CS academics have been slowly taking control of more of our own journals. It somewhat helps that there is already relatively little editing/layout work done on the journal's side, because computer scientists like to produce publication-ready PDFs straight out of LaTeX, and a journal is more or less just a concatenation of those. In some other fields it's more common to send a draft-formatted manuscript, which the journal staff will format into professional-looking magazine style; that requires more of a "real" staff, which then means you need a budget and source of funds and such. This is one reason some open-access publications in other areas are turning to charging the authors to fund the operations (although reputable journals will waive the fee for authors who don't have a source of funds such as a grant to cover the fee, so it's not fully "pay to play").

In AI, two of the top journals are now open-access and run on a shoestring budget by academics, so also don't have any publication charges: http://jmlr.org http://jair.org

Here was the 2001 letter that kicked off JMLR: http://jmlr.org/statement.html