Well since it works in camera space of the ones are allways <> because thats the eye ray in camera space and the second is cameraspace face normal! (which in turn can be computed by taking cross(point1-point2,point3-point2) or any combination really), maya knows this at all time tough so you can just ask maya for it)
Offcourse if you do some sort of raytracing then the eye ray can vary.
Anyway understand that back face culling does NOT hide all faces, just the ones hat absolutely wont participate in the render, theres a lot of extra faces that don't contribute and tahts the job of the z buffer, or trace (in which case you do not really need back face culling).