"You've watched it, you can't unwatch it" is not just a funny quip from Futurama, it's also real-life. The only real way for this sort of content to be reduced is for finer screening of user accounts in signup, which is contrary to the business model of frictionless signup and subsequent steps towards the abyss / whirlpool of social sludge. Another way is to distribute this task among the many accounts, but then you just give PTSD to everyone for not a lot of benefit. A third way would be to pay people to look at obscene content, but you can also tell them this and not lie that they're doing translation, like it said in the article for this particular gentlemen. In theory the image recognition software is able to see this better and better, to avoid humans needing to inspect, but if something is contested or appealed again human eyes get involved. How to observe without being altered, oneself? Impossible, I say.