I wish citing a paper is as easy as using apt in Ubuntu or importing a module in Python. We need an open access journal. Open source too. Where the papers and their associated datasets are hosted as well. That way we can just fork them if we think of something more interesting to do with the original research.
I was doing a Cost Benefit Analysis project for a local non-profit that had a few working economist in their research department. When I presented by work they were mildly interested in my results, but had a ton of questions about my source material, which were mostly academic papers that a mid size non-profit couldn't hope to buy access to. The fact that there were reams of pertinent papers on energy economics which working energy economists had no ability to read seemed a horrible waste of research and dollars.
As is always the case for policy change, it would require lots of people arguing the case to the relevant government agency, as money is more important these than the general good.
Thankfully the NIH is instituting policies that any research performed with NIH funds must become open within a certain amount of time (I believe it is several months). That said, I think these paywalls will fall in time. If other authors cannot read my work, they won't cite my work. Getting your work cited is an important component of a scientific career. It's very advantageous to get published in a journal like Nature, which every institution carries. However, at least to me, middle level paywall journals are becoming much less attractive when compared to http://www.plosone.org.