Reading · The Paper Pilot's Digital Garden

Cybernetics is the science of organization, and the field of study - while not explicitly anarchist nor do the most notable cyberneticians identify as anarchist - aligns well with anarchist goals and values, and can help solve the scalability problem of anarchist organization. What follows is how to make an equitable large-scale organization that, due to its dynamicism, is viable in a complex world. It will discuss the structure and governance of such organizations, and tools to decide upon and enforce formal rules within the organization.

Anarchist Cybernetics goes in more detail on what cybernetics is and its relation to anarchism, including a reference to how our own brains are structured not dissimilar to anarcho-syndicalism. I’ll be applying the principles discussed therein, like the VSM, but won’t be exhaustively explaining them.

I hope the ideas here can help readers structure their organizations, present and future ones, with better governance than the “benevolent dictator for life” model oft defaulted to today. It might also help give an idea to how large-scale projects like rail networks and power grids would look under anarchism.

Cybernetics

Ashby's Law of Requisite Variety states that an organization must have as much variety as its environment in order to respond effectively. Which is to say, an organization must be able to adapt well to survive in our highly complex world. Cybernetics tells us we can achieve the requisite variety using many autonomous units that each have local decision-making power (operational governance) informed by their local knowledge and feedback from the environment. Only truly constitutional governance, like the values guiding the entire organization, necessitate the coordination of large parts of the organization.

This adaptability is useful not only for making a stable and coherent organization, but for fulfilling the vision of prefigurative anarchism. It also means if we can define a general structure that works well enough to have these effects, it can then handle the context-specific specializations the organization requires. That’ll allow this page to give broadly applicable advice for many governance contexts, from open-source software to a casual discord server.

Note that cybernetics doesn’t suggest hierarchical organizations can’t also be stable - they demonstrably very much are, at least for a long time - but the point is that cybernetics shows we don’t need a hierarchy to be viable. Indeed, Beer wrote about attenuating the environment’s variety as a way of solving Ashby’s law, and he had ample experience applying cybernetics to quite hierarchical institutions.

Reputation

Since the advent of easily accessible generative AI, organizations like open source projects or niche gaming communities like the incremental games community have had to deal with how to handle a sharp increase of newcomers that are much less likely to become long-term parts of the community. This is something game theory can help us with: The tit-for-tat strategy in dyadic games as one that is nice (won’t defect first), provocable (defects once defected), forgiving (co-operates after defection stops), and clear (doesn’t act unpredictably). There is also a generous variant that is more forgiving for noise (like the messiness of human social interaction). This strategy has shown to be remarkably successful in a variety of situations, including those with bad actors. Nicky Case’s Evolution of Trust demonstrates why this strategy performs so well.

It doesn’t apply generally though. For n-person games, you need some level of “indirect reciprocity” - a reputation system. Further, if we want to handle the aforementioned cases of an influx of participants who are unlikely to have repeated interactions - which is essentially an unintentional Sybil attack - we need to add friction to the initial interactions. And I am deliberately excluding identity/personhood verification from our solution space. I don’t want to attach our governance to any external institutions that have their own hierarchical issues - so money, government identity verification, and the like - unless absolutely necessary, so we need a local reputation. Ideally this should be resolvable to a binary “in good standing”: Overly stratifying the organization across trust levels reintroduces hierarchical power.

Keeping reputation local also keeps it within a handful of non-strangers, where we can be forgiving, and use generous tit-for-tat if not full-on restorative justice. We do not need to keep people’s grievances globally legible, with all the surveillance, centralization, and mistrust-building that comes from that. In the next section we’ll be discussing rare moments of constitutional governance that require large parts of the organization - determining who is involved with that should not involve a global reputation system, so instead should be the aforementioned “in good standing” flag, which each autonomous unit can determine the criteria for itself.

Governance

As an anarchist, I’m opposed to even direct democracy, which is still majoritarian rule, and want to rely entirely on consensus-based decision making. Consensus is often criticized as not being feasible on large scale, but it can be by using many autonomous groups with recursively local operational governance. Ultimately, operational concerns do not need the full consent of the entire organization, and keeping them local to the ones it directly affects keeps the entire organization adaptable and running smoother.

These operational concerns also line up with the first system under Beer’s “Viable System Model”. The higher systems deal with constitutional concerns that are increasingly more deserving of the time and effort that goes into workshopping consensus solutions - anything demanding urgency becomes a de facto operational concern. The systems also recurse, meaning each System 1 can be considered its own full Systems 1-5 one layer down. This means additional layers can be utilized to ensure decisions stay as local as possible and thus keep consensus feasible. For the truly org-wide decision making, in orgs where consensus is not logistically feasible across all members in good standing, then autonomous groups can workshop the bounds of their position and send a delegate, who can re-convene with the group should there be a novel proposal. Naturally, delegates should be messengers and not decision makers, and rotate so as to resist ossification of a delegate-class.

Implementations

Let’s take a look at how applying these ideas would look in a couple common types of large-scale organizations.

Infrastructure

I’d like to start with infrastructure - like highways, rail networks, and power grids - because they’re pointed to regularly when discussing how anarchy doesn’t scale. I’ll also apply the VSM more thoroughly to this example than I did when discussing organizing more abstractly.

System 1 is straightforward - each part of the network (e.g. train stations, power generation nodes, interchanges, etc.) is an autonomous unit that performs local decision making.

System 2 is operational coordination across these units, which happens through automatic control loops, protocols, and schedules set by higher systems. This is already largely automated today and does not require any human-level governance on this level.

System 3 would involve large sections of the network and only be employed rarely. It would handle things like scarce resource allocation, and once again should workshop ideas until a reasonable solution reaches consensus.

Systems 4 and 5 would be infrequent assemblies to make constitutional changes that affect how the lower systems operate. Ideally these too would utilize consensus.

Businesses

Igalia is a fully worker-owned cooperative consulting firm based in Spain whose internal organization maps cleanly onto the VSM, and is therefore a great real-life example of these ideas working in practice. They have no CEO or hierarchical governance, keep operational concerns local to small teams, and use full-Assembly consensus for org-wide concerns. It’s a firm of well over 100 workers and over two decades old.

Open Source Projects

Already most OSS projects have a class of maintainers who are roughly equal in power and can act autonomously, but when it comes to constitutional governance a lot still defer to a head maintainer being the “benevolent dictator for life”, (BDFL) which has led to many cases of drama within the wider community. Often the BDFL cites the ability to fork the project to avoid adopting a more egalitarian governance structure. I believe it’ll be easier to accrue and retain maintainers by switching to constitutional changes requiring consensus from all maintainers.

Beyond that, OSS projects also need a way to prevent drive-by PRs (especially AI generated ones) from those unlikely to be recurring contributors from using the scarce time of the maintainers better spent on contributors who may promote to maintainers themselves. This is something the community is actively experimenting on how to do in ways that still support onboarding people likely to stick around as maintainers.

Finally, it should become easier to become maintainers, by subdividing maintainers into groups that have purview over specific aspects of the project and the autonomy to promote contributors into maintainers within that group without needing to coordinate with the other groups.

Discord Servers

Discord servers are similarly structured to OSS projects: a BDFL, a few appointed moderators, and everyone else. Similar to before, the first step should be making constitutional changes go through consensus between all moderators. Then, as before, splitting the moderators up more - perhaps per-channel group, or even per-channel depending on the exact structure of the server. Each group of moderators would handle operational governance for the part of the server they look over, including promoting chatters into moderators. All regular chatters should become moderators of the sections they’re regular within. Per the forgiving tit-for-tat mechanism, conflicts should be treated with forgiveness, and only after repeated violations should things be brought before larger groups of moderators for more severe actions.

Due to the limitations of discord, server-wide governance like adding emojis are really more operational than constitutional. To account for this, having channels for things like emoji requests that are auto-granted unless blocked by other moderators can keep these server-wide operational concerns low-friction.

Expanding this from Discord to Social Media at large is tricky since most social media doesn’t have strictly drawn boundaries between communities. Places like Reddit do, although I don’t think Reddit makes it easy to sub-divide any further than communities. In any case, I believe Digital Locality is the solution for moderating social media at large, which does not rely on any formal governance systems.

Pitfalls

The Tyranny of Structurelessness argues that no organization is truly “structureless”, but rather that the structures may present informally, invisibly, and thus unaccountably. These informal in-groups may coordinate outside formal processes, shape agendas, and steer outcomes through peer pressure. Newcomers initially unfamiliar with the social dynamics are then left without knowledge of the actual rules they’re playing by.

The framework we’ve setup here is already fairly resistant to these concerns. Local decision-making and relying on consensus for constitutional change keeps the informal power of in-groups limited and scope and allows individuals to block constitutional changes. But, they can still happen from things like an in-group peer pressuring full-org assemblies into not blocking, or non-rotating delegates ossifying. In addition to rotating delegates, orgs can ensure there’s mechanisms for the decisions made in assemblies to be audited by each autonomous group, that nothing was decided upon outside the original instructions provided to the delegate. Also, an autonomous group may still have these informal structures arise within them, and may need ways to resolve conflicts by moving up a level, or enabling mobility between autonomous groups.

Freeman’s warnings can also apply to a simply charismatic individual holding informal influence over the organization, in a way no governance system can truly eliminate. We can do our best to keep that influence out of formal mechanisms, but it’s the kind of thing members of organizations will always have to look out for.

Tools

Loomio is a premium governance platform that looks great, if you can afford it. Igalia uses it for their full-Assembly decisions.

PolicyKit is an open source tool for adding governance systems to platforms like slack and discord. It has a couple example recipes, but this would realistically need you to write code to use yourself. Since it integrates into those platforms, it can handle things automatically like deleting posts members voted to delete. Of particular note, I think the governance system itself should be under the purview of the governance system, which in this case would mean proposing code proposals, which might be tricky to setup fully automatically.

It should also be noted that the usual issues associated with electronic voting - which center around a large attack surface for large-scale vote manipulation in a context where we need accurate tallies without exposing individual votes - do not apply at all to this form of governance! Blocks are public, and it’s immediately obvious to the blocker if their position was changed by a bug or compromised part of the system.

Anarchist Cybernetics