Let's be precise about the terminology here. When we talk about the computer “seeing” the differences between documents, that is not to imply that the computer actually understands those differences in a literal sense. Perhaps a better way of putting it is that the computer “registers” the distinction between The Shining and the Apple manual, because it can describe that distinction numerically. The numbers reveal something about the meaning of each document, even if that revelation is an oblique one, like a low-frequency tone shimmering in a puddle. The computer can't hear the meaning directly, but it can catch glimpses of it in its statistical reflections.