Nice article in The Atlantic on recommendation algorithms and those that rely less on social endorsement than automatic other things. The important thing is the bias that systems always exhibit, and that we treat these as black boxes. (I described our approach to our learning as a black box in the first lecture, same idea.)