I still don't understand how a Fermi question makes for a good interview question. How can you make reasonable estimates about something you know nothing about?
It's likely everyone you interview knows nothing about it, so it isolates logic and reasoning skills from previous knowledge. The accuracy isn't important.
I don't like those type of questions either, I said it would make a good Fermi question, not a good interview question :-) For some reason, some people love to ask such questions. I'd rather ask how to perform face detection on 1 trillion images.