I wonder if the problems the author describes can be solved by artifically downvoting and not showing spam and flamewar content, not banning people.
- Spam: don't show it to anyone, since nobody wants to see it. Repeatedly saying the same thing will get your posts heavily downvoted or just coalesced into a single post.
- Flamewars: again, artifically downvote them so that your average viewer doesn't even see them (if they aren't naturally downvoted). And also discourage people from participating, maybe by explicitly adding the text "this seems like a stupid thing to argue about" onto the thread and next to the reply button. The users who persist in flaming each other and then get upset, at that point you don't really want them on your platform anyways
- Insults, threats, etc: again, hide and reword them. If it detects someone is sending an insult or threat, collapse it into "" or "" so that people know the content of what's being sent but not the emotion (though honestly, you probably should ban threats altogether). You can actually do this for all kinds of vitriolic, provocative language. If someone wants to hear it, they can expand the "" bubble, the point is that most people probably don't.
It's an interesting idea for a social network. Essentially, instead of banning people and posts outright, down-regulate them and collapse what they are saying while remaining the content. So their "free speech" is preserved, but they are not bothering anyone. If they complain about "censorship", you can point out that the First Amendment doesn't require anyone to hear you, and people can hear you if they want to, but the people have specified and algorithm detects that they don't.
EDIT: Should also add that Reddit actually used to be like this, where subreddits had moderators but admins were very hands-off (actually just read about this yesterday). And it resulted in jailbait and hate subs (and though this didn't happen, could have resulted in dangerous subs like KiwiFarms). I want to make clear that I still think that content should be banned. But that content isn't what the author is discussing here: he is discussing situations where "behavior" gets people banned and then they complain that their (tame) content is being censored. Those are the people who should be down-regulated and text collapsed instead of banned.