Most people had a hard enough time telling the difference between man made fact and fiction, now they have to tell the difference between AI fact and fiction on top.
Well AI fact, in this use has always been made up of a combination of man's fact and fiction. Nobody's been smart enough to make an AI that can reliably separate the two, to my knowledge.
I believe the solution is curated data models with the top members of the applicable field determining validity or a stack overflow model.
I think you're on the right track here, but will retain the same flaws ultimately this way.
Personally, I believe the models should be open and all interested parties have varying degrees of influence over the accepted truth. That's going to be a complicated in itself.
By limiting it to "trusted people", you only have to corrupt enough of them, and eventually you end up with the same shitty problems, but with bots too.