The article is written more from a let's-get-together perspective, which is nice, but also consider any system that tends to help people find more of what they already know they like, based on their similarity to others -- the only variation you'll get is if "friends" happen to have already tried and liked something new, and you might then try and like that same thing.

What we should all want (yes, even you, dear reader!) is a system that not only helps us find what we already know we like, but what we might not know we like because we haven't tried it yet -- a sampler, if you will. (Or as my friends know it, the "mixed grill". Tasty!) At the very least it should find representative entities for each group, and have a user try them out one at a time; that way they don't have to explore everything to get a better feel for what they'll like.

But there's also the idea of randomly introducing items into the preferred list without telling the user -- they'll just think it's a mistake, but maybe they'll give it a try, and maybe we can help friends discover how different they are from each other in addition to what they already knew -- how similar they are. Also, consider that tastes change over time; a long-lived system shouldn't continue to suggest that you listen to the same music you liked when you were a teen; it should constantly force you to give other music another try, to see if your tastes have changed. You can't do that without a little randomness, a little slyness.

The "related items (implicit)" system we use on this site is a case of only finding what you already know you'll like, and as such, doesn't help people browse and sample the whole thing. Then again ... for a stand-alone site, that's probably not what most of our (rare) visitors want anyway.