My specific, stated metric was pigeonholing.

You receive less information online when these systems are employed, ie more repetition in the same space. Youtube optimizes for maximizing the viewing time across the most common types of viewing associations, namely people watching for entertainment, not looking for information. If you ever look for information in a new topic, you tend to be signposted back towards your old time-sink staples from your history instead.

Certainly, one can also point to algorithmic enforcement of echo chambers, which would be a subjective value judgment as well. It feels good in immediate practice (and can increase viewing time), but many don't view it as a net positive at scale for social issues. Yet for things like small-scale hobbies, it can be beneficial.

This is basically the obesity/fast food scenario, but just for ideas. YouTube Facebook etc being the new McDonalds.

Out of interest, do you have a source for that "petabytes of analytics every day"? It sounds like the kind of thing that would come out of an interesting "How we cope at scale" kind of article.