The most well known example of recommendation services that rely almost exclusively on your social graph is that of Facebook, where it's generally assumed that if many of your close friends like a topic, that you will as well. My common example is that if your friends like Mamma Mia, it's assumed you like Broadway plays and if many of your friends have given a like to one establishment or another that you would of course be more interested in that. This is extended with tools like Foursquare and others, who give the social connections a big push to bring you new stuff you might like.
There's no question that collaborative filtering brings value. But it's not enough. What's more accurate is that you are driven by your own preferences and just because your friends like something doesn't guarantee that you will. Only you have your unique makeup and interests, and the knowledge of how to rank those interests. For example, you may like both Katy Perry and Depeche Mode, and you might prefer Depeche Mode to Katy Perry, or specifically prefer one song of Depeche Mode or the entire album continuous. That's your preference, and what's needed is the ability to learn from your own behavior. The Web sites and applications that do this in the right way will start to separate themselves from the competition.
So look at the news just from the past few days. Google is now going to personalize your Ad experience based on your activity in Gmail. That's a good move, based on your own implicit history. But in parallel, we have more chances to tap social signals with +1 in the search results (for starters). At my6sense, we talk a lot about the best signals to watch for to deliver a real human experience that's personal. And it takes more than just collaborative filtering. See the embedded presentation to see just what I mean.
Disclosure: I am vice president of marketing at my6sense. I was not compensated by Kynetx for my presentation, but they did cover expenses for the trip.