Return-Path: Delivered-To: alvarochf@traceseis.com Received: from cs510.bluehost.com by cs510.bluehost.com with LMTP id MKbWFsV562lKOwkAWiBQxw (envelope-from ) for ; Fri, 24 Apr 2026 08:10:13 -0600 Return-path: Envelope-to: alvarochf@traceseis.com Delivery-date: Fri, 24 Apr 2026 08:10:13 -0600 Received: from v517.v5375b7fa.use4.send.mailgun.net ([159.112.244.17]:54986) by cs510.bluehost.com with esmtps (TLS1.2) tls TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256 (Exim 4.98.2) (envelope-from ) id 1wGHES-00000002YfN-2CrZ for alvarochf@traceseis.com; Fri, 24 Apr 2026 08:10:13 -0600 DKIM-Signature: a=rsa-sha256; v=1; c=relaxed/relaxed; d=mg2.substack.com; q=dns/txt; s=mailo; t=1777039796; x=1777046996; h=List-Unsubscribe-Post: List-Unsubscribe: List-Post: List-Id: List-Archive: List-Owner: Reply-To: In-Reply-To: References: Sender: Sender: Date: Message-Id: To: To: From: From: Subject: Subject: Content-Type: Mime-Version; bh=oJMUjvxTMSy69C3m3fl7cks8BGFXUPVuHbbZCQ+CKYI=; b=lemuDUhZ0negT1u80NqsGqZ57N/CgPHgHGJK8y/jdLUiE/5rtpp0kuFD0ilaSDq/3eyxCg2HYQha+4B3CQkvLONkHTr3EsENdJHZ/+MMgq2+s7qNuQ5Cd1OZJrJ1ZEXsDQNjKcPs3pis44jfYRvsuHdvGTIDPiRjCuHDZ2o0OWw= X-Mailgun-Sid: WyI3MDdlYyIsImFsdmFyb2NoZkB0cmFjZXNlaXMuY29tIiwiNGQxODU4Il0= Received: by 95fdb6aedeed05e28adcc51ca727cb31c7a611cae776b42997d6ff1c16420ced with HTTP id 69eb79b2f4ece30470fee40c; Fri, 24 Apr 2026 14:09:54 GMT X-Mailgun-Sending-Ip: 159.112.244.17 X-Mailgun-Batch-Id: 69eb79b2d86abfee860780c7 Mime-Version: 1.0 Content-Type: multipart/alternative; boundary="09c6004c1da58507216cbae79678f670fd8f06d229bf1098badc06d3d771" Subject: When Decentralization Fails From: Cosmos Institute To: alvarochf@traceseis.com X-Mailgun-Tag: post X-Mailgun-Track-Clicks: false Message-Id: <20260424140345.3.1a38406d32f701af@mg2.substack.com> Date: Fri, 24 Apr 2026 14:03:45 +0000 Feedback-Id: post-195335695:cat-post:pub-2225794:substack Sender: Cosmos Institute References: In-Reply-To: Reply-To: Cosmos Institute List-Owner: List-Url: List-Archive: List-Id: List-Post: List-Unsubscribe: List-Unsubscribe-Post: List-Unsubscribe=One-Click X-Mailgun-Variables: {"category": "post", "email_generated_at": "1777039791987", "is_freemail": "true", "post_audience": "everyone", "post_id": "195335695", "post_type": "newsletter", "pub_community_enabled": "true", "publication_id": "2225794", "subdomain": "cosmosinstitute", "user_id": "347258985"} X-Spam-Status: No, score=-1.5 X-Spam-Score: -14 X-Spam-Bar: - X-Spam-Flag: NO --09c6004c1da58507216cbae79678f670fd8f06d229bf1098badc06d3d771 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable View this post on the web at https://blog.cosmos-institute.org/p/when-decen= tralization-fails Francis Bacon=E2=80=99s New Atlantis describes a utopian island whose ruler= s rely on an institution called Salomon=E2=80=99s House for knowledge and d= iscovery. This society of the learned conducted experiments under condition= s of strict secrecy, deciding among themselves which of their findings they= should share with the sovereign. Bacon spent much of his political career = unsuccessfully trying to convince King James I to establish a real-world Sa= lomon=E2=80=99s House. Thomas Hobbes, who had served as Bacon=E2=80=99s secretary in the early 162= 0s, came to a much darker view of such institutions. In Leviathan, publishe= d a quarter of a century after his old employer=E2=80=99s death, he compare= d corporations that wielded such independent judgment to worms in the entra= ils of man, sapping the undivided sovereignty he thought essential to peace= =2E Questions about science, power, and accountability date back through centur= ies of political thought. In today=E2=80=99s world, a handful of companies = control the compute, data, and frontier models that are restructuring how b= illions of people interact with the world. Existing institutions are strugg= ling to keep up. The concentration of power in AI labs is now one of the de= fining political questions of the decade. Many are unhappy about this development, with groups like the AI Now Instit= ute [ https://substack.com/redirect/6b492d43-f2c0-4e44-afd6-c8cbb67b732d?j= =3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ], the D= istributed AI Research Institute (DAIR) [ https://substack.com/redirect/aef= 600c0-5a09-4246-adad-ac31e7c3c9f0?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudlu= PwbMYWklSu8UKGoFv6eWS-HyoA ], and the Algorithmic Justice League [ https://= substack.com/redirect/e0ff14c5-d973-47a8-95b5-8349f1719e38?j=3DeyJ1IjoiNXFx= eXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] arguing that AI devel= opment as currently constituted is irredeemably centralizing. They believe = that we need to relocate authority away from corporations and regulators to= wards the communities most affected by these systems. When policymakers loo= k for alternatives to the status quo of corporate self-governance and light= -touch regulation, these groups are frequently in the room. Ideas around participatory AI governance draw on a deep intellectual tradit= ion that integrates technology and power, dating back to nineteenth century= anarchism and running through twentieth century American social theory. Wh= ile elements of the diagnosis have force, both the analysis and the prescri= ptions suffer from fatal flaws that become even more acute in the AI age. Anarchy and utopia Marxism dominated nineteenth and twentieth century left-wing thought, but i= t had a serious rival. Where the Marxists wanted workers to seize state pow= er and wield it as an instrument of transformation before allowing it to wi= ther away, anarchists like Pierre-Joseph Proudhon and Peter Kropotkin argue= d that concentrated power always reproduces itself. A workers=E2=80=99 stat= e would just produce a new ruling class of bureaucrats and party officials.= The record of twentieth-century communism suggests they had a point. Their alternative was a rejection of authority in favor of decentralized, s= elf-governing communities. They quickly identified large-scale industrial p= roduction as part of the problem: the modern factory introduced managerial = hierarchy and stripped the worker of control over what he made. The early anarchists inspired a generation of twentieth century writers who= reflected on technology and dominance in greater detail. The American soci= ologist Lewis Mumford was one of the first people to place a theoretical fr= ame [ https://substack.com/redirect/f1c06333-8355-4761-9213-89bc50822c74?j= =3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] around= the idea that a technology=E2=80=99s underlying logic could be inherently = centralizing. Mumford distinguished between two types of technology: =E2=80=98democratic = technics=E2=80=99 that individuals could understand, maintain, and operate,= and =E2=80=98authoritarian technics=E2=80=99, which are large-scale system= s that subordinate individuals to their operational logic. In his account, = medieval and early modern economic activity had been artisanal. It had been= based on wood, water, and wind power, which were embedded in local communi= ties. The rise of the factory system destroyed this and subordinated worker= s to machines. Mumford=E2=80=99s work became progressively more pessimistic. In The Myth o= f the Machine, published in the late 1960s, Mumford wrote of the =E2=80=98m= ega machine=E2=80=99 =E2=80=93 machines that were organized physical system= s rather than human tools. The first example was the apparatus used by the= Pharaohs to build the pyramids, which treated human labor as a raw materia= l. Mumford believed that the modern bureaucratic state and corporation were= both manifestations of the same phenomenon. He came to believe that these = systems could not be dismantled from within and that the only way out was a= society-wide =E2=80=98great refusal=E2=80=99. Mumford=E2=80=99s work in part inspired Ivan Illich, an Austrian Catholic p= riest writing in the 1970s. Illich=E2=80=99s central argument was that some= tools and institutions, as they scale, go from extending human capabilitie= s to restructuring their environment so that the activity they were designe= d to serve relies on them. When this happens, the tool achieves a =E2=80=98= radical monopoly.=E2=80=99 For example, the car produces sprawl and redesig= ns cities so that walking becomes impossible. While you can break up a normal monopoly through anti-trust action, it is h= arder to make an unwalkable city walkable again. Illich argued that we had = to keep the development of technology below a certain threshold. For exampl= e, in Energy and Equity, he argued that the speed of vehicles should be cap= ped at 15 miles per hour, to prevent the bicycle =E2=80=93 a technology tha= t merely extended human capabilities =E2=80=93 from being replaced by the c= ar. While Mumford and Illich provided much of the theory of technology, America= n social theorist Murray Bookchin contributed the political program. Bookch= in extensively studied Ancient Greece, revolutionary Paris, Spanish anarchi= st collectives, and the politics of New England, and developed a model of h= ow to organize decentralized, community-governed life. Bookchin outlined a model where citizens=E2=80=99 assemblies at the municip= al level make decisions by direct deliberation and majoritarian voting. The= y would oversee administration and economic life, such as land use and reso= urce allocation. Municipalities would then federate into larger networks fo= r coordination across regions, but ultimate sovereignty would always remain= at local level. Patterning justice The main planks of the modern movement for participatory AI governance come= from this lineage: the inherently centralizing force of technology, the ne= ed to keep capabilities below a certain threshold, and a federated system o= f governance. They began to enter academia in the 1970s and 1980s through t= he emerging field of Science and Technology Studies, which focused on how s= cientific knowledge was socially constructed. Works like Safiya Umoja Noble= =E2=80=99s Algorithms of Oppression (2018) and Ruha Benjamin=E2=80=99s Race= After Technology (2019) would add race as a central analytical category, w= hich this earlier work had lacked. The AI Now Institute, for example, has argued [ https://substack.com/redire= ct/1430801a-da00-4669-8565-666b677a7adc?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64c= TNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] that: =E2=80=9CAI as a field has been no= t just co-opted but constituted by the logics of a few dominant tech firms.= It is no coincidence that the =E2=80=98bigger-is-better=E2=80=99 paradigm = that dominates the field today=E2=80=A6lines up neatly with the incentives = of Big Tech.=E2=80=9D Activists routinely draw on three solutions. The first is locating authority in the communities affected by AI systems, = circumventing expert regulators or corporate self-governance, which they be= lieve are biased and liable to capture. A Kenyan content moderator at an ou= tsourcing firm used by Meta will know things about the conditions and psych= ological toll of their work that an AI ethics researcher in an American uni= versity can=E2=80=99t. Much like Bookchin studying patterns of governance a= round the world, DAIR conducted its Data Workers=E2=80=99 Inquiry [ https:/= /substack.com/redirect/f777ddee-e5e5-46b5-b80f-abc186bd9016?j=3DeyJ1IjoiNXF= xeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ], a global participat= ory research project, inspired by Marx=E2=80=99s 1880 inquiry [ https://sub= stack.com/redirect/e332031b-606c-409c-a0b5-d39ed4824741?j=3DeyJ1IjoiNXFxeXF= 4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] into the conditions of t= he French working class. DAIR paid data workers in Kenya, Syria, Venezuela,= and Germany to document their own conditions as community researchers Second, abandoning corporate AI development altogether. In its place, we wo= uld see federated, community-owned AI infrastructure [ https://substack.com= /redirect/969c8c9f-95f8-4f3f-920b-c28ca6318540?j=3DeyJ1IjoiNXFxeXF4In0.h9dE= DbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ], employed to develop small, task-= specific models. The infrastructure, training data, and resulting models wo= uld be treated as resources belonging to the communities that generate and = are affected by them. Finally, there are certain types of systems that should just not be built u= nder any circumstances, because they will inevitably end up being coercive.= The most common examples are usually facial recognition and autonomous wea= pon. Imagined communities=20 Bluntly, there is a reason that the anarchists lost the battle of ideas on = the left the first time around. Many of these proposals disintegrate in the= face of reality and scale. For a start, it=E2=80=99s often unclear what community governance means her= e. Which community? Defined by whom? With what boundaries? How do we weigh = up impacts on different communities? Kropotkin and Bookchin had assumed tha= t communities would be determined by geography, which is clearly not what m= odern-day activists mean. Even if we can determine who the community is, neither Bookchin nor modern-= day activists developed mechanisms to prevent capture by an organized or ar= ticulate minority. The people who show up to participatory design sessions = are not a random sample of affected populations =E2=80=93 they=E2=80=99re u= sually activists, academics, and professionals in the participation industr= y. The retreat to a world of =E2=80=98communities=E2=80=99 also faces a seriou= s coordination challenge. A federated commons of small, task-specific AI mo= dels requires compute infrastructure, which someone has to fund and maintai= n. It requires interoperability standards, which someone has to set and enf= orce. It requires data governance rules, which someone has to adjudicate wh= en communities disagree. It requires protection from being outcompeted and = absorbed by well-resourced corporate alternatives, which means either subsi= dies or regulatory barriers or both. None of this can be coordinated by voluntary mutual agreement among communi= ties, because they don=E2=80=99t have the resources, the technical capacity= , or the legal authority. This was ultimately the charge Marx leveled at Pr= oudhon [ https://substack.com/redirect/8affd296-40d6-4620-83d0-8f2eed843848= ?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ]: you= cannot decentralize production by changing who owns it if the production p= rocess itself requires centralization. Even if we again suspend all the practical arguments and accept that this i= s possible, a world of small, task-specific models comes with big trade-off= s. As the authors of these proposals tend to view LLMs as =E2=80=98stochast= ic parrots [ https://substack.com/redirect/47247c66-3d79-4482-9e5e-7038391b= 2c2f?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ],= =E2=80=99 they can take refuge in a Mumfordian nostalgia for the small-scal= e and the artisanal. Their framework doesn=E2=80=99t account for the possib= ility that bigger models could produce genuine public goods, such as AlphaF= old [ https://substack.com/redirect/e1d858e8-8e27-41d5-8f99-0309b9f8415a?j= =3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ], so th= ey simply disregard it. It=E2=80=99s not that these activist groups are interpreting the ideas of M= umford, Illich, or Bookchin incorrectly. In fact, these proposals are very = faithful renditions, which serve to highlight the flaws in the originals. All three thinkers built their ideas on a romantic anthropology that treate= d decentralization and face-to-face deliberation as the natural human condi= tion from which industrial modernity is a deviation. This is why the tradit= ion can=E2=80=99t navigate trade-offs. If your starting premise is that human flourishing is what happens when the= megamachine gets out of the way, you don=E2=80=99t need to weigh the goods= it produces, because they aren=E2=80=99t really goods. You don=E2=80=99t n= eed a theory of when expertise is legitimate, because expertise is a sympto= m of the problem. You don=E2=80=99t need mechanisms against capture, becaus= e capture is what happens under the current system and will dissolve along = with it. The intellectual apparatus is structured to avoid the questions that a func= tional governance regime has to answer. What looks like a program for radic= al democracy turns out to be a refusal of the conditions under which democr= atic decisions about complex systems can be made at all. A different tradition The failure of the participatory alternatives doesn=E2=80=99t force us to a= ccept centralization passively. There is a rich alternative tradition runni= ng through Alexis de Tocqueville, the American federalists, and the work of= Elinor Ostrom. Where the anarchist tradition attempts to relocate power ba= ck to the community, this contrasting liberal tradition aims to ensure that= no single locus of authority =E2=80=93 whether state, corporation, or comm= unity =E2=80=93 acquires comprehensive jurisdiction over any domain of life= =2E The liberal tradition rejects the anarchist conflation of freedom with dece= ntralization. Tocqueville=E2=80=99s great insight was that a democratic com= munity can be both decentralized and unfree, because of the social pressure= to conform. Instead, the distinction between a free and an unfree society = is determined by the institutional life within it. When Tocqueville visited America, he was struck by its thick layer of overl= apping, competing associations. He wrote of how =E2=80=9CAmericans of all a= ges, all conditions, and all dispositions, constantly form associations. Th= ey have not only commercial and manufacturing companies, in which all take = part, but associations of a thousand other kinds =E2=80=93 religious, moral= , serious, futile, extensive, or restricted, enormous or diminutive.=E2=80= =9D To most anarchist thinkers, institutions like professional bodies, religiou= s organizations, and commercial associations should be regarded with suspic= ion. They are hierarchical, exclusionary, and reproduce entrenched interest= s, so should be replaced with direct participation. By contrast, Tocqueville saw their partiality as their strength. Each has l= imited jurisdiction and none claims authority over the whole person. Citize= ns belong to many simultaneously, and the overlapping, competing claims cre= ate space in which individuals can appeal from one authority to another. Wi= thout this intermediary layer, the state would fill the gap. Its authority = would be =E2=80=9Cabsolute, minute, regular, provident, and mild,=E2=80=9D = and while it might not coerce people, it would =E2=80=9Ckeep them in perpet= ual childhood,=E2=80=9D as they lost their capacity to exercise free will. Tocqueville was reporting on the constitution working as intended. The Amer= ican federalists had started from the premise that concentrated power, even= democratic power, tends toward its own abuses. Their response, set out mos= t clearly in Federalist 51 [ https://substack.com/redirect/a9048c80-1bec-41= c6-a8c8-147dce10886c?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UK= GoFv6eWS-HyoA ], was to divide authority so that no institution held compre= hensive jurisdiction =E2=80=93 =E2=80=9Cambition must be made to counteract= ambition=E2=80=9D =E2=80=93 and to leave space below the federal level for= states, municipalities, and voluntary associations to govern their own aff= airs. Averting the tragedy of the commons Tocqueville described how such an arrangement looked from the outside. Othe= rs have since done the harder work of showing how it holds together. Elinor Ostrom was a political economist who studied how communities around = the world manage shared resources =E2=80=93 fisheries, forests, irrigation = systems, grazing lands =E2=80=93 without either state regulation or privati= zation. The conventional wisdom, encapsulated in ecologist Garrett Hardin= =E2=80=99s description of the =E2=80=9Ctragedy of the commons=E2=80=9D, hel= d that this couldn=E2=80=99t work. Ostrom found that communities in Switzerland, Spain, Japan, the Philippines= , and dozens of other settings had developed sophisticated, durable governa= nce arrangements for shared resources, some lasting centuries. At the same = time, Ostrom concluded that many commons governance arrangements failed. Th= e examples that endured shared a number of common features. The most important of these is that successful commons are governed by appr= opriators =E2=80=93 people who use the resource and bear the consequences o= f how it=E2=80=99s governed. The second is that the governance activity itself generates the knowledge r= equired [ https://substack.com/redirect/d406098f-6456-425b-8a3a-36fb4ca4455= c?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] for= governance to work. In her studies of fishing communities and irrigation s= ystems, monitoring was done by users as a byproduct of their own activity. = Conflict resolution was handled internally by people who understood the con= text, while violations surfaced information about the extent to which the r= ules were well-calibrated. A third feature is redundancy: long-enduring commons tend to have multiple = overlapping mechanisms for monitoring and enforcement, so that the failure = of any one doesn=E2=80=99t cascade into general rule-breaking. For AI, this framework points toward distributing governance across domain-= specific intermediary institutions. For example, the relevant knowledge for= governing AI in cancer medicine is held by oncologists. The oncologists do= n=E2=80=99t mentally store the information, waiting to deploy it, but gener= ate it through their use of technology in context. A central regulator cann= ot have this knowledge, because it is produced through practice, not availa= ble prior to it. In other words, governance must go where the knowledge is. This could be pr= ofessional bodies, academic institutions, or open-source communities. They = would each govern usage within the domain where their members have the requ= isite competence and stakes. Fortunately, most of these institutions alread= y exist. They do not need to be designed from first principles or assembled= by the participation industry. While the vast majority of governance questions are deployment problems whe= re domain-specific institutions have the advantage, there are a handful of = bigger challenges that sit above this. Problems like the security of fronti= er model weights or thresholds for certain dangerous capabilities sit at a = higher layer that require a degree of either state or interstate coordinati= on. This is why Ostrom wrote about =E2=80=9Cnested enterprises.=E2=80=9D For ex= ample, in eastern Spain, farmers have managed shared water for close to a t= housand years through a tiered structure. At the local level, irrigators=E2= =80=99 associations allocate water within each canal and monitor compliance= among their own members. Above them sits the Tribunal de las Aguas of Vale= ncia, which has met every Thursday morning outside the Apostles=E2=80=99 Do= or of Valencia Cathedral for at least five centuries to resolve disputes be= tween communities. The higher level didn=E2=80=99t replace local governance= , but complemented it by providing the baseline rules that allowed it to fu= nction. Ostrom extended this thinking to global problems late in her career. On cli= mate change, the conventional view was that only an enforceable global trea= ty could work and that subnational efforts were a distraction. She argued t= he opposite [ https://substack.com/redirect/a3ba46c3-2b16-49d4-9826-bbd0320= defb6?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ]= =2E Letting cities, regions, nations an= d blocs cut emissions in parallel mean= t that different approaches were tried, more was learned, while the people = making commitments were accountable to constituents who could see whether t= hey kept them. Waiting for global consensus before allowing anything else t= o happen maximized risk for everyone. The argument for a single framework or set of rules to break the power of t= he AI labs would fall into the same trap. By contrast, a polycentric world = in which medical bodies, universities, industry associations, or open sourc= e communities independently develop their norms restores this redundancy. Life on the frontier No quantity of nested enterprises can resolve the production-side concentra= tion of frontier AI. A handful of labs control the most powerful models, an= d no amount of deployment-side checks and balances can change that. But a thick ecosystem of intermediary institutions on the deployment side c= reates countervailing power. The labs must satisfy many masters rather than= capturing one regulator, or, as the anarchist model would have it, being r= eplaced by a constellation of community-run alternatives that will never ma= tch their capabilities. These checks could vary widely. A medical licensing body can refuse to cert= ify practices built on tools its members judge unsafe. A malpractice insure= r can price risk based on whether clinicians follow professional norms. A p= rocurement officer can refuse to buy systems that don=E2=80=99t meet standa= rds set by a professional body. None of this is regulation in the conventio= nal sense, and none of it requires a legislature to act or a global governa= nce regime to take shape. But collectively, these institutions force the la= bs to satisfy demands they didn=E2=80=99t set and can=E2=80=99t unilaterall= y override. Freedom has never depended on power being small. It has depended on power b= eing answerable to more than one authority at a time, and on citizens belon= ging to institutions that can push back on their own terms. The task ahead = of us is building that intermediary layer. Cosmos Institute [ https://substack.com/redirect/3c2a6240-4ca8-447e-9957-41= 4be4c28817?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-H= yoA ] is the Academy for Philosopher-Builders, technologists building AI fo= r human flourishing. We run fellowships, fund AI prototypes, and host semin= ars with institutions like Oxford, Aspen Institute, and Liberty Fund. Unsubscribe https://substack.com/redirect/2/eyJlIjoiaHR0cHM6Ly9ibG9nLmNvc21= vcy1pbnN0aXR1dGUub3JnL2FjdGlvbi9kaXNhYmxlX2VtYWlsP3Rva2VuPWV5SjFjMlZ5WDJsa0= lqb3pORGN5TlRnNU9EVXNJbkJ2YzNSZmFXUWlPakU1TlRNek5UWTVOU3dpYVdGMElqb3hOemMzT= URNNU56a3hMQ0psZUhBaU9qRTRNRGcxTnpVM09URXNJbWx6Y3lJNkluQjFZaTB5TWpJMU56azBJ= aXdpYzNWaUlqb2laR2x6WVdKc1pWOWxiV0ZwYkNKOS5tR3hwUFRQOElpWGgwTlFRNGJaY1lqR0t= IcHdpUXFNaHNmajR2MzIxenpVIiwicCI6MTk1MzM1Njk1LCJzIjoyMjI1Nzk0LCJmIjp0cnVlLC= J1IjozNDcyNTg5ODUsImlhdCI6MTc3NzAzOTc5MSwiZXhwIjoyMDkyNjE1NzkxLCJpc3MiOiJwd= WItMCIsInN1YiI6ImxpbmstcmVkaXJlY3QifQ.SGhf5-kXQh9IzAAsA8ZyApMwdm-z_P1ClS1vA= 7HhIg0? --09c6004c1da58507216cbae79678f670fd8f06d229bf1098badc06d3d771 Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable = When Decentralization Fails
And how it succeeds
͏     &= #173;͏     ­͏     ­͏   =   ­͏     ­͏     ­͏= ;     ­͏     ­͏     &#= 173;͏     ­͏     ­͏   &= #8199; ­͏     ­͏     ­͏=     ­͏     ­͏     = 73;͏     ­͏     ­͏   &#= 8199; ­͏     ­͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­͏     ­͏     ­= ͏     ­͏     ­͏   ̳= 9; ­͏     ­͏     ­͏ &nb= sp;   ­͏     ­͏     ­&= #847;     ­͏     ­͏    = ; ­͏     ­͏     ­͏ &nbs= p;   ­͏     ­͏     ­&#= 847;     ­͏     ­͏    = ­͏     ­͏     ­͏  = ;   ­͏     ­͏     ­= 47;     ­͏     ­͏     = ­͏     ­͏     ­͏  =   ­͏     ­͏     ­T= 7;     ­͏     ­͏     &= #173;͏     ­͏     ­͏   =   ­͏     ­͏     ­͏= ;     ­͏     ­͏     &#= 173;͏     ­͏     ­͏   &= #8199; ­͏     ­͏     ­͏=     ­͏     ­͏     = 73;͏     ­͏     ­͏   &#= 8199; ­͏     ­͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­͏     ­͏     ­= ͏     ­͏     ­͏   ̳= 9; ­͏     ­͏     ­͏ &nb= sp;   ­͏     ­͏     ­&= #847;     ­͏     ­͏    = ; ­͏     ­͏     ­͏ &nbs= p;   ­͏     ­͏     ­&#= 847;     ­͏     ­͏    = ­͏     ­͏     ­͏  = ;   ­͏     ­͏     ­= 47;     ­͏     ­͏     = ­͏     ­͏     ­͏  =   ­͏     ­͏     ­T= 7;     ­͏     ­͏     &= #173;͏     ­͏     ­͏   =   ­͏     ­͏     ­͏= ;     ­͏     ­͏     &#= 173;͏     ­͏     ­͏   &= #8199; ­͏     ­͏     ­͏=     ­͏     ­͏     = 73;͏     ­͏     ­͏   &#= 8199; ­͏     ­͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­͏     ­͏     ­= ͏     ­͏     ­͏   ̳= 9; ­͏     ­͏     ­͏ &nb= sp;   ­͏     ­͏     ­&= #847;     ­
<= td height=3D"16" style=3D"font-size:0px;line-height:0;"> 
Forwarded t= his email? Subscribe here for more<= /span>

When Decent= ralization Fails

And how it succeeds

<= tr>
Alex Chalme= rs
<= tbody>
Apr 24
 
<= /tr>
<= img class=3D"icon" src=3D"https://substackcdn.com/image/fetch/$s_!PeVs!,w_3= 6,c_scale,f_png,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack.com= %2Ficon%2FLucideHeart%3Fv%3D4%26height%3D36%26fill%3Dnone%26stroke%3D%25238= 08080%26strokeWidth%3D2" width=3D"18" height=3D"18" style=3D"border: none;v= ertical-align: middle;max-width: 18px" alt=3D"">
3D""
3D""
3D""
READ IN APP3D""
<= /table>
Giovanni Paolo Panini, A Capriccio of Cla= ssical Ruins with the Pyramid of Cestius and Figures in a Landscape (1730s)=

Francis Bacon’s New Atlantis describes a utopian island whose rulers rely = on an institution called Salomon’s House for knowledge and discovery.= This society of the learned conducted experiments under conditions of stri= ct secrecy, deciding among themselves which of their findings they should s= hare with the sovereign. Bacon spent much of his political career unsuccess= fully trying to convince King James I to establish a real-world SalomonR= 17;s House.

Thomas Hobbes, who had served as B= acon’s secretary in the early 1620s, came to a much darker view of su= ch institutions. In Leviathan, published a quarter of= a century after his old employer’s death, he compared corporations t= hat wielded such independent judgment to worms in the entrails of man, sapp= ing the undivided sovereignty he thought essential to peace.

Questions about science, power, and accountability date back throu= gh centuries of political thought. In today’s world, a handful of com= panies control the compute, data, and frontier models that are restructurin= g how billions of people interact with the world. Existing institutions are= struggling to keep up. The concentration of power in AI labs is now one of= the defining political questions of the decade.

Many= are unhappy about this development, with groups like the AI Now Institute, t= he Distribut= ed AI Research Institute (DAIR), and the Algorithmic Justice League= arguing that AI development as currently constituted is irredeemably centr= alizing. They believe that we need to relocate authority away from corporat= ions and regulators towards the communities most affected by these systems.= When policymakers look for alternatives to the status quo of corporate sel= f-governance and light-touch regulation, these groups are frequently in the= room.

Ideas around participatory AI governance draw= on a deep intellectual tradition that integrates technology and power, dat= ing back to nineteenth century anarchism and running through twentieth cent= ury American social theory. While elements of the diagnosis have force, bot= h the analysis and the prescriptions suffer from fatal flaws that become ev= en more acute in the AI age.

<= span style=3D"color: #ffffff;text-decoration: none;">Upgrade to paid=

Anarchy and utopia

Marxism = dominated nineteenth and twentieth century left-wing thought, but it had a = serious rival. Where the Marxists wanted workers to seize state power and w= ield it as an instrument of transformation before allowing it to wither awa= y, anarchists like Pierre-Joseph Proudhon and Peter Kropotkin argued that c= oncentrated power always reproduces itself. A workers’ state would ju= st produce a new ruling class of bureaucrats and party officials. The recor= d of twentieth-century communism suggests they had a point.

Their alternative was a rejection of authority in favor of decentralized, = self-governing communities. They quickly identified large-scale industrial = production as part of the problem: the modern factory introduced managerial= hierarchy and stripped the worker of control over what he made.

The early anarchists inspired a generation of twentieth century= writers who reflected on technology and dominance in greater detail. The A= merican sociologist Lewis Mumford was one of the first people to place a theoretical fr= ame around the idea that a technology’s underlying logic co= uld be inherently centralizing.

Mumford distinguishe= d between two types of technology: ‘democratic technics’ that i= ndividuals could understand, maintain, and operate, and ‘authoritaria= n technics’, which are large-scale systems that subordinate individua= ls to their operational logic. In his account, medieval and early modern ec= onomic activity had been artisanal. It had been based on wood, water, and w= ind power, which were embedded in local communities. The rise of the factor= y system destroyed this and subordinated workers to machines.

Mumford’s work became progressively more pessimistic. In <= /span>The Myth of the Machine, published in the late 1960s, = Mumford wrote of the ‘mega machine’ – machines that were = organized physical systems rather than human tools. The first example was = the apparatus used by the Pharaohs to build the pyramids, which treated hum= an labor as a raw material. Mumford believed that the modern bureaucratic s= tate and corporation were both manifestations of the same phenomenon. He ca= me to believe that these systems could not be dismantled from within and th= at the only way out was a society-wide ‘great refusal’.<= /p>

Mumford’s work in part inspired Ivan Illich, an Austr= ian Catholic priest writing in the 1970s. Illich’s central argument w= as that some tools and institutions, as they scale, go from extending human= capabilities to restructuring their environment so that the activity they = were designed to serve relies on them. When this happens, the tool achieves= a ‘radical monopoly.’ For example, the car produces sprawl and= redesigns cities so that walking becomes impossible.

While you can break up a normal monopoly through anti-trust action, it is = harder to make an unwalkable city walkable again. Illich argued that we had= to keep the development of technology below a certain threshold. For examp= le, in Energy and Equity, he argued that the speed of= vehicles should be capped at 15 miles per hour, to prevent the bicycle = 211; a technology that merely extended human capabilities – from bein= g replaced by the car.

While Mumford and Illich prov= ided much of the theory of technology, American social theorist Murray Book= chin contributed the political program. Bookchin extensively studied Ancien= t Greece, revolutionary Paris, Spanish anarchist collectives, and the polit= ics of New England, and developed a model of how to organize decentralized,= community-governed life.

Bookchin outlined a model where c= itizens’ assemblies at the municipal level make decisions by direct d= eliberation and majoritarian voting. They would oversee administration and = economic life, such as land use and resource allocation. Municipalities wou= ld then federate into larger networks for coordination across regions, but = ultimate sovereignty would always remain at local level.

= Patterning justice

The main planks of the modern mov= ement for participatory AI governance come from this lineage: the inherentl= y centralizing force of technology, the need to keep capabilities below a c= ertain threshold, and a federated system of governance. They began to enter= academia in the 1970s and 1980s through the emerging field of Science and = Technology Studies, which focused on how scientific knowledge was socially = constructed. Works like Safiya Umoja Noble’s Algorithms of= Oppression (2018) and Ruha Benjamin’s Race Aft= er Technology (2019) would add race as a central analytical cate= gory, which this earlier work had lacked.

The = AI Now Institute, for example, has argued that: “AI as a field has been= not just co-opted but constituted by the logics of a few dominant tech fir= ms. It is no coincidence that the ‘bigger-is-better’ paradigm t= hat dominates the field today…lines up neatly with the incentives of = Big Tech.”

Activists routinely draw on three s= olutions.

The first is locating authority in the comm= unities affected by AI systems, circumventing expert regulators or corporat= e self-governance, which they believe are biased and liable to capture. A K= enyan content moderator at an outsourcing firm used by Meta will know thing= s about the conditions and psychological toll of their work that an AI ethi= cs researcher in an American university can’t. Much like Bookchin stu= dying patterns of governance around the world, DAIR conducted its Data Workers’ I= nquiry, a global participatory research project, inspired by Marx’s 188= 0 inquiry into the conditions of the French working class. DAIR p= aid data workers in Kenya, Syria, Venezuela, and Germany to document their = own conditions as community researchers

Second= , abandoning corporate AI development altogether. In its place, we would se= e federated,= community-owned AI infrastructure, employed to develop small, ta= sk-specific models. The infrastructure, training data, and resulting models= would be treated as resources belonging to the communities that generate a= nd are affected by them.

Finally, there are certain = types of systems that should just not be built under any circumstances, bec= ause they will inevitably end up being coercive. The most common examples a= re usually facial recognition and autonomous weapon.

Imag= ined communities

Bluntly, there is a reason that the anar= chists lost the battle of ideas on the left the first time around. Many of = these proposals disintegrate in the face of reality and scale.

For a start, it’s often unclear what community governance means = here. Which community? Defined by whom? With what boundaries? How do we wei= gh up impacts on different communities? Kropotkin and Bookchin had assumed = that communities would be determined by geography, which is clearly not wha= t modern-day activists mean.

Even if we can determine who t= he community is, neither Bookchin nor modern-day activists developed mechan= isms to prevent capture by an organized or articulate minority. The people = who show up to participatory design sessions are not a random sample of aff= ected populations – they’re usually activists, academics, and p= rofessionals in the participation industry.

The retreat to = a world of ‘communities’ also faces a serious coordination chal= lenge. A federated commons of small, task-specific AI models requires compu= te infrastructure, which someone has to fund and maintain. It requires inte= roperability standards, which someone has to set and enforce. It requires d= ata governance rules, which someone has to adjudicate when communities disa= gree. It requires protection from being outcompeted and absorbed by well-re= sourced corporate alternatives, which means either subsidies or regulatory = barriers or both.

None of this can be coordinated by = voluntary mutual agreement among communities, because they don’t have= the resources, the technical capacity, or the legal authority. This was ul= timately the charge Marx leveled at Proudhon: you cannot decentralize product= ion by changing who owns it if the production process itself requires centr= alization.

Even if we again suspend all the pr= actical arguments and accept that this is possible, a world of small, task-= specific models comes with big trade-offs. As the authors of these proposal= s tend to view LLMs as ‘stochastic parrots,’ they can take refuge= in a Mumfordian nostalgia for the small-scale and the artisanal. Their fra= mework doesn’t account for the possibility that bigger models could p= roduce genuine public goods, such as AlphaFold, so they simply disregard it.<= /span>

It’s not that these activist groups are interp= reting the ideas of Mumford, Illich, or Bookchin incorrectly. In fact, thes= e proposals are very faithful renditions, which serve to highlight the flaw= s in the originals.

All three thinkers built their ideas on= a romantic anthropology that treated decentralization and face-to-face del= iberation as the natural human condition from which industrial modernity is= a deviation. This is why the tradition can’t navigate trade-offs.

If your starting premise is that human flourishing is what ha= ppens when the megamachine gets out of the way, you don’t need to wei= gh the goods it produces, because they aren’t really goods. You don&#= 8217;t need a theory of when expertise is legitimate, because expertise is = a symptom of the problem. You don’t need mechanisms against capture, = because capture is what happens under the current system and will dissolve = along with it.

The intellectual apparatus is structured to = avoid the questions that a functional governance regime has to answer. What= looks like a program for radical democracy turns out to be a refusal of th= e conditions under which democratic decisions about complex systems can be = made at all.

A different tradition

The f= ailure of the participatory alternatives doesn’t force us to accept c= entralization passively. There is a rich alternative tradition running thro= ugh Alexis de Tocqueville, the American federalists, and the work of Elinor= Ostrom. Where the anarchist tradition attempts to relocate power back to t= he community, this contrasting liberal tradition aims to ensure that no sin= gle locus of authority – whether state, corporation, or community = 211; acquires comprehensive jurisdiction over any domain of life.

The liberal tradition rejects the anarchist conflation of freedom wi= th decentralization. Tocqueville’s great insight was that a democrati= c community can be both decentralized and unfree, because of the social pre= ssure to conform. Instead, the distinction between a free and an unfree soc= iety is determined by the institutional life within it.

Whe= n Tocqueville visited America, he was struck by its thick layer of overlapp= ing, competing associations. He wrote of how “Americans of all ages, = all conditions, and all dispositions, constantly form associations. They ha= ve not only commercial and manufacturing companies, in which all take part,= but associations of a thousand other kinds – religious, moral, serio= us, futile, extensive, or restricted, enormous or diminutive.”

To most anarchist thinkers, institutions like professional bodies= , religious organizations, and commercial associations should be regarded w= ith suspicion. They are hierarchical, exclusionary, and reproduce entrenche= d interests, so should be replaced with direct participation.

By contrast, Tocqueville saw their partiality as their strength. Each = has limited jurisdiction and none claims authority over the whole person. C= itizens belong to many simultaneously, and the overlapping, competing claim= s create space in which individuals can appeal from one authority to anothe= r. Without this intermediary layer, the state would fill the gap. Its autho= rity would be “absolute, minute, regular, provident, and mild,”= and while it might not coerce people, it would “keep them in perpetu= al childhood,” as they lost their capacity to exercise free will.

=

Tocqueville was reporting on the constitution working as= intended. The American federalists had started from the premise that conce= ntrated power, even democratic power, tends toward its own abuses. Their re= sponse, set out most clearly in Federalist 51, was to divide authority so tha= t no institution held comprehensive jurisdiction – “ambition mu= st be made to counteract ambition” – and to leave space below t= he federal level for states, municipalities, and voluntary associations to = govern their own affairs.

Averting the tragedy of= the commons

Tocqueville described how such an arrangement= looked from the outside. Others have since done the harder work of showing= how it holds together.

Elinor Ostrom was a political econo= mist who studied how communities around the world manage shared resources &= #8211; fisheries, forests, irrigation systems, grazing lands – withou= t either state regulation or privatization. The conventional wisdom, encaps= ulated in ecologist Garrett Hardin’s description of the “traged= y of the commons”, held that this couldn’t work.

The most important of these is that successful commons are governed by= appropriators – people who use the resource and bear the consequence= s of how it’s governed.

The second is that the = governance activity itself generates the knowledge required for governance to= work. In her studies of fishing communities and irrigation systems, monito= ring was done by users as a byproduct of their own activity. Conflict resol= ution was handled internally by people who understood the context, while vi= olations surfaced information about the extent to which the rules were well= -calibrated.

A third feature is redundancy: long-end= uring commons tend to have multiple overlapping mechanisms for monitoring a= nd enforcement, so that the failure of any one doesn’t cascade into g= eneral rule-breaking.

For AI, this framework points toward = distributing governance across domain-specific intermediary institutions. F= or example, the relevant knowledge for governing AI in cancer medicine is h= eld by oncologists. The oncologists don’t mentally store the informat= ion, waiting to deploy it, but generate it through their use of technology = in context. A central regulator cannot have this knowledge, because it is p= roduced through practice, not available prior to it.

In oth= er words, governance must go where the knowledge is. This could be professi= onal bodies, academic institutions, or open-source communities. They would = each govern usage within the domain where their members have the requisite = competence and stakes. Fortunately, most of these institutions already exis= t. They do not need to be designed from first principles or assembled by th= e participation industry.

While the vast majority of govern= ance questions are deployment problems where domain-specific institutions h= ave the advantage, there are a handful of bigger challenges that sit above = this. Problems like the security of frontier model weights or thresholds fo= r certain dangerous capabilities sit at a higher layer that require a degre= e of either state or interstate coordination.

This is why O= strom wrote about “nested enterprises.” For example, in eastern= Spain, farmers have managed shared water for close to a thousand years thr= ough a tiered structure. At the local level, irrigators’ associations= allocate water within each canal and monitor compliance among their own me= mbers. Above them sits the Tribunal de las Aguas of Valencia, which has met= every Thursday morning outside the Apostles’ Door of Valencia Cathed= ral for at least five centuries to resolve disputes between communities. Th= e higher level didn’t replace local governance, but complemented it b= y providing the baseline rules that allowed it to function.

Ostrom extended this thinking to global problems late in her career.= On climate change, the conventional view was that only an enforceable glob= al treaty could work and that subnational efforts were a distraction. She <= /span>argued the op= posite. Letting cities, regions, nations and blocs cut emissions = in parallel meant that different approaches were tried, more was learned, w= hile the people making commitments were accountable to constituents who cou= ld see whether they kept them. Waiting for global consensus before allowing= anything else to happen maximized risk for everyone.

Life on the frontier

No quantity of nes= ted enterprises can resolve the production-side concentration of frontier A= I. A handful of labs control the most powerful models, and no amount of dep= loyment-side checks and balances can change that.

But a thi= ck ecosystem of intermediary institutions on the deployment side creates co= untervailing power. The labs must satisfy many masters rather than capturin= g one regulator, or, as the anarchist model would have it, being replaced b= y a constellation of community-run alternatives that will never match their= capabilities.

These checks could vary widely. A medical li= censing body can refuse to certify practices built on tools its members jud= ge unsafe. A malpractice insurer can price risk based on whether clinicians= follow professional norms. A procurement officer can refuse to buy systems= that don’t meet standards set by a professional body. None of this i= s regulation in the conventional sense, and none of it requires a legislatu= re to act or a global governance regime to take shape. But collectively, th= ese institutions force the labs to satisfy demands they didn’t set an= d can’t unilaterally override.

Freedom has never depe= nded on power being small. It has depended on power being answerable to mor= e than one authority at a time, and on citizens belonging to institutions t= hat can push back on their own terms. The task ahead of us is building that= intermediary layer.

<= hr style=3D"margin: 32px 0;padding: 0;height: 1px;background: rgb(0,0,0,.1)= ;border: none;">

Cosmos Institute is the Academy for Philosoph= er-Builders, technologists building AI for human flourishing. We run fellow= ships, fund AI prototypes, and host seminars with institutions like Oxford,= Aspen Institute, and Liberty Fund.

Upg= rade to paid

=
3D""
 
3D""Like
Comment
3D""Restack
 

= © 2026 Cosmos Institute
500 W 2nd Street, Suite 1900, = Austin, TX 78701
Unsubscribe

3D"Start3D"" --09c6004c1da58507216cbae79678f670fd8f06d229bf1098badc06d3d771--