Teaching an AI to Lie at One Night Ultimate Werewolf

Поделиться
HTML-код
  • Опубликовано: 19 ноя 2024

Комментарии • 5

  • @MattEland
    @MattEland  Год назад

    An article write-up of this video is also available at accessibleai.dev/post/teaching_ai_to_lie/ along with links to code, videos covering gameplay, and a few more goodies.

  • @telph3223
    @telph3223 Год назад +1

    I know in one night ultimate werewolf, there are some situations where the good team are incentivized to lie. For example, the troublemaker can lie about switching players A and B when they actually switched player C and D. Then player A might admit truthfully that they had the werewolf card and claim B is now the werewolf, because player A thinks they are now playing for the good team. After player A has made their role claim, the troublemaker can change their story and say that they actually switched players C and D, and tell everyone to vote out the player A because he was tricked into admitting he was a werewolf.
    If you remove the assumption that good roles always tell the truth, the game of one night ultimate werewolf becomes 10x more complicated.

    • @MattEland
      @MattEland  Год назад +1

      Correct. That's why that assumption is in place during phase 1 of development. Very good justification of my scoping decisions!

  • @tomat0o
    @tomat0o Год назад +1

    Is it kind of like The Resistance?

    • @MattEland
      @MattEland  Год назад

      Avalon / the resistance are social reduction card games like ONUW, so yes, there's a high degree of similarity.