I am optimistic about human in the middle. I imagine a large grid of small thumbnails and the Google Captcha style "Choose the pictures that are not gruesome" (i.e. differentiating between violence, a video game, and a red scarf). Many reports could be resolved quickly this way. I would be interested in a more proactive solution that is well documented publicly so that we can apply the tech to blogs/news.