Again that level of AI/ML means being able to state what the intent of a comment or document is. If we ever get close to that point, that programs can reliably mark something as "risky" content, human thought is in big trouble.

"Risky" material is subjective. Something even humans can't make out reliably very often.

I think, this is the result of the tech firms putting content creators of old out of business, but not investing in the necessary editorial power, subscribing instead to the "wisdom of the crowds" school of thought.

They have not paid for (and will not be able to afford) the distributed human editorial power which was lost.

if you go to youtube for the corolla video, you see one ad, see the video and be done with it.

the average bla bla bla video instead has a much higher chance of driving multiple view per user via the suggestion mechanism, giving a multitude of views on each video and ads.

of course that strangles quality, focused content in favor of generalist episodic blogs/devs, but a single view video will earn them many times more than a video with a strong network effect trough past/previous episode suggestion, watch again, see this other blogger on the same topic etc.