I think part of the problem at FB's scale is getting any level of consistency across human moderation. If you ask 10 good, experienced, human moderators whethe

I think part of the problem at FB's scale is getting any level of consistency across human moderation. If you ask 10 good, experienced, human moderators whether a controversial post is allowed, you'll probably get at least 15 different answers and justifications. FB would probably find it harder to train from this, as each moderator will be using different justifications, and reaching different decisions. At FB's scale, they want something cheap and straightforward which they can implement automatically and go Kafka on the lack of appeals. The reality people need to wake up to is that when a business is at FB scale, people who fall through the cracks due to a broken algorithm simply don't matter. Google doesn't care if you get locked out of your Gmail, in the same way FB doesn't care if you get spuriously banned. When users place more value on a service than the service places on them as a customer, it feels like a market failure scenario that leads to these kinds of outcomes. When the go-to price is $0 though, it isn't clear how you resolve this.