4 ms·
>Can you crowdsource the moderation task to double check what the ai is flagging? This would help with false positives, which is a noble goal. But it would not
by filoleg 6y ago
>Can you crowdsource the moderation task to double check what the ai is flagging?
This would help with false positives, which is a noble goal. But it would not help with false negatives, which is imo a very disastrous consequence in this specific scenario.
Something that isn't CP got marked as CP? Sure, a moderator can manually doublecheck that flag and override if needed. Something that is CP didn't get marked as CP? You got a problem on your hands, because there is no way to manually doublecheck that, since that would mean manually checking every single post marked as "no issues here", which gets us back to square one of manual moderation of all posts.
Not even mentioning the exposure of FB users to CP that it would bring, which would mentally ruin quite a lot of people and bring very negative press to FB on the level of severity that we haven't seen before.
- landryraccoon 6y agoSpeaking as someone who has to deal with supervised learning models, human beings falsely tag positive examples as negative all the time. Merely having "average" human beings tag examples doesn't guarantee perfect recall. Or to restate, sites that rely purely on manual moderation of content that breaks rules still has content that breaks rules leak through all the time.