Skip to content
TODD BROWN
« The Anatomy of Corruption and Corrigibility I — The Stance · Chapter 3

Three Kinds of Together: Grex, Collegium, Egregore

Just before sunset, a few hundred starlings lift off a stubble field and start to circle. Within a minute there are thousands. Watch from the ground and the flock stops looking like birds and starts looking like a single dark shape breathing over the fields. It folds in on itself, stretches into a ribbon, collapses into a ball, then pours sideways like smoke caught in a draft. A hawk cuts in low and the whole shape flinches away from it as one surface, a hole opening and closing around the predator's line of attack faster than the eye can follow the birds inside it. It looks choreographed. It isn't. No starling has seen the shape from outside, the way you are seeing it now. No starling has been told where to fly, and there is no lead bird calling the turns. Field studies suggest each bird is tracking a handful of its nearest neighbors, about six or seven, and adjusting to them over and over: matching their speed, closing the gaps, peeling away from the ones peeling away from something else. The shape you're watching is not a decision anyone made. It is what happens when thousands of small, local adjustments add up. The flock behaves like a single animal dodging a hawk, and there is no animal there. There is only the rule, running in every bird at once.

Now walk into a different kind of room. It's a physics seminar, sometime mid-century: chalk dust, a projector, forty people who have spent decades arguing with each other for a living. A junior researcher presents a result. Someone in the third row raises a hand and points out that the error bars on slide six don't match the ones in the paper. The presenter checks, admits it, promises a correction. Nobody's career ends over this. Nobody even raises their voice. Six months later a referee at a journal on the other side of the country catches a different flaw in the same line of work, and that gets fixed too, by someone who has never met the original author and never will. The correction gets published, cited, absorbed, and within a few years nobody remembers whose mistake it was, only that the number is now right.

Extend this room across fifty years and every face in it changes. The graduate students become the senior researchers become the emeritus professors become names in the acknowledgments. What doesn't change is the machinery: the seminars, the referee reports, the standing expectation that a wrong number gets found and fixed by someone who didn't make it, and that finding it counts as a contribution rather than an accusation. The knowledge outlives every person who ever held it.

And in your wallet, or somebody's, is a twenty-dollar bill. Cotton and linen fiber, a little ink, a portrait, a serial number. As a physical object it is worth approximately nothing. You cannot eat it or build anything out of it. It buys groceries only because everyone around you has agreed, without ever holding a meeting, to keep treating it as if it were worth groceries. Carry it far enough from the people who keep that agreement and it is worth only what someone there will give you for it, which may be much less. The agreement is the actual thing of value. The paper is just where the agreement gets written down.

Agreements like this can also end. Think of an online forum that was, for a few years, the place everyone in some hobby went: thousands of posts a day, regulars who knew each other by handle, in-jokes, a shared sense that this was where the conversation happened. Then people started drifting elsewhere, a few at a time. Nobody voted to close it. The servers kept running. The database still held every thread ever written. What went missing was the agreement: thousands of separate, ordinary decisions to keep showing up, reversing one by one until the place was quiet. Everything physical survived. The thing itself did not.

Three scenes. Three kinds of many-minds-behaving-as-one. A murmuration, a seminar room, a currency. All three coordinate large numbers of separate individuals into something that acts, from a distance, like a single system with a shape and a direction. And all three fail in completely different ways, for completely different reasons, which is exactly why lumping them together is a mistake. Before this book can talk about how collective systems go wrong, it needs three separate names for three separate ways they hold together.

Grex

Call the first kind a grex (it rhymes with "flex"). A grex coordinates through local rules, applied by many individuals at once, with no shared model of the whole and no deliberation about it. The starlings are a grex. So is a crowd leaving a stadium after the last encore. Nobody consults a map. Each person reads the backs and shoulders of the people directly ahead, and a slow, braided river of exits forms out of thousands of local guesses, with no one holding the whole flow in mind. So is a school of fish tightening into a ball when a predator approaches. So, in a slower and more abstract register, is a market price. Thousands of buyers and sellers, none coordinating on purpose, each acting on some sliver of information the others lack, produce a single number that responds to a total picture no participant ever sees. A viral post is a grex event too. Nobody plans that a video will be shared four million times. Each share is a local decision by someone reading their own feed, and the total is a wave nobody authored and nobody controls.

A grex computes without knowing it computes. That sentence is doing real work, so sit with it. The murmuration is solving a hard problem (stay close to your neighbors, don't collide, evade the hawk) and solving it well, in real time, with no central processor and no bird holding the whole picture. The direction the flock actually moves, its telon, is implicit in the local rules each bird follows. Nobody chose it and nobody could tell you in advance what it will be.

This is also the grex's limit. You cannot argue with a murmuration, because there is no one inside it to hear the argument. You cannot ask a market price to reconsider. You can only change the incentives that feed into it and watch what new price falls out. A grex has no capacity for reflection anywhere in it, which means it has no built-in capacity for self-correction either. Its characteristic failure is a very old one: the crowd that surges the wrong way, the stampede, the market that panics. A grex can be moved, quickly and completely, without anyone in it choosing to move it. Whatever a grex gets right, it gets right because the rules happen to be good, not because anyone is checking.

Collegium

The physics seminar is a different animal. Call it a collegium. A collegium coordinates through shared models, shared standards, and distributed oversight. Its members don't just read their neighbors. They hold roughly the same picture of what the group is doing and the same rules for judging whether it's being done well. Science at its best is a collegium. So is a functioning court system, where any single ruling can be appealed, reviewed, and overturned by people who didn't make it. So is a serious engineering culture, where a design gets checked by people whose job is to find what's wrong with it before it ships.

Here is the property that defines a collegium, and it is worth setting apart from everything else in this chapter: a collegium can be wrong, and it knows how to find out. Not "a collegium is never wrong." The opposite. Its members are wrong constantly. That's what the third row raising a hand was doing. What makes it a collegium, rather than a group of smart people, is that being wrong is a normal, survivable, expected event, with a built-in path to discovering it and a built-in path to fixing it, without the discovery costing anyone their place in the room. A collegium's authority does not come from its confidence. It comes from its error-correction. A community that never admits fault can look more impressive from outside than one that publishes corrections and retractions. It is in fact weaker, because its confidence is untested, while the other community's has survived real attempts to break it.

The same shape shows up in machinery that has nothing to do with physics or robes. Consider a large, crowdsourced encyclopedia that anyone with an internet connection can edit. On paper this should be chaos: no editor is vetted, no entry has a single accountable author, anyone can change anything at any time. What keeps it from being chaos is not the absence of error but a working correction loop. There is a citation standard that says claims need a traceable source. There is a visible edit history that shows who changed what and when. And there is a body of editors watching each other's work and reversing what doesn't hold up. It has real vulnerabilities. An edit war can bog down a contested page, and a small group of committed editors can end up with outsized control of certain topics. But on the whole it runs the same structure as the physics seminar: not an absence of mistakes, but a working, distributed apparatus for finding them. A stone tablet can't be corrected this way. A body of knowledge maintained by a collegium can be, continuously, by people who never met.

This property is not guaranteed to last, and it's worth flagging now, because a later part of this book lives here. A collegium can keep every visible piece of its structure (the journals, the review process, the boards, the credentials) while quietly losing the thing that structure was for. A field can keep holding peer review while peer review stops being able to overturn its central assumptions. The costume survives. The property underneath it doesn't. That's a problem for a later chapter. For now, hold the definition: a collegium is measured not by how right it is, but by what happens the moment it's wrong.

Egregore

The twenty-dollar bill points to the third kind. Call it an egregore. An egregore coordinates through shared belief, ongoing participation, and mutual recognition. Money is the clean case, because you can hold it in your hand and still see that the paper isn't the point. The agreement is. But egregores are everywhere once you know the shape: a nation, a brand, a religious tradition, a sports fandom, a language, an online community. None of these are physical objects with edges. All of them exist only as long as enough people keep participating as if they exist: singing the anthem, recognizing the logo, showing up to the game in the colors, using words the way other speakers will understand. A language, as the last chapter showed, survives the death of every speaker alive when you were born, carried by whoever picked it up next. Stop speaking it, enough people for long enough, and it doesn't get retired by a vote. It just goes quiet, the way the forum did.

The word has an odd pedigree worth clearing up in a sentence or two. It traces back to a Greek word for "watchers," used for a class of angels in the Book of Enoch, and in the nineteenth century French occultists took it up for a collective thought-form sustained by a group's shared attention. This book borrows the word for something un-mystical and fully observable: a coordination pattern, not an entity with its own consciousness floating above the participants. Nothing in this chapter asks you to believe in group minds. An egregore is not something over and above its participants, any more than a wave is something over and above the water. It is a pattern the participants make, together, by acting.

Here is the part of the definition the rest of this book keeps returning to. The constitution runs both ways. You sustain the egregore: your participation, your recognition, your willingness to keep treating the twenty as worth something, is part of what keeps it real. But the egregore also shapes you. A Marine's sense of who they are is not a badge sitting on top of an unchanged self. Thirteen weeks of recruit training rebuilt something load-bearing in there. A lifelong company man doesn't just work for the company. Some real portion of who he is runs through it. A devoted fan of a band that's been touring for fifty years isn't performing an identity at the show. For those three hours, and often well past them, that identity is the operating one. None of this is a flaw. It's close to how meaning works for social animals. We become who we are partly through what we belong to, and belonging that changes nothing about you isn't really belonging. File one question away without answering it yet, because a later chapter needs it: what happens when a system with this much power to shape a person's identity turns that power toward defending itself against being questioned?

Hybrids

Almost nothing in the world is purely one of these three. Real institutions are hybrids, and the useful move is not to force one into a single box but to ask which mechanism is doing the coordinating work for the question in front of you.

Take a software company. Ask how a production outage at two in the morning gets handled, and you're mostly watching a grex: an on-call rotation, alerts firing, whoever is awake reading the same dashboards and reacting to each other in a chat channel, with no one directing the whole response in advance. Ask how a risky change gets into the main codebase, and you're watching a collegium: code review, automated tests, a design document that can be argued with, reviewers whose job is to find the flaw before it ships and who can be overruled by evidence. Ask an engineer who has been there ten years what the company means to her, and you get something closer to an egregore's answer. Not a description of the org chart, but an account of a place she has given a decade to, which has become part of how she understands her own work: what kind of engineer she is, what she's for, who she'd be without it.

All three answers are about the same company, sometimes the same week. None of them is mistaken. They answer different questions, and the error to avoid is answering the wrong one: treating the on-call scramble as if it needed a design review, or treating a decade of identification with the place as if it were just a scheduling system that happens to issue paychecks.

This will matter later, so name it now. The question you're asking fixes which mechanism you're looking at, the same way the boundary of a system depends on the question being asked of it. "How does this institution allocate a resource?", "how does this institution correct its own mistakes?", and "what does this institution mean to the people inside it?" are three different questions. A single company, hospital, or nation answers all three at once, through three different pieces of itself running at the same time.

What This Isn't

It's worth being blunt about the boundary, because the vocabulary invites overreach. Nothing above is a claim about group minds, collective consciousness, or some spooky property that emerges over and above the individuals involved. This is a functional taxonomy: three different ways that separate people's behavior adds up to a pattern that acts, at a distance, like one thing. That's all. The starlings don't share a mind any more than the seminar room does. They share a coordination mechanism, and different mechanisms deserve different names because they behave, fail, and get fixed differently.

Not everything three people do together earns one of these labels, either. A filing cabinet full of signed contracts is not an egregore. The agreements inside don't need anyone to keep believing in them moment to moment. They sit there, enforceable, whether or not anyone is thinking about them. That's a difference worth marking: an egregore requires ongoing participation, not a one-time signature. And a handful of strangers at a bus stop, all facing the same direction, is barely a grex worth naming. There's a local rule at work (face the street the bus comes from), but the coordination is thin enough that nothing interesting follows from calling it anything. The typology is a tool for cases where the coordination is doing real work, not a label for every instance of more than one person in a room.

Why the Distinction Pays Rent

The reason to sort collective systems into three types, instead of treating "the group" as one blur, is that each type fails differently, each failure needs a different repair, and confusing them wastes the repair.

You cannot peer-review a murmuration. There's no author to submit a correction to and no committee to review it. The flock doesn't hold a position it could be talked out of. You cannot rally a collegium the way you'd rally a crowd. Showing up with more passion and a bigger banner doesn't move a body whose currency is evidence and standards, or shouldn't. And you cannot fact-check a currency into stability. A run on a bank is not resolved by a press release full of correct statistics, because the thing failing isn't a claim. It's an agreement to keep believing.

Reform efforts in every domain (corporate, civic, religious, personal) often fail by treating one type of system as though it were another. Someone inside a group organized around shared identity and belonging gets handed a carefully sourced spreadsheet of counter-evidence, and it changes nothing. The spreadsheet answers a question about facts, and the group is running on a question about who its members are and who they stand with. Someone trying to fix a broken engineering culture with a rousing speech gets applause and no change in the defect rate, because a collegium's failures live in its review process, not its morale. Bringing footnotes to a fight that is really about belonging fails the same way bringing a rally to a fight that is really about evidence fails: right tool, wrong system. Everyone involved usually blames the audience instead of the mismatch. A later chapter looks closely at what happens when a correction aimed at an egregore's participants gets heard as an attack rather than received as information. For now, notice that the mismatch is a common and avoidable failure on its own, well before anyone reaches the harder question of why a correction was rejected.

An Exercise

Pick three groups you currently belong to: a workplace, a family, a hobby, a faith community, a fan base, whatever comes to mind first. For each one, ask three questions:

  • What actually coordinates this group's behavior: local rules nobody discusses, shared standards everyone could name, or shared belief and participation?
  • If the group turned out to be wrong about something central to it, how would that get discovered, and by whom?
  • What would it take for the group to change its mind about that central thing?

If your honest answer to the last question is "nothing could," with no evidence, argument, or event able to move it, that's not proof the group is broken. But it tells you which kind of together you're standing in, and it's worth knowing that before you need to.

Close

Part I is now assembled. What a system does is where you look to find its direction, and what it is for is a separate question you answer from documents, not guesses (Chapter 1). What persists in an institution is a pattern, not a roster: the hospital survives every employee's departure the way a language survives every speaker's (Chapter 2). And collective systems hold together in three distinct ways, which means they come apart in three distinct ways too. That's the full kit: direction, persistence, coordination. What's next is failure, and not some rare, dramatic kind. The next chapter starts with the most common way a system with the best intentions quietly stops doing what it's for.

Sources

Contents · Glossary · How this was written