relative to the length of each user's arm
It's easy to snidely dismiss that idea, but seriously, how would you define a unit of length in an environment that is supposed to translate between multiple devices and media types? A "pixel" makes little sense when drawing on a sheet of paper or a super-high DPI screen where pixels aren't individually visible. It can't be a fixed real-world distance, because pixels would sensibly be larger at movie-screen distance than at arms' length. They basically defined it in terms of the arc-length of the picture upon your eye, which is completely reasonable to me.