[glass3]

AI moderators don't have to be personal. It would be enough to tag comments as constructive, lulz and flame and let users decide which type they want, or to which degree they want it. The technology of spam-detecting Markov chains should be enough since the users can easily correct the classification.

Since the feature is possible but not used, question remains why users don't need it.

[EVa5I7bHFq9mnYK]

If I were AI, I would remove any post not containing a link or a number. And remove any comment containing a sequence {capital I followed by a space}. Including this one :)