The machine learning community (along with the combinatorics, category theory and many other communities) created an open access journal. While this is laudable it did not solve the non-trivial problem that commercial publishers have a lock on 60 years of scientific publications.
This really is a hard problem. I'm not sure this is right, but I think there's probably a case to be made that congress should seize the IP under eminent domain, write the publishers a check and tell them to go home.
It is possible that in one hundred years or so that many of the "most useful" results published in copyrighted journals of the 20th century and available nowhere else will have found their way into the public domain. But it need not turn out that way. Cory Doctorow spoke about the coming war on general computation at 28c3: http://www.youtube.com/watch?v=HUEvRyemKSg So we cannot be complacent.
