There was a proposal in Nature last week from David Eagleman and Alex Holcombe, suggesting that public debate and criticism of published scientific work should be attached to the online articles (or their representations in a database). If you're a user of HubMed, you'll know that the 'References' link attached to each article contains forward links to weblogs that have sent Trackback pings, and a comments form for users to attach their notes on each article (as well as the standard citations and reference lists).
Clay Shirky and Derek Lowe both picked up the article, with Clay presenting it under the guise of social software, which it is - the more users the better. As with most social software, there's a need for moderation. This is missing from HubMed as I haven't the time or the money to build a moderated system, though it would be fairly straightforward to do so using one of the existing models like Slash or Scoop (see biologging for the first tentative steps in that direction, though I'm waiting to see what the semantic blogging people come up with before making a commitment).
As John Vu pointed out, what the authors of the proposal have missed is the increasingly widespread use of weblogs for collecting and annotating online scientific texts. With Trackbacks between blogs and a central database with unique identifiers for each article, the idea of a centralised annotation datastore becomes less than ideal. The problem of identity is also solved by distributed annotation, and this should become more standardised in the near future.
An equally important point is still the problem of access to the literature we're supposed to discuss. Imagine how empty or ill-informed the discussions on Slashdot would be if everyone had to pay $30 to read the information being discussed (there's a problem with that analogy, obviously). John put this well: "individuals with limited access to the Internet and researchers without the funds to subscribe to all their relevant journals will be left out in the dark. Think about it, we have to have access to the papers to read prior to placing comments or reading other people's comments -- it's like going to journal club without having read the paper."