 Hello everyone, my name is Zarin Karazin, I'm a PhD student at the University of Washington in the Department of Human-Centered Design and Engineering. Today I'm going to be presenting research done with my advisors and co-authors, Benjamin Mako Hill and Kate Starbird on governance capture in self-governing communities, a qualitative comparison of the CERBO correlation Wikipedia's. Some background. For about a decade, at least 2011 to 2020, a small group of far right editors took over the governance of the Croatian Language Edition of Wikipedia. It's meant that they installed themselves and their allies in administrative positions by rigging local elections, and they systematically started introducing far right bias, including fascist revisionism about World War Two into the project's article base. They also harassed and stamped out dissenting editorial voices on the project. After years of calls from local editors to the foundation to intervene, the situation started getting some media attention. Eventually, through the work of the local editing community, the offending editors on the project were removed. The Wikimedia Foundation published a retrospective assessment of the situation and confirmed that indeed this was a case of project capture where the project's mechanisms of governance were taken over by a few users in powerful administrative roles. The interesting thing about the Croatian case is that CERBO Croatian is a plural-centric language, and there are actually four separate language editions. In addition to the CERBO Croatian, there's the Serbian, Croatian and Bosnian. But despite similarities among these language editions, Croatian alone that we know of succumbed to this sort of capture, all of these projects to some extent faced vandalism and disinformation campaigns and things that are challenges for all language editions, but none of them quite had the takeover of the project's entire mechanism of governance that Croatian Wikipedia suffered. So the question that arose was why? Why did Croatian Wikipedia to succumb to the sort of systematic disinformation campaign when other language editions in the same language family did not? To answer this question, we conducted 15 interviews with participants who were familiar with the situation on Croatian Wikipedia. They were editors and admins in the local language communities, as well as contributors that were involved in global governance issues. This included stewards, as well as foundation staff members dealing with trust and safety. And based on our grounded theory analysis of these interviews, we arrived at these three propositions. The first was perceived value. Is the project worth expending the effort to capture? On this we found that Croatian and Serbian Wikipedia, the two projects that we compared, were roughly similar because they had both the sort of editing community where nationalist narratives would resonate because they were both tied to a particular national identity. And they were of a sufficient size that they had an audience that would be attractive for any sort of motivated actors to target with these sorts of narratives. But the projects then differed on in two respects that made Croatian really vulnerable and Serbian comparatively more resilient. The first was bureaucratic openness. How easy is it for contributors outside the core founding team of users to ascend to these local governance positions on the project such as becoming an admin? And we found on Serbian Wikipedia there was a deliberate attempt at the beginning to keep the project open, make it pretty easy for people to ascend to these positions, and that created a relatively diverse initial administrative team. Croatian instead became more and more bureaucratically closed relatively early on, which kind of created this core group of users that had pretty powerful positions on the project and also created an environment that was not very transparent bureaucratically. The second dimension on which these two projects differed was institutional formalization. To what degree did the project preference personalistic informal forms of organization over formal ones. So through our interviews we found that Croatian Wikipedia a lot of the actual governance discussions and work was done off wiki in a very non transparent fashion. And the primary way that people on the project ascended to these sorts of governance positions and that decisions were made about governance was through these personalistic informal forms of organization rather than more formal ones such as rules about admin power and rules about when users can block and when they can't. Particularly admins. So we synthesize these propositions into a conceptual model of risk that represents the varying levels of vulnerability that various Wikipedia communities have to capture along these different institutional configurations. So according to this, the projects that are actually most vulnerable to the kind of capture that we saw in Croatian Wikipedia are projects that have a really insular bureaucracy, and have personalistic institutions, and are relatively high value so particularly ones that are, for example, tied to a national identity. Some quotes from our participants. Croatian Wikipedia. A few participants described the system of governance on Croatian Wikipedia is very much like feudalism one former editor on the project said we have admins who behave like medieval Lords and we are basically serves. Serbian Wikipedia editors on that project described it rather differently. They explained that Serbian Wikipedia was is a bit larger the editing community is a bit larger than Croatian Wikipedia but they also have a little bit more diversity and polarization among different factions and there's not really any had ever gained an upper hand over the other so power on Serbian Wikipedia is was much more balanced than on Croatian. So what does this all mean for the over 300 language editions on Wikipedia. Well, the goal here is not to design power out of these socio technical systems. I really like this quote by James Grimman Grimman for example that highlights that doing so is actually impossible. He wrote there is no way to redesign the technologies of social software so that technical power disappears for the reason that it is the social power that gives the technical power its bite. But this study shows that both social and technical power can be checked, not necessarily by particular technical design choices but by these certain governance arrangements. And so a lot of focus on there's a lot of focus within the Wikimedia Foundation, as well as social media companies and other online communities on developing automated tools to help deter things like vandalism or fix disinformation issues. But these tools are rather limited in that they don't address these more fundamental challenges to knowledge integrity that are rooted in the that attack the very governance of these systems or rather their mechanisms of self governance. So more deliberate attention is needed to foster government governance arrangements on these projects that support more democratic participation and decision making. So future work in this for this project is going to involve conducting a quantitative study that empirically tests hypotheses derived from our three propositions across the hundreds of language editions on Wikipedia.