This is a little naive. Running a good journal takes a lot of time and effort (not to mention a little bit of money).

In practice, paper selection is done in a bottom-up fashion already. An individual lab usually decides whether a paper is worth publishing. A minor conference decides which of the papers submitted to it are worth featuring. Bigger conferences look at the results of the minor ones. And so on. There is not a guy at Elsevier whose job it is to read the entire internet. Like the rest of their business, it's parasitic off the publicly funded university research system.

If you really think that no-one employed by journals is actually doing any work, I'm not sure what to say. The reality is that editing and publishing are time consuming, and that if this work is not being doing by paid employees of journals, it's going to have to be done by someone else.

Paper selection is not really the issue. That is largely done by (unpaid) reviewers in any case. It's everything else that goes into running a journal that's potentially going to give rise to problems.

Again, I'm not saying that it's impossible for a group of academics to get together and run a journal entirely by themselves. I think it's less clear whether this model can scale to replacing all current traditional journals. It's very easy to confuse work that's easy with work that's quick. Yes, it's easy to set up a website with a few PDFs on it, and it's easy to forward papers to reviewers, and it's easy to come up with a LaTeX stylesheet for papers, etc. etc. But maintaining all of that stuff and keeping it running smoothly eats up a lot of time, even though there's nothing fundamentally difficult about it.

Yes, it's less clear because we don't have that kind of system yet to see. We need to try it. Perhaps we'll find we do not need all current traditional journals. Perhaps we'll be fine with online archive for all academic publications and a good search engine.