> I hate the idea that just because it's fuzzy to define what constitutes unwanted behaviour online that we must allow all behaviour. That we'll be censors destroying free speech if we ban trolls.

I think the correct idea is somewhat different. That we should be very careful in classifying speech as "troll" and extend the maximum effort in not using "unwanted behavior" as an easy substitution for suppressing dissenting opinion. There should be boundaries, and they should be wide enough to allow proper discussion - including challenging dominating opinions and expressing controversial and upsetting views.

> Automatic moderation won't be perfect, but neither will human moderation.

The concern is not that it will be imperfect, it's a given. The concern is that it would be less perfect than human one, while being perceived of being more perfect, due to the aura of "objectivity" that algorithms have. If a moderator unfairly targets you, you could complain about this moderator's biases, you could quote your interaction, you could appeal to other moderators or wider audience, you have plenty of options human interaction allows. You would not always get a just outcome, but there's some chance. If you are declared toxic by Google bot, what your options are? Argue with a bot?