AI improvements

[Chromatophore]

Please explain, really do because i believe they are good they way are, buff them too much to hard to farm therefore no point.

If the a.i. is better and its too hard we can make the fg's advantages less.  I'd rather play against a smarter opponent who cheats less.

[Bhlewos]

^This isn't about buffing AI in terms of deckbuilding, this is a topic about mistakes that the AI makes in terms of logic. For example, when you have more than three creatures on the field and the AI has multiple Plagues in its hand, it will waste all of them in one turn instead of saving a couple for later, which a human player would do. Another example is, until recently when you were playing an RoL-Hope deck the AI wouldn't target the RoLs specifically to break the combo. Now, it'll treat RoLs (and other light-producing creatures) as primary targets if a Hope is out. It's stuff like that -- we don't necessarily want to make the AIs stronger, just less stupid.

[Calindu]

When Ai lobotomize my Archangels instead of RoL's.This is stupid because Ai kills RoL's instead of Archangels
When Ai summons little creatures even if my oty is out.

[wavedash]

AI seems to prefer to discard Ghosts of the Past instead of Fractals. Seems like a problem.

[sbluen]

The AI sometimes does not target its opponents weapons to prevent its otherwise obviously assured death.

[Terrilocks]

Ai had a PAed Pulverizer out, then drew an eternity and played over it -.-

All in one turn.

[Bulrog]

The AI dont target his Phoenix with Cremation.

[Tom Bombadil]

Ai likes to lobotomize Archangels over rols even if there's a hope out

[Calindu]

When Ai lobotomize my Archangels instead of RoL's.This is stupid because Ai kills RoL's instead of Archangels

Ai likes to lobotomize Archangels over rols even if there's a hope out

[Ravensvoice]

When Ai lobotomize my Archangels instead of RoL's.This is stupid because Ai kills RoL's instead of Archangels

Could not find the original post this quickly, but archangels with 7Hp are a lot harder to kill than rol's with 1, so if the AI can kill a rol with 1 spell or need sometimes 3 for an archangel, killing rols makes sense even if he would consider the healing of an archangel to be the bigger threat. Lobo on the other hand does not care about how strong a creature is, only how strong its ability is. and since abilities on creatures with more health are more usefull if removed since those are likely to survive longer, the Ai can be forgiven for removing a useless healing power over a usefull bioluminessence since AI has to be written to do a fair job in al cases. this means it must also work in a deck with lightning storm and lobo, where loboing the angels could definately be the best move. Now making an AI that does the best thing in all cases is both undesirable(makes the game too hard for most people) as well as a nigh impossible task

[Chromatophore]

Maybe far far down the road (in hundreds of years even) we could look forward to an a.i. that keeps a database of commonly used player strategies and uses some sort of probablistic mechanism to adjust its play based on the cards it has seen.  (I.e. an a.i. that makes educated guesses about what the human's strategy is and reacts to it).

A practical (though still arduous) step towards this goal would be an a.i. that recognizes what cards are in its deck and plays accordingly, waiting for combinations and so on.  It would not be necessary (even if it would be nice) for the a.i. to actually analyze its deck.  Instead (at least at first), each deck could come with its own a.i. file that tells it how to play.  For example, a deck whose success is predicated on quinting its only creature could demand that the a.i. not play the creature without quintessence in hand and sufficient quanta.

Ambitious?  Certainly.  But i think we can set our sights on the horizon and aim high.

[wavedash]

Maybe far far down the road (in hundreds of years even) we could look forward to an a.i. that keeps a database of commonly used player strategies and uses some sort of probablistic mechanism to adjust its play based on the cards it has seen.  (I.e. an a.i. that makes educated guesses about what the human's strategy is and reacts to it).

A practical (though still arduous) step towards this goal would be an a.i. that recognizes what cards are in its deck and plays accordingly, waiting for combinations and so on.  It would not be necessary (even if it would be nice) for the a.i. to actually analyze its deck.  Instead (at least at first), each deck could come with its own a.i. file that tells it how to play.  For example, a deck whose success is predicated on quinting its only creature could demand that the a.i. not play the creature without quintessence in hand and sufficient quanta.

Ambitious?  Certainly.  But i think we can set our sights on the horizon and aim high.

Marvel vs. Capcom 3 has a mode where you can play against the computer that tries to emulate the playstyle of some of the developers. It'd be cool to see something like this.