Current systems are indeed very vulnerable to professional manipulation - there is no real defence yet - and there are powerful players. But democratic sampling won't help much. Only the wrong guys are interested in voting - and once mass hysteria has set in any democratic majority will vote to censor anything that brings them out of their panic loop. Witch hunting lasted for hundreds of years.

I think just tagging things accordingly would be a lot better than raw censorship. In good old places of Usenet just tagging things as Spam worked quite well. Just filtering out some tags and putting some guys in a kill-file was good enough. But it required manual labour - and eventually that was too much. But with AI now I think tagging could be done efficiently.

If people like to filter out all the tags (sarcasm, math, physics, ...) they can have it - but the way how things work now is that a lot of important information just gets censored by stupid people everywhere. Just hiding information from everybody is quite harmful - being seriously uninformed already killed a lot of people...