I start from the position that academic writing is primarily produced in order to add to the sum total of human knowledge and contribute to the conversation that constitutes science and scholarship. The author writes with the intent and desire for their work to be read (the pre-requisite to having their work understood, cited, have their experiments replicated or extended, have their theories confirmed or refuted, have their theses argued with or against) and/or the funder of the author and the specific work in question has that desire. (some authors only write because they are required to do so by funders or employers but I suspect they are rare in general, though individual pieces may be more commonly written to fulfil specific obligations).
Sometimes academic writers receive financial rewards for their writing, but except in the case of textbooks for students, such monetary income is tiny and while few refuse such supplements to their financial situation, the work would almost always be done without such rewards. These rewards are almost always for monographs in the humanities or social sciences. While access to these elements of the literature is important, it is more complicated than my focus here and hence is not included in this essay.
The target of this essay is the reviewed journal (and for some disciplines such as Computer Science the reviewed conference) literature. The salient factors of this literature is that the author receives no financial income from the "sale" of the work, neither when they provide it to a publisher nor when a reader obtains access to a copy.
At present, academics and a small percentage of others1 write papers, submit them to journals (or conferences) where their submissions are considered by editors and peer reviewers for some approximate measure of quality and if deemed acceptable by the editors, they are then officially added to the scientific/scholarly corpus. Professional publishers (mostly these days large commercial multimedia organisations, but also including some universities, scholarly societies and other NPOs and even governments) are involved in this process, though in most cases their involvement is limited to managing the administrative aspects of the system including the printing and delivery of hard copy of journal issues bundling multiple papers and more recently providing an online database of papers. They often perform typesetting, though these days that involves much less work than previously because the papers are required to be submitted in a suitable digital form which is automatically run through typesetting systems. These publishers charge large fees to readers through various mechanisms for access to these papers. These fees have increased well beyond inflation in the last twenty years and no institution and no individual2 can afford to preemptively purchase access to all the papers that they would like to consider reading.
Authors (and/or the people who legitimately require authors to write) produce their work for it to be read. Readers want to read it but cannot because middlemen charge too much. That is the problem. What is the solution? Readers need access to the authors text after it has been quality checked. This is not about changing peer review. This is not about library budgets directly. This is not about text-mining rights. The first problem we must solve is the basic one of access to the text. Only once that has been solved should be consider other problems.
1 Including but not limited to: workers in industrial/government/NPO research labs; retired researchers; spare time "amateurs", in the pure sense that they are just not paid for doing the research or writing the papers.
2 Barring someone like Bill Gates.