"Machine learning will help you find your best friends."

Yes, but how? How will it avoid optimizing on the wrong thing?

"It will gather a lot of data and then use algorithms to predict your affinities from it."

Yes, but how? What algorithm would do that? Ordinary recommender systems, neural networks, something else? And tight knit communities of close friends rarely have a lot of churn, but at-scale systems often do - how will it reconcile these tendencies?

I guess I was looking for something a little more like a research paper.