices are BAD.

[Nexhro]

AI always atatches Kudzu to one of its own lands when one of its lands enchanted by Kudzu is destroyed. It used to attach it to the player's lands in the vanilla version from the 90s.

[Nexhro]

Will of the Council AI sucks. Was just testing Council's Judgment, and the only logical thing to do is for the AI to pick whatever I picked, but it sometimes doesn't (almost always will pick an artifact if one is available after I pick a creature). Don't know if this is card specific or global, but the AI's choices are BAD.

Considering Manalink only supports 1v1 games anyway, it would be best to have Council's Judgment use an approximation that just exiles target nonland permanent the opponent controls. Allowing the AI to "vote" just makes the card overpowered.
The only downside to that approximation I can think of is the fact that Brago's Representative, depending on its controller, wouldn't enable the removal of multiple targets (though that interaction could probably be hardcoded) or help the opponent to have Judgment exile their least important nonland perm, respectively.

It's a card that would go into almost every EDH deck that has access to white mana, but it feels like cheating to run it given the state it's currently in.

[Nexhro]

The Ai stops playing lands as soon as the player controls Exotic Orchard, even if Orchard can already produce mana that the lands in AI's hand produce. For example, if teh AI plays an island and teh player follows up with an Orchard, the AI doesn't play its other Islands.
A turn 1 Orchard will lock the AI out of the game.
The AI will resume playing lands once Orchard can produce all five colors.

[HarlequinCasts]

The AI doesn't seem choose a target for Vesuva. It always just comes into play as Vesuva with no abilities regardless of other lands on the battlefield.

[BAgate]

The AI chose Swords to Plowshares when damaged by Sphinx Ambassador. Talk about useless.

[HarlequinCasts]

Just curious, do we have any idea why sometimes an AI just chooses to do nothing? Specifically what I mean is, I sometimes see an AI that will play no lands or spells, and then discard a land at the end of turn. Sometimes they may even play a Sapphire Mox or something, but will still discard a land.

(28 Feb 2014, 07:29)


I thought I would bring this up again. I see this behavior once every several dozen games. The AI will just decide not to play lands for the first few turns, or never at all.

In the attached screenshot, you can see the AI did nothing on its turn and even discarded a land at the end of the turn. No idea what occasionally causes the AI to just not load apparently...

[Aswan jaguar]

HarlequinCasts wrote:
I thought I would bring this up again. I see this behavior once every several dozen games. The AI will just decide not to play lands for the first few turns, or never at all.

As I have told at other times AI has serious problem handling some lands (I made a huge but unfinished survey for that)some cause AI serious problems especially if they are in the opening hand they can cause AI to miss 2-3 turns land drops so AI can't play spells either.Others lands can't be played by AI all the time and cause AI to miss casting spells that would need +1 mana than AI has in play.
From AI's hand 99,9% it is Dimir Aqueduct causing trouble as AI sees it is the best land to play but then has no land to return,again best land to play,....... so after making this computation many times AI ends up doing nothing.

These lands as Dimir Adequate at least didn't cause any trouble before the last update or the one before as Jatill had a made a hack for AI only,to play instead of Dimir Aqueduct the equivalent in colours Underground Sea for example.

Because AI will ONLY play these lands if AI has already a land in play and at the beginning of the game AI hasn't a land in play because AI thinks highly of these lands and prefers them from most lands that makes AI not to play nothing.
Also other parameters or code can mess AI computation concerning lands.

[HarlequinCasts]

Aha I was secretly wondering if it was the bounce lands but wanted to check here first. That makes sense from an AI perspective at least... interesting.

[Nexhro]

As mentioned here, AI doesn't even play a bounceland if it already has a land on the battlefield. It waits until it controls at least 3 lands before it starts playing bouncelands.

[Nexhro]

The AI loves to sacrifice all its permanents to Renounce, regardless of its life total.

[HarlequinCasts]

Has the AI ever changed to add randomness to its decision making in the recent-ish past? I seem to recall the AI used to be a lot better at maximizing its board position, but more often I see it choose the correct spell but then pick a seemingly random (sub-optimal) target.

For example, I have a Noble Hierarch and Ulamog, the Infinite Gyre in play and the AI has a Control Magic in hand. It seems like 50% of the time the AI will choose the Noble Hierarch...