How Does AI Determine Its Targeting?

2»

Comments

  • Kinesia
    Kinesia Posts: 1,621 Chairperson of the Boards
    octal9 said:
    wereotter said:
    I concede it may be limited to certain cards, but the same cards the AI uses to kill its own creatures, it has also used to kill mine meaning there is some degree of generic risk assessment it’s applying to all available targets without regard to who controls the creature. 

    correct, this is the difference between a hostile targetresolver and a neutral one

    just read this closer; it's only partially correct. the degree of generic risk assessment does not always treat your creatures as "more dangerous" - side of board is irrelevant in the eyes of the neutral resolver

    how old are you guys??? ;)


    lol

    It's not polite to ask a lady his age!
  • Kinesia
    Kinesia Posts: 1,621 Chairperson of the Boards
    James13 said:
    It's as Octal says.  If it's improperly considering it's own creatures as targets it just goes by "threat" priority, ignoring ownership.  Which I don't have all the details on, but it considers card rarity and power stat as part of that calculation.
    It also seems to take number of reinforcements into account now.
    Testing with the 2 leviathian's it kills the cheaper one with the highest power, whereas most other examples it's usually the most expensive thing unless there is a stack of cheaper things.

    My current theory is that it uses "cost bands" and then compares based on other things including stack size and power.
  • FindingHeart8
    FindingHeart8 Posts: 2,731 Chairperson of the Boards
    Kinesia said:
    octal9 said:
    wereotter said:
    I concede it may be limited to certain cards, but the same cards the AI uses to kill its own creatures, it has also used to kill mine meaning there is some degree of generic risk assessment it’s applying to all available targets without regard to who controls the creature. 

    correct, this is the difference between a hostile targetresolver and a neutral one

    just read this closer; it's only partially correct. the degree of generic risk assessment does not always treat your creatures as "more dangerous" - side of board is irrelevant in the eyes of the neutral resolver

    how old are you guys??? ;)


    lol

    It's not polite to ask a lady his age!
    say what?

    *brain cramp*
  • span_argoman
    span_argoman Posts: 751 Critical Contributor
    edited July 2018
    Casting cost also seems to play a part in the AI's creature assessment. I've had the opponent Bolas with a charged 1st ability skip killing my Ghalta for a turn when its cost had dropped to zero on the board and it didn't have reinforcements. This is with me having only a Ghalta and a Gaea's Revenge on the board.
  • Bil
    Bil Posts: 831 Critical Contributor
    edited July 2018
    Casting cost also seems to play a part in the AI's creature assessment. I've had the opponent Bolas with a charged 1st ability skip killing my Ghalta for a turn when its cost had dropped to zero on the board and it didn't have reinforcements. This is with me having only a Ghalta and a Gaea's Revenge on the board.
       I actually noticed that same thing, AI seems to avoid ghalta or even tishana despite their big bodies if there is another creature with higher casting cost but lower stats ( zacama in my case).  That would also explain why creature tokens are safe as long as a creature is on the field.
  • Kinesia
    Kinesia Posts: 1,621 Chairperson of the Boards
    It seemed like it was originally _just_ casting cost.
    At the time I started watching this behaviour tokens had printed casting costs on them too.

    This "theoretically" is a reasonable approach for an AI targetting judgement as long as cards are costed appropriately (but we know they aren't!)

    Tokens had costs taken off display but they might still have "hidden" ones for this purpose.

    They then _seemed_ (this is all from observation and word of mouth) to start multiplying these costs by stack size to get a rough "danger level" of the stack.

    My new experience with Slinn Voda and Aethersurge Leviathan suggests that it is not strict cost, but cost "ranges" and it starts taking power (or something else) in account for ties, because Bolas ALWAYS targets Slinn Voda even though Aethersurge costs more and they are the same rarity.

    I suspect that something like 16-19 is a cost "bracket" for example. Perhaps I should start filling in a spreadsheet for Bolas games (when you can be sure there is lots of targeting to record!)