When this was a topic in University a few years ago, it was still very hard to properly detect humor, especially sarcasm. If I understood this correctly, this whole episode was kicked of by Stephen making a joke about some lady's attire. I don't know what would be worse; the sanctimonious fallout or his tweet being censored by an automatic insult detector in the first place. (actually I do: the second).

I was thinking about a tool, not a universal censor :) Something like a special place for notifications: "Hate mail" - some people with loads of noise replies would definitely appreciate that. Also offensive replies could be shown last when displaying tweet with replies, for example.

How many false positives would you be willing to put up with before you left a platform altogether?

I'm not saying the detector should filter all tweets everywhere, was just thinking about an option for some sensitive people to turn on which would affect only tweets mentioning them, for example.