More fundamentally, you want to be able to distinguish the representation(s) of true from false as efficiently as possible, so to work on most machines the choice is either zero vs nonzero or negative vs nonnegative.