For the past year, after reading in the Wikipedia Signpost newsletter on multiple occasions that the number of active Wikipedia Administrators has been in relentless decline since 2008. I was hoping that the final report from the foundation research effort that began last summer would lay the foundation for bold action. And, while the report does confirm that several large Wikipedia projects (i.e. languages) have a problem recruiting new admins, the recommendations are mostly confined to areas for additional study.

My feedback will proceed as follows: identifying what I believe is the most concerning problem from the report’s own data, articulating the potential risks to projects from administrator decline, discussion of technical tools that could reduce the risks posed by inexperienced and malicious admins, discussion of training frameworks that could improve potential administrator supply and streamline the RfA process.

Pipeline problems

From the quantitative analysis on Administrator tenure, it’s clear that most of the current active admins have been doing this job for more than 15 years. On a purely demographic level, this should be setting off alarm bells. Without a steady stream of new recruits, this pool of volunteers is almost certainly facing the potential for a accelerating decline.

As with problems that economists are beginning to discuss regarding falling fertility and, correspondingly populations, there is good reason to believe that the decline in administrators could accelerate. One obvious cause would be the rise in administrative workload required per admin.

The report does not acknowledge the degree that present administrators are involved in the RfA process, and that, in the absence of a more formal requirements specification, the set of administrators acts more as a club than a professional practice.

Project Capture - Why it Matters

Wikipedia, especially the large English Wikipedia, has a difficult to quantify role in powering current technological productivity. With large language models often using the comprehensive and readily shareable content as the factual training material, it would not be unreasonable to suggest that trillions of dollars of economic activity depends upon the project. The damage that could be done by a rogue group of admins working in concert is substantial.

And, this is not a theoretical problem. The 2021 report on the compromise of the Croatian Wikipedia demonstrates that the risks are real and poorly quantified. Because there are substantial economic benefits to Wikipedia manipulation, and the difficulty in detecting COI edits, especially edits performed using generative models, it seems important that the administrator pool have explicit targets based upon expected workloads.

Many wikipedians consider the creation of commons goods to be a movement. And, if so, the decline in administrators is a warning sign that the movement is faltering. At a time when so many public institutions, some storied and long lived, are finding themselves compromised by internal or external forces, it seems unwise to expect worrisome trends to work themselves out. Just consider the damage done to a 150 year old organization like the National Rifle Association by corruption, or the collapse of the Green party in the United States. In both cases the failures were caused, in part, by people who were ideologically aligned with the goals of the organization.

In short, if the organization is a movement, and it is not growing then it’s probably in trouble. Some have argued that, as the project matures and the need for new policy creation declines, so perhaps the decline in active editors and administrators is not a concern. This perception - that the project is basically done - is perhaps even more damaging to recruitment. Being an administrator is a difficult task, as is amply documented in the report. If there is a loss of enthusiasm for the ongoing work of Wikipedia, why would people step forward to take this responsibility? I think the foundation could do a lot more to encourage people to become contributors - my own back of the envelope estimate is that the English Wikipedia is 20% complete - but that is a topic that is beyond the scope of the present discussion.

Reducing Risks from Bad Admins

The primary reason that the RfA process is arduous and exhausting is because admins have such powerful control over Wikipedia content and users. Broadly speaking the administrative controls can be broken down into page actions and user actions. While it is important that administrators be able to act quickly to address vandalism or distributed attacks, the most important, and often controversial actions involve reading complex arguments and resolving disputes. These types of actions are already subject to the AARV process, and almost all administrator actions are subject to easy reversal by other administrators.

The problem of distributed administration of large systems is hardly unique to Wikipedia. There are a lot commercial organizations that administer even more complex systems (or, sometimes, fail to) that could provide lessons. But I will just list some relevant ideas:

  • finer grained admin ACLs that can allow people to progressively become admins
  • provisional admin status, subject to automatic review
  • multi-party authorization for admin tasks that are not time sensitive
  • improved monitoring for anomalous administrative activity

Overall, the important point here is that AfD would be less fraught if the power granted to administrators was lessened. While the topic of improving automation use to reduce the administrative workload, it more likely than not that cleaning up the busy work would leave only the most exhausting task: dispute resolution.

Building up governance infrastructure on the mediawiki platform has always been a difficult process both socially and technologically. But I would have liked to see the report on Administrator retention to spend more time on potential engineering solutions, even if they are speculative.

Formalizing and Mentoring

Administration on Wikipedia is a job, and the foundation would most benefit from treating it as such. Jobs have requirements, and for Wikipedia demonstrating familiarity with the project and foundation’s policy framework is an obvious requirement. At present, candidates have to prove their familiarity with these policies by citing past on-platform instances where they had interactions involving them. This is not the only way to demonstrate knowledge, and developing educational materials, distinct from the policy documents, to train people in dispute resolution is entirely possible.

Having read through several RfAs, especially the withdrawn or failed RfAs, one is struck by the unforgiving regard that admins have for people who have made mistakes in their past interactions. The general tone is that the candidate is perhaps not the “right kind of person” for Adminship. In this regard, Wikipedia seems to be copying many of the biases and shortcomings of the corporate recruiting system. For an organization founded on open principles, it’s more than a little disappointing that more isn’t done to foster candidates through mentoring.

I think there are lessons to be learned from the readability process used at Google and documented in the Software Engineering at Google book. This program is a formal mentoring process where new employees are required to obtain additional code reviews from language domain experts until the demonstrate mastery of the company’s particular policies and practices.

Wikipedia has been slowly rolling out mentorship programs, including featuring them on the relatively new homepage feature. And, while mentioned there does not appear to be a formal Administrator mentoring program.