Self-Organizing

What I have learned about self-organizing communities, I learned from the inside. Not as the person who creates them — that is rarely how these things actually start — but as one of the people who joins, finds substantive work to do, earns trust along the way, and eventually finds themselves part of the small group that decides whether the effort sustains long enough to become something durable.

The interesting work is rarely the moment of ignition. It is the months that follow, when the initial urgency fades and internal pressures threaten to fracture what was built — disagreements over leadership legitimacy, exhaustion, ambiguous authority, the gap between what volunteers can do and what the situation now demands. That is where most self-organized efforts dissolve. It is also where I have repeatedly found my role.

I call this self-organizing rather than collective action because the distinction matters. Collective action implies an existing movement, an organization that convened the group, a prior social infrastructure. None of that was present in the moments below. The community did not exist as a community until the right signal named the problem — and then organized itself before anyone asked it to.

What I keep noticing

Once is a story. Twice is a pattern. Three times is something worth naming.

1. A public signal names an unaddressed problem and surfaces a latent community.
2. Anonymous, distributed self-organization produces immediate outputs faster than any institution could.
3. Socioemotional bonds form during the urgent phase — these are the load-bearing structure, not just a byproduct.
4. Internal pressures arrive — leadership legitimacy, alignment, exhaustion — and most efforts fail here.
5. If the group survives the internal pressures, durable outputs follow: a nonprofit, a peer-reviewed paper, a platform.

My contribution is overwhelmingly inside steps three through five. I am not the person who sends the signal. I am the person who shows up, makes themselves useful, and helps the community navigate the unsexy organizational and socioemotional work that determines whether anything durable comes of it.

Guildmaster

EverQuest, then World of Warcraft

The honest origin of all of this is online gaming. I was a guildmaster in EverQuest and World of Warcraft. In WoW I led 40-person raids through Molten Core, Onyxia’s Lair, Blackwing Lair, and Ahn’Qiraj Temple — content that genuinely required forty people who had never met in person to coordinate complex sequences under time pressure, with the difference between success and failure measured in minutes of synchronized execution.

The technical objective in those raids was the loot and the boss kill. The actual work was something else: keeping forty volunteers showing up week after week, mediating disputes about loot distribution, navigating burnout cycles, replacing officers without losing the group, holding the social fabric together when the urgency of progression dropped. Nobody trained me to do that. The lessons were earned the way they always are — by watching guilds I admired collapse under entirely solvable internal pressures, and trying not to let mine do the same. By the time disaster response and academic governance asked the same questions, I had been answering versions of them for a decade.

Saying so up front is partly to humanize the story. The capability I am describing did not arrive in a credentialed package. It came from how I actually live.

CRHQ → CEDR

Disaster response, 2017–present

In late August 2017, Hurricane Harvey was hours from landfall near Houston. Anyone watching the forecasts could see that emergency systems were going to be overwhelmed — not if, when. There was no organization standing by to handle the overflow. There was no protocol. There was a gap, and the gap was visible to anyone paying attention.

Someone else put out a tweet. I joined it as one of hundreds of pseudonymous volunteers who arrived in Slack within hours. They came from nowhere in particular — tech workers, data people, concerned citizens who had seen the tweet or seen someone retweet it. Nobody appointed them. Nobody told them where to go. They found each other because the signal was specific enough to name the problem and broad enough to reach everyone who cared about it.

What followed was remarkable. The group jointly leveraged Google, Tableau, and ESRI tools at no cost, built a triage workflow from scratch, and began human-reviewing rescue requests submitted via Twitter — real people reading real distress signals, verifying them, routing vital information to volunteer disaster responders on the ground. Faster than any institution could have assembled a response team. The effort scaled through Hurricanes Irma and Maria, developed coordination relationships with NAPSG and Firemappers, and eventually pivoted to wildfire response. CrowdRescue HQ formalized into CEDR Digital Corps, a 100% volunteer 501(c)(3). I currently serve as Co-Chair of its Board of Directors.

I joined as a volunteer. I found a role manually geocoding the difficult-to-resolve addresses — the ones algorithms could not place, where someone had to read the original message, parse it against local landmarks, and produce a usable point. Tedious, careful work that was directly load-bearing for rescue routing. From there I led the GIS team. From there I was asked into the small group that helped the effort transition into a formal nonprofit.

Trust in that environment was not granted on credentials. It was granted incrementally as people watched each other handle hard work under pressure. There was, fairly, internal skepticism of leadership — volunteers had seen the worst of bureaucratic emergency response and were not eager to recreate it. The work I ended up doing, more than anything technical, was helping alleviate the organizational and socioemotional pressures that build up in a distributed group operating at the edge of its capacity. Keeping people aligned. Surfacing disagreements before they fractured trust. Helping the group remember why it was doing what it was doing when the urgency was no longer the same.

What CRHQ was doing in 2017 — human-in-the-loop curation at scale, trust calibration with anonymous contributors, socioemotional infrastructure as load-bearing — predates “participatory AI/ML” as a named research area. Volunteer fatigue, distributed authority, the calibration of trust between strangers doing high-stakes work: these were worked out first in disaster response, under real conditions rather than laboratory ones. That shapes how I think about what participatory AI/ML actually requires.

ICML Position Paper

AI research community, 2024–2025

Seven years later, the same pattern surfaced in a completely different domain.

A small public critique of “AGI as the north-star goal of AI research” began to attract responses from researchers across institutions — people thinking about adjacent problems, some of whom had never met. About fifteen of us self-organized around a shared concern. No conference convened the group. No funding agency organized it. People found each other because the critique named something they already knew was true but had not seen named publicly.

The result was a peer-reviewed ICML 2025 publication: Blili-Hamelin, Graziul et al. (2025), “Position: Stop treating ‘AGI’ as the north-star goal of AI research” (arXiv:2502.03689). The paper identifies six specific failure modes in AGI discourse: the Illusion of Consensus, Supercharging Bad Science, Presuming Value-Neutrality, Goal Lottery, Generality Debt, and Normalized Exclusion. It is, in miniature, an academic community performing its own version of self-organization around a shared concern — distributed across institutions, accelerated by the urgency of wanting to say something before the discourse calcified further.

Why IDEP

The synthesis, 2024–present

The earlier moments are evidence. The Illinois Data Equity Project is the bet that follows from the evidence.

The pattern only works under conditions you usually do not get to choose: a hurricane, a discourse failure that finally reaches a tipping point, a guild full of people who happened to be online at the same time. The communities most exposed to data-driven systems do not get to wait for those conditions. They need durable infrastructure that lets a latent community become an operational one without first requiring a crisis.

IDEP is what I am building because I have lived through the pattern enough times to know what makes it survive. Its purpose is to develop and propagate the resources for communities to form data stewardship organizations that self-organize, self-govern, and self-determine how to advance their collective and individual data rights. It is the institutionalization of what I learned by being inside the earlier moments — the trust work, the post-urgency repair, the alignment maintenance, the willingness to do the unsexy parts that determine whether a community-led effort survives long enough to matter.

A capability, not a coincidence

What I have learned by doing this in different rooms is that the work which makes self-organized efforts survive is rarer than the moment that founds them, harder to fake, and almost never visible from the outside. It is also exactly the kind of work that determines whether community-led data governance becomes a slogan or an institution. IDEP is the bet that this kind of work can be made into infrastructure.