I think you underestimate the extent to which companies try to do both. I know Twitter and Facebook both spent big for years and years on increasingly sophisti

I think you underestimate the extent to which companies try to do both. I know Twitter and Facebook both spent big for years and years on increasingly sophisticated ways to automatically recognize problematic content. Even they couldn't afford to just throw labor at the problem; a few years back when I ran the numbers Twitter's revenue was about $1/user/month, or about ¢1 per tweet. No matter where you outsource too, that doesn't buy much labor. If you look at the manuals for the outsourced moderators [1], it's clear they were being asked to work at a pretty sophisticated level. That level certainly wasn't available from ML systems before a couple of years ago, and I'm skeptical that it's there now, given that many of the hard moderation decisions come down to contextual judgements, and given how good awful people are at innovating their way around automated detection systems. [1] which you can do here among other places: https://www.theguardian.com/news/series/facebook-files https://www.theguardian.com/news/series/facebook-files