Gaps and oddities of the computer A.I.

[masterpoker]

I'm starting this thread in an attempt to better understand the computers AI in the game Planeswalker (2010/2014 mostly).

I've noticed sometimes the computer doesn't always seem to make the best move, a 2010 example would be with the Cloud Sprite 's in Jace's deck, even with a clear sky they refuse to attack as a 1/1. The Goblin Sky Raider in Chandra's deck from 2010 acts almost random when it decides to attack, not always opting for growth first.

Playing just now on 2014, the computer had a 1/1 Squirrel Token (using Chatter of the Squirrel ) and I had nothing on the field but a plains, it refused to attack (the time thinking on combat phase almost nothing).

Bating both 2010 and 2014 has never been difficult, sending out a 1/1 with a Giant Growth in hand is a pretty classic move. With VERY few exceptions the computer will risk almost anything to stop the little critter, seemingly disregarding possible bonuses.

Targeting seems to be a big issue, and it shows heavily when playing the Relics of Doom deck with the Root of the Firemind deck. We were playing yesterday, and he had a pair of Gelectrode 's out just absolutely decimating the other players creatures. The computer virtually ignores Gelectrode, opting instead to fry my Bottle Gnomes who are virtually no threat to anybody.

Both are set to Planeswalker difficulty, with 2010 set for Auto AI Time and Throttling enabled.

I'm not trying to trash the A.I., which in both games is good enough to present a challenge to most seasoned vets. Just thought I'd share my experience playing against it, show possible faults that could be corrected.

I've noticed the 2014 Deck Editor has an AI Personality option, with about 40 different ones in there. Can anybody tell me more about what the differences are, I've usually just tried matching colors to AI (monored decks get Chandra, blue get Jace, ect).

Thanks for anybody who read this far, and consider giving your own insights on the game's AI for me to read.

[RiiakShiNal]

The AI is basically just a min/max engine that tries to best guess what the next moves will be. It works fairly well with the cards that come with the game, but is notoriously bad at trying to play modded cards. This is likely due to the fact that it doesn't understand how many of the modded abilities work and thus can't put a proper value on them to minimize or maximize in its calculations. Though it can be noted that the AI does get to "peek" at your cards so it is in a way cheating (especially at the higher difficulty levels) as it attempts to check on how certain abilities will play.

So with modded decks and cards the AI's ability to play a specific deck is pretty much a crap shoot, sometimes it works often it doesn't.

As for the differences in AI Personalities, just the images, names, and background music. There is no other difference between them.

[masterpoker]

Thanks for the reply sir.

I mentioned Squirrel tokens before, having played with them more the last few days I am pretty confident in saying they seem to be broken (DOTP 2014). The enemy never blocks my Squirrels, not if I'm totally tapped out and he's got a 5/5 on the other side. He also won't block to save himself with a squirrel token. They don't act like most of the other game tokens, in terms of how the computer reacts with them. Is it possible something with the token is broken?

For my Gelectrode (DOTP 2010) example, is there any way to teach the computer that this is a potentially dangerous card. It seems to understand that Prodigal Pyromancer or Glaze Fiend is something that needs to be destroyed asap, but not this one.

I understand the AI won't always make the best choice given all the options, it just feels like all the options aren't on the table for it to decide, and I'm wondering if there's a way to change that?

[RiiakShiNal]

As for the Squirrel tokens it is entirely possible there is something wrong with the tokens, but without looking at the code it is impossible to say.

For Gelectrode it is probably just a matter of giving the card/ability a higher score to tell the computer this is a card that should be dealt with somehow.

Sometimes a card or ability can be "fixed" by giving the AI additional information about it, but some abilities are just impossible to tell the AI how to cope/understand them.