Problem is that AI has no “common sense” and there is a whole lot of garbage mixed in with any actual usable knowledge on the internet.
A human can sift the garbage out while he’s synthesizing, but even most humans don’t do that very well.
Yes, there will still need to be a human in the loop to do quality control so to speak. But even that can probably be achieved algorithmically to a substantial degree, like by, say, looking for indications of consensus. For instance, if your bot scans through a thousand postings and articles in the area of basic math and find that most of them contain the expression “2 + 2 = 4” with just a few outlier variations, then maybe it would be reasonable for it to treat 2 + 2 = 4 as good information.
What we are sorely in need of is artificial common sense, and lots of it ;-)