Requests for AI improvements

[Aswan jaguar]

Flatten, Grasp of Darkness and others that share code with them are not used very good by AI.If I am not wrong AI is set to only cast them targeting 4 toughness creatures or creatures that need 4 damage to die.So in my tests:
- AI doesn't cast the card even to save itself from death by unblocked attacker if it doesn't have 4 toughness. or
- to take advantage by killing humans 3/3 creature that holds back AI's 3/3 creatures, or
- to deal with a creature with trample like Ball lightning or
- and possibly not deal with an indestructible creature of toughness less than 4.(this not tested) etc.

[counterspell15]

Have you ever played against the AI and AI was using Fact of Fiction? I noticed that AI doesn't make good decisions when taking cards to the hand. I tried to do the following: choose a pile of lands and other pile. The AI takes the lands to its hand and drop the rest to the graveyard

[FastEddie]

Some general observations, not so much about specific cards.

The AI often plays (correctly imo) cards before attacking that support the attack like Muscle Sliver or Sinew Sliver in a Sliver deck. However, a Bonesplitter Sliver is usually played after attacking. Same for a second Cranial Plating in an Affinity deck. Imo these cards should be played before attacking.

Generelly the AI seems to plan only one step ahead. For example, I play Eureka, AI plays Phantasmal Forces without having blue mana. The Phantasmal Forces die during the next upkeep. Similar (at least in the old Shandalar, but I presume the same holds for Manalink) the AI plays Stasis without having a blue mana left.
My impression is that it uses a tree with only one level, weights the leaves and chooses the best one. It would be cool if the same algorithm could be run based on the outcome of either each leaf or at least certain leaves, based on some pruning strategy if the former is too computationally intensive.

Similar but maybe not quite the same: the AI does not attack when it needs two turns to win or would loose creatures on the way. Say the AI has enough creatures to deal enough damage to win over two turns, even if I block perfectly during both turns and it looses some creatures during the first turn. Ignore hand cards on my side that might change the outcome like Giant Growth, Fireball or Wheel of Fortune for a hail mary pass. I think a human player would employ that strategy at least once in a while.

[Aswan jaguar]

The AI often plays (correctly imo) cards before attacking that support the attack like Muscle Sliver or Sinew Sliver in a Sliver deck. However, a Bonesplitter Sliver is usually played after attacking. Same for a second Cranial Plating in an Affinity deck. Imo these cards should be played before attacking.

The "Lord" type of cards that boost others are set to be played for AI in both main phases as AI most of the times plays them well in first main and few times if AI will not attack in 2nd. If we make AI to play them only in 1st main then we get another issue that to my experience is more often than AI make a bad use of "Lords" and not cast them before combat although AI did attack or would attack if it cast them before.
When AI draws cards in combat or 2nd main can't cast the drawn "lords" the turn AI got them so loses a turn and wastes mana.
For Cranial Plating and most equipment AI is set to play them only in 1st main but activate them anytime so AI can equip attack and then reequip a creature that didn't attack and defend with it, AI uses that correctly sometimes.

Generally the AI seems to plan only one step ahead.

Unfortunately AI can see only until end of it's current turn so we should give hints to AI for these cards to cast them only if AI has such mana available.

Similar but maybe not quite the same: the AI does not attack when it needs two turns to win or would loose creatures on the way.

That is happening in the vast, vast majority of AI attacks however I have seen AI to attack and win in two turns even if AI would loose creatures. Either AI has some algorithm which we don't know about that will make it attack even in two turns to win and loose creatures doing so or it is possible in these few times that AI does something like that to think through bad code or functions we don't understand well that can win that turn. Still some of those times AI didn't have anything so complicated to mess it's calculations.

[FastEddie]

The AI is too trigger happy with Strip Mine (probably also Wasteland) and Zuran Orb. What I mean is that it immediately sacs either Strip Mine or arbitrary lands to Zuran Orb, thereby depleting its mana base.

For Strip Mine (and other "Destroy-target-land" lands) it would make sense for the AI to wait until it has enough mana, either lands or other sources. "Enough" could be the average CMC across all cards in the deck or maybe even one mana more, either in play or on the hand (it could make sense to sac Strip Mine on turn one if you have four more lands in hand, for example). Once the mana base is established it could sac the Strip Mine without undue damage to itself.

Zuran Orb is used immediately now, i.e. every new land is sacced, thereby the AI never has enough mana. Instead it should sac lands only when its life falls below 1 during a phase and then sac only enough lands to get back to 1 or 2 at most.

[Dav3377]

The AI will block a 2/1 Dwarven Soldier with a 2/2 Orcish Veteran with First Strike activated. It does not appear to consider that doing so gives the Dwarven Soldier +0/+2 (making it a 2/3). It will also attack with the 2/2 orc when a Dwarven Solider can block it, become a 2/3, kill it, and survive. Typically the AI does not seem to sacrifice 2/2s to 2/3s in this way.

[etphonehome]

AI never leaves a creature to defend against critical damage.

I've seen this a lot of times.

AI always atacks if there are no creatures to block and next turn your creatures get untapped and make critical damage and you win the game.

THX

[woghomm]

1. Gatekeeper of Malakir
AI always choose it to cast on second and third turns without kicker payment, afterfore have problems with creatures coming on table. Of course, most players will wait it for cast for BBB cost.

2. Boros Reckoner
AI not feel it ability so oftenly it cause to blocking it with a creature or do many damage to him, that will cause to losing a game.
Or AI just do damage bolts on him, that will cause a problem too.

[Blyfycyfyfyf]

Hi, common situation - Pillory of the Sleepless on AI's creature, yet AI still pumps the 'pacified' creature with various effects, even though this creature cannot attack. And it happens mostly in first main phase.

Is this something that actually can be changed? The target is of course legal, but makes no sense.

[Konatsu]

This may be a known issue, but I'm attaching the save data.
Land that becomes a creature will upset the AI's judgment.
In particular, when paying for echoes, the AI doesn't do that and pours all its mana into those lands.

[Konatsu]

AI does not appear to be able to use Odessey's "filter land".
(e.g. Skycloud Expanse)
I would be happy if you could fix this.

-I have seen AI using Azorius Signet etc. successfully.

[Aswan jaguar]

I have already fixed AI with Odyssey lands and other lands or mana producers for AI, check here.

[Konatsu]

Faith's Fetters has a big problem.

The AI attaches it to its own creatures when my creatures are present.

When this happens, the AI then goes violently crazy.
All or some of the following behaviors occur.
-AI makes a disadvantageous attack and self-destructs the creature.
-AI can block without loss, but does not do so and sandbags.
-AI stops casting spells.

[drool66]

After allowing EVENT_SHOULD_AI_PLAY through in IS_AURA_EVENT(), AI no loner plays Faith's Fetters on its own permanents in any way I could ascertain.

EDIT: Commmitted in 8257bd5

[Konatsu]

The AI tries to attach Cartouche of Strength to my creature, fails, and it is placed in the graveyard.