Here's my answer: Let's say n=10^10. In one context (if these are iid draws from some 1-d distribution) this would be considered good size dataset, and it would be appropriate to model them by drawing each point on a single line, and estimating statistics or density. But what if these points represent pixels in a 10^5 x 10^5 grey image? You could find the average pixel intensity in the image, and think of the image as a distribution of pixel intensities, and scrutinize it to death. "Studying the image." On the other hand, you could chunk up to a more appropriate level of abstraction, and think of the 10^10 pixels as being just a single point in a high dimensional space.