• Stovetop@lemmy.world
    link
    fedilink
    English
    arrow-up
    6
    ·
    2 days ago

    I am not sure what the user above is thinking, but to play devil’s advocate:

    One thing that modern AI does well is pattern recognition. An AI trained on player behavior, from beginner level all the way up to professional play, would be able to acquire a thorough understanding of what human performance looks like (which is something that games have been developing for a long time now, to try to have bots more accurately simulate player behavior).

    I remember someone setting up their own litmus test using cheats in Tarkov where their main goal was just to observe the patterns of other players who are cheating. There are a lot of tells, a big one being reacting to other players who are obscured by walls. Another one could be the way in which aimbots immediately snap and lock on to headshots.

    It could be possible to implement a system designed to flag players whose behavior is seen as too unlike normal humans, maybe cross-referencing with other metadata (account age/region/sudden performance anomalies/etc) to make a more educated determination about whether or not someone is likely cheating, without having to go into kernel-level spying or other privacy-invasive methods.

    But then…this method runs the risk of eventually being outmatched by the model facilitating it: an AI trained on professional human behavior that can accurately simulate human input and behave like a high performing player, without requiring the same tools a human needs to cheat.

    • zqps@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      4
      ·
      2 days ago

      Cheating humans already perform closely enough to trick such a system. Many cheaters are smart enough to use an aimbot only for a split-second to nail the flick. With a tiny bit of random offset, those inputs indistinguishable from a high-skill player.