We've now created an economy where the only way to get clicks is to make an article that is more angry, more provocative, and more biased than the competing articles. The market is optimizing for values that are far removed from what actually benefits society. And the machine will keep optimizing those values to the point of absurdity. To the point where it tears society apart. To the point where the average reader spends their entire life and never hears an opposing point of view.
Obviously facebook doesn't have a "filter_bubble.php" and literally no one is claiming that. It's understood this is just a natural result of using machine learning to customize newsfeeds. But it is a real result, that has real consequences.
You could maybe even argue that people want this. I'm not sure if that's really true. But it's irrelevant to the argument that it's bad for society as a whole.
Just answer this one question. Do you honestly believe the world is better off with this, than if facebook did something else?
Now as to whether people actually want this, I don't think they do. The problem with collaborative filtering is that the basic assumption is wrong. That if people click or like a post, that means they actually want to see more content like it. Or that if they don't click or like it, they want to see less like that.
This is just obviously false. People like and share things to advertise their political positions. Or at least in the hope that other people will see it. Likewise they don't vote on things they don't want to advertise, or don't want other people to see.
If that sounds confusing, concrete example. On reddit I sometimes read an article that I disagree with slightly. I don't upvote it, because I don't want it to be ranked higher so more people see content I disagree with. Or I don't want to give social rewards to the poster. But I still might want to see more content like it in the future. Maybe I'm interested in the subject, or like arguing about it, or just don't care about occasionally reading stuff I disagree with.
The only way to properly do collaborative filtering is if "likes" and votes are private, and are only used to predict what articles you will like or dislike, nothing else. It can't feel like "liking" is equivalent to a vote for that position, as opposed merely "show me more articles like this".
Then facebook also uses information like whether you clicked on a post. But that just encourages clickbait, articles with provocative or misleading titles. You should only take into account whether people like the article or not, after reading it. Otherwise you are just optimizing for clickbait even the users don't actually want.
Lastly there is just standard false information. If I read an article filled with incorrect arguments and information, I might not know the information is incorrect and like it. On a meta level, I don't really want to see articles with false information. But as a random user, I have no way of knowing that, so just my like behavior doesn't take that value into account, and so wouldn't optimize for it.