Well, of course, you've never seen any of the videos that have zero views, and you're more than a million times less likely to see a video with one view than a video with a million views. Definitely the YouTube recommender will recommend it at least a million times less often, and I suspect much more than that.

This kind of selection bias pops up in a lot of contexts. When you ride the bus, for example, you're disproportionately more likely to be on a bus that's over half full than on a bus that's mostly empty. And most of your friends probably have more friends than you do. (Not just you. I'm not saying you're unfriendly or asocial. It's true for most people.)

The guy I was talking to, on the other hand, could just run a database query over all the videos, and he did.

It's possible that YouTube has pruned those zero-views videos since I talked to him.