Inspired also by the series on algorithm regulation on this blog, I am currently working together with Claudia Müller-Birn on the issue of algorithmic governance in the case of Wikipedia. In the course of this research project, I stumbled upon the case of flagged revisions/sighted versions, which very nicely illustrates the concept of algorithmic governance.

With Wikipedia Germany taking the lead in 2006, some Wikipedia language versions introduced sighted versions of articles as a measure to secure against vandalism and improve article credibility. The concept is described in the English language Wikipedia as

a system whereby users who are not logged in may be presented with a different version of an article than users who are. Articles are validated that they are presentable and free from vandalism. The approved versions are known as Sighted versions. All logged-in users will continue to see and edit the most recent version of a page. Users who are not logged in will initially see the most recent sighted version, if there is one. If no version is sighted, they see the most recent one, as happens now. Users looking at a sighted version can still choose to view the most recent version and edit it.

Since its introduction in the German Wikipedia, the concept has evolved in a complex set of rules determining how Wikipedia edits are sighted. The core idea is that registered Wikipedia editors automatically receive the status as a passive or active “reviewer” depending on their respective editing history. Edits of a user with the status of a passive reviewer are automatically considered to be sighted; active reviewers have additional rights such as actively marking versions as sighted or removing the respective status.

To become a passive reviewer in the German Wikipedia, the following criteria have to be fulfilled (shortened and translated by myself):

  • The user is registered for at least 30 days.
  • The  user has at least 150 edits in the article namespace or at least 50 edits, who were later sighted. Deletions do not count. Edits of the last two days do not count.
  • There are at least 7 edits of the user with at least 3 days time in between
  • At least 8 different pages in the the article namespace have been edited.
  • In the course of at least 20 edits the edit summary was used.
  • The user has never been blocked

The rules for becoming an active reviewer follow a similar logic, albeit setting higher standards that have to be met. In addition, editors not (yet) meeting the criteria can apply for getting reviewer rights.

I find several aspects of the complex set of rules around the issue of sighted revisions in Wikipedia interesting. First, the whole system works mostly automatically, guided by a software algorithm. Many if not most editors receive reviewer status without having to know anything about the concept of sighted revisions. Comparing such “silent” with more explicit forms of algorithmic governance seems to be an interesting avenue for research on the issue.

Second, the criteria considered by the algorithm are transparently spelled out and have – as can be seen in the revision history of the respective wiki page – been subject to recurrent adaptations. In other contexts such as, for example, the slashdot community, making algorithmic regulations transparent has led to an increasing number of attempts to game the system. Kraut and Resnick (2011: 58) have therefore suggested that “nontransparent eligibility criteria and unpredictable reward schedules lead to less gaming of the system than do predictable rewards.” Wikipedia is therefore an interesting case for researching the question, when and how transparent forms of algorithmic governance work.

Third, a proposal to introduce a system similar to the German one within the English Wikipedia has been rejected by the community, pointing to the potentially controversial nature of algorithmic governance in online communities (see the more than lengthy discussion pages on the issue). Comparing debates around the proposals to introduce the system in the German and the English Wikipedia might provide insights into the historically and/or culturally contingent nature and acceptance of algorithmic governance measures.

Taken together, I am optimistic that we have a promising research agenda ahead of us.