Assuming the youtube recommendation algorithm is mostly just "people who watched this also watched", then couldn't the asymmetry in the recommendation graph str

Assuming the youtube recommendation algorithm is mostly just "people who watched this also watched", then couldn't the asymmetry in the recommendation graph structure simply be a relatively accurate reflection of asymmetry in the underlying data? It seems plausible to me that the probability that people who hold "extreme" views will watch "moderate" content is lower than the probability that people who hold "moderate" views will watch "extreme" content. I haven't run any experiments, but it seems to me that a difference in probabilities like this would induce the kind of "closed group" structure being discussed on the resulting recommendation graph. The low number of steps to the closed groups shouldn't be that surprising, given previous famous results about degrees of separation in these kinds of graphs.