preventing filter bubbles (so to speak) is an explicit design goal of recommender systems. Of course it's a goal. They're just not very good at achieving that

preventing filter bubbles (so to speak) is an explicit design goal of recommender systems. Of course it's a goal. They're just not very good at achieving that goal. I wouldn't expect someone who makes their living building these things to be unbiased on the question of whether or not they're currently a bad thing. If I search incognito youtube for "Trump Speech" and select the first video, 3 of the next 4 "up next" are also Trump speeches, and the other one is "McCarthy SHREDS Pelosi". If I click on of those 4, the next "up next" videos are different, but also 3 of the first 4 are Trump speeches. The other one is a Trump-Biden debate. This is a filter bubble. Sure, human psychology plays a part, that's why algorithms like this work in the first place. But compare this to the opposite extreme pre 24-hour news networks and there was simply much less opportunity for playing into this. That's what algorithms provide: The ability to "instantiate" filter bubbles much more readily than content that is passively delivered without consideration for past choices, and the consumer has to make all of their choices without that outside influence.