If you imagine a cartesian coordinate space where your samples are clustered around the origin, then a zero vector will tend to be close to everything because it is the center of the cluster. Which is a different way of saying that there's a bit of nothing in everything I guess :)