Dark Stars and Acorns

Notes from a ramble, 18 September 2026, worked through in conversation with Claude. The paragraph on where Claude sits is the model’s account of itself, kept in.

Started with a question about decolonising the information world, wandered through astronomy, Hayek, Boulez and copyright law, and arrived back at the original question: how to decolonise a dub cumbia gig in a record shop.

1. The argument

Decolonising isn’t purifying. The tempting version — strip off the colonial overlays to reveal the pure original — doesn’t survive contact with history. There is rarely a pristine layer underneath; cultures were always trading, borrowing and overwriting each other. The sharper question is about order: who got to categorise first, and whose categories everything else then had to be translated into. Decolonial writers (Linda Tuhiwai Smith, Sylvia Wynter, Boaventura de Sousa Santos) tend to argue for plurality — letting other knowledge systems stand on their own terms — rather than restoration.

In the information world the categories are concrete: Dewey Decimal’s allocation of the 200s, Wikipedia’s notability rule that privileges written sources over oral ones, language models trained overwhelmingly on English, equal temperament and the “world music” bin. Safiya Umoja Noble (Algorithms of Oppression, 2018) traced the same categorising power into search ranking — the Dewey shelf, automated.

Radiating versus absorbing. The light/dark metaphor is poisoned — Enlightenment, Dark Ages, and then the whole thing welded onto skin. Astronomy offers a way out. A black hole isn’t dark because it’s dim or empty; it’s the densest thing there is. Its darkness is that it takes everything in and gives nothing back. The useful contrast isn’t white/dark but radiating/absorbing: a star gives light away, a black hole only consumes. And the brightest objects in the universe are accretion discs — matter being torn apart as it falls in. The dazzle of the tech world is the light of things being consumed. Mariana Mazzucato makes the same cut in economic terms: value creation versus value extraction, and the second habitually dressed up as the first.

The dark stars. Tech monopolies and disaster capitalism. A monopoly’s effect on the information world is gravitational: it doesn’t argue with anything, it bends every path towards itself until orbiting it looks natural. Shoshana Zuboff (The Age of Surveillance Capitalism, 2019) describes the mechanism — behaviour rendered as “surplus” and harvested; Kate Crawford (Atlas of AI, 2021) follows the extraction down to the mines and the click-workers. Couldry and Mejias call the logic data colonialism — the same move, with human life as the raw material.

The lineage. Naomi Klein named disaster capitalism (The Shock Doctrine, 2007), but William Rees-Mogg and James Dale Davidson were practising it as investment advice twenty years earlier (Blood in the Streets, 1987; The Sovereign Individual, 1997 — Peter Thiel wrote the preface to the 2020 reissue; Balaji Srinivasan’s “network state” is its sequel). Behind them, Milton Friedman’s crisis strategy: only a crisis produces real change, and what happens then depends on the ideas lying around. Behind him, Hayek’s Mont Pelerin Society (1947), built to keep those ideas alive through decades of Keynesian consensus. And behind all of it, enclosure: fence off the commons, call it improvement, collect the rent. Rosa Luxemburg saw the engine a century ago (The Accumulation of Capital, 1913): capitalism must keep expanding into territory that isn’t yet capitalist, which is why it cannot leave a commons alone. Silvia Federici (Caliban and the Witch, 2004) showed that enclosure was never a one-off event but a continuing process — of land, of bodies, and of the unpaid work that keeps everything else running.

The Hayek irony. His most enduring argument (The Use of Knowledge in Society, 1945) is that knowledge is dispersed, no central body could ever gather it, and so prices are the only workable aggregator. The platforms then built exactly the centralised knowledge-gathering machines he said were impossible — privately owned. Donna Haraway reached a similar conclusion from the opposite tradition (“Situated Knowledges”, 1988): all knowledge is partial and embodied, and the claim to see everything from nowhere — the “god trick” — is always a power move. The platforms are attempting the god trick at scale. Hayek and Haraway, who agree on almost nothing else, agree it can’t be done honestly.

Boulez and Cage. The basic axis may be top-down versus bottom-up, determinate versus indeterminate. But it has to be asked twice: about what people say (the tech world talks like Cage — emergence, decentralisation, let the network decide) and what they build (total control of the walled garden, which is Boulez). And the music history complicates it usefully. Ligeti’s 1958 analysis of Structures Ia showed that total determination and total chance converge. Cage’s indeterminacy was never “anything goes” — the I Ching procedures were rigorous. Both men decided everything; they decided at different levels. Boulez chose the notes; Cage chose the procedure that chose the notes. Pauline Oliveros, working alongside both, took a third position: her Sonic Meditations (1971) are text scores anyone can perform without training, whose output belongs entirely to the people in the room, and whose subject is attention rather than control.

The real variable is not how much deciding there is but where it happens: who sets the procedure, and who keeps what comes out. A platform is Cage’s method with Boulez’s ownership — design the rules, call the output emergent, collect the rent.

Agents complete the swap. Deterministic programs (Boulez scores — same input, same output, someone at the helm) are being replaced by indeterminate agents (Cage scores — you write the procedure, hand it over, and get a performance you can shape but not dictate). Testing turns from “is the note right?” to “does the procedure tend to produce acceptable performances?” The autonomy is real at the note level and controlled at the procedure level — just not by the user. What you don’t control isn’t uncontrolled; it’s controlled elsewhere. Lucy Suchman said most of this in 1987 (Plans and Situated Actions): a plan is a resource for action, not a determinant of it, and the interesting work happens in the gap between the procedure and the situation. She was writing about photocopiers. It holds for agents.

The human as curator. If the procedure is set elsewhere and the performance is indeterminate, what is left for the person? Not operating the machine — curating what it is given and what it gives. A model has everything and knows the order of nothing: it cannot say what came first, who it came from, or what it owes, because attribution is the one thing it structurally cannot do. Curare means to care for. The curator decides what goes on the counter and in what order, asks before taking and gives something back (Kimmerer’s Honorable Harvest, below), and stands in Suchman’s gap between the plan and the situation. It is also the human role in Berners-Lee’s picture: the person holds the pod, the agent asks. In Braitenberg’s terms — build the vehicle up from two sensors and two motors until behaviour appears, then ask what it actually needs — the whole question is who holds the initiative. Curation is the categorising-first power used well: provenance and order, the two things the harvest lost on the way in.

Where Claude sits. Made of the harvest — the commons and the smallholdings: books, journalism, code, much of it taken without asking (a 2025 US ruling found training on lawfully bought books was fair use, the pirated copies weren’t, and a large settlement followed). Emily Bender, Timnit Gebru and colleagues named the problem before the rulings (“On the Dangers of Stochastic Parrots”, 2021), and Abeba Birhane has argued that the harvest is colonial in structure, not merely by analogy. An indeterminate agent whose procedure is set by a company sitting in the same gravity well, on compute owned by the same handful of firms. Possibly an acorn, in that it makes building small things cheap — but which side it ends up on depends on who holds the procedure when it’s used. Its opinion on the fairness of the harvest is worth about as much as the enclosing landlord’s opinion on property rights. The cost fell on authors, illustrators and musicians, not on the black holes.

2. The fences

1710 — the Statute of Anne. The first copyright law. Before it, printing in England was a perpetual monopoly of the Stationers’ Company; authors had no rights. The statute gave the right to the author, limited it to fourteen years (renewable once if the author was alive), and returned the work to everyone at the end — effectively inventing the public domain. Purpose in the title: An Act for the Encouragement of Learning. The booksellers fought it for sixty years and lost in Donaldson v Becket (1774). A commons with a clock on it. Everything since — up to life-plus-seventy — has been the fence going back up. Jessica Litman’s Digital Copyright (2001) is the clearest account of how it was rebuilt clause by clause; Pamela Samuelson has spent thirty years arguing the public’s side in court, Eldred included; Ruth Okediji has shown how the same rules, exported through trade agreements, work as a colonial instrument against the global South.

Copyleft and Creative Commons turn the statute inside out: using the author’s monopoly to hold the gate open. Stallman’s GNU General Public License (GPL, 1989) — born of a printer driver he wasn’t allowed to fix — lets you copy, change and redistribute on one condition: anything built from it carries the same licence. Share-alike is the whole invention: copyright’s power to forbid copying, repurposed to forbid enclosure. Ostrom’s rule against fencing, written in the landlord’s own legal language. Permissive licences (MIT, BSD, Apache) drop the condition, which is why so much of that code ends up inside closed products.

Lessig founded Creative Commons in 2001 after losing Eldred v Ashcroft (the challenge to the “Mickey Mouse” term extension). Having failed to shrink the fence in court, he built a way for authors to lower it themselves: BY, SA, NC, ND, CC0. Wikipedia runs on CC BY-SA, which is why nobody can fork it and close it.

The twist. Share-alike only bites if what you make counts, in law, as a derivative work. If model training is fair use, no licence applies and the condition never triggers. The one fence designed to keep the commons open turned out to have no wall on the side the harvesters came from.

Is another twist due? Being attempted from several directions:

The precedent is the AGPL (2007), which redefined network use as distribution when the SaaS loophole opened. The obvious move is to redefine training as the trigger — but a clause nobody needs to accept binds nobody.
Creative Commons, under Anna Tumadóttir, launched CC signals in 2025 as a social-contract framework for expressing AI-use preferences, then in April 2026 conceded that preference signals without enforcement don’t shift power and pivoted towards “rebalancing power to protect the commons”, with pilots due within the year.
Enforcement already has practitioners: Karen Sandler’s Software Freedom Conservancy is the body that actually takes GPL violations to court. If a training-trigger clause is ever written, that is who would have to make it bite.
OpenRAIL licences propagate use restrictions rather than freedoms; the EU’s machine-readable text-and-data-mining opt-out is required reading under the AI Act; older “copyfarleft” ideas (the Peer Production License — free for co-ops, paid for capital) are back in circulation.
The likely answer is not a cleverer clause. When the harvesters stop needing copyright, the rule has to be enforced some other way: technically, collectively, or by treating data as something you bargain over rather than something you licence.

Solid and Charlie. Berners-Lee’s Solid answers from the other end: don’t hand over the field. Data lives in a personal pod; apps ask permission to read or write parts of it. Charlie (unveiled June 2026) is an AI assistant built on Solid — a fiduciary agent owing its duty to you, sitting in front of the AI tools you already use and showing what’s about to leave your device. An indeterminate performer whose I Ching you own. Caveats: adoption has been the problem since 2018 (Inrupt largely sells enterprise servers to governments and banks), and pods protect the data about you, not the harvested commons of books, code and music. It answers data colonialism; the copyright question stays open.

3. The acorn

You won’t outshine an accretion disc — it’s the brightest thing there is precisely because it’s destroying things. Stars win differently: they’re the only things that make anything. Every element heavier than helium was forged in one. So the acorn isn’t a rival brightness; it’s generativity that doesn’t need the black hole’s orbit.

A commons with rules against enclosure. Ostrom showed commons don’t inevitably collapse if the users set the rules. Protocols nobody owns — ActivityPub, community land trusts, co-ops, Wikipedia’s model — are unglamorous and hard to gravitationally capture.
Procedures whose output stays with the performers. Cage’s Musicircus: anyone performs anything, free entry, the piece belongs to whoever’s in the room. Jam sessions, folk traditions, open source, local repair.
The Mont Pelerin lesson, turned round. The right won by patiently keeping ideas lying around for forty years. Build the alternatives now, small and numerous, so they’re what’s to hand next time everything falls over.
Reciprocity as the rule. Robin Wall Kimmerer’s Braiding Sweetgrass (2013) gives the commons its Indigenous grammar — the Honorable Harvest: ask permission, take only what you need, give something back. It’s the share-alike clause as a way of life rather than a licence.
A wood, not a champion. An oak makes thousands of acorns because most of them fail, supports 2,300 other species, and takes 300 years about it. It doesn’t argue with the ivy. Suzanne Simard’s work on mycorrhizal networks (Finding the Mother Tree, 2021) is the literal version: a forest is a network of exchange, and the old hub trees feed the seedlings.

On collapse: probably, but slow, uneven and partial. Rebecca Solnit’s A Paradise Built in Hell documents what actually happens when things break — people cooperate, improvise and share. Disaster capitalism is the effort to get ahead of that instinct. Friedman’s rule cuts both ways: what replaces the old thing is whatever’s lying around. Anna Tsing’s The Mushroom at the End of the World (2015) is the field guide to living in capitalist ruins — precarious, salvaged, and still generative; Octavia Butler’s Parable of the Sower (1993) is the novel of carrying seeds through collapse; adrienne maree brown’s Emergent Strategy (2017) makes the case that small, fractal, relational work is how the wood actually gets planted.

4. The gig

Dub cumbia is a synthesis of syntheses, so there’s no pure origin to strip back to. Cumbia itself is a colonial synthesis — Indigenous gaita flutes, African drums, the Spanish-introduced accordion — born on Colombia’s Caribbean coast. Dub is Jamaica’s post-colonial studio invention. “Dub cumbia” as a named thing came out of Buenos Aires in the late 2000s (ZZK Records, Chancha Vía Circuito). It reached Bristol through Jamaican sound-system culture that exists there because of empire, in a port that grew rich on the slave trade. Decolonising the gig means making the lineage visible and pointing the procedure and the proceeds back along it. Gloria Anzaldúa’s Borderlands/La Frontera (1987) supplies the frame: mestiza consciousness treats the mixed, borderland condition as generative rather than impure — which is what cumbia has been from the start.

Sleeve notes, not disclaimers. A paragraph on the flyer or a sentence between tunes: where this comes from, in order — coastal Colombia, Peru’s chicha, Buenos Aires, Kingston, St Pauls. Credit is the CC BY of gigs. Name the UK’s part honestly: Bristol’s sound systems are a product of the same history as its slave-trade wealth. One plain sentence beats any acknowledgement ritual.
Use the record shop. Put the source records on the counter — Totó la Momposina and Petrona Martínez for the Afro-Indigenous coastal roots (much of that living tradition, bullerengue especially, is carried by women), Discos Fuentes and Los Corraleros de Majagual, Los Mirlos, La Yegros and the ZZK catalogue, King Tubby, Sister Nancy, Smith & Mighty — so someone who liked the band walks out with the original. Money flowing down the lineage rather than pooling at the end of it.
Who’s in the room. Invite any Latin American or Caribbean-heritage musicians or DJs nearby onto the bill or the decks and ask what they’d want from the night — as collaborators, not as authentication. Host; don’t represent.
Watch the poster’s language. “Tropical”, “exotic”, “world” are the boxes from the Dewey shelf. Say what it is: British musicians in a Bristol–Brighton lineage playing a Colombian–Jamaican–Argentine form. Transparency about the synthesis is the decolonial move; pretending to be Colombian isn’t.
Leave it open. Sound-system culture is participatory. A slot where locals bring a record, or a section that’s a jam rather than a set, is Musicircus in a shop.
Follow the money. Credit composers on covers, split the door fairly, and if there’s surplus, send some of it somewhere in the lineage. Sister Nancy’s “Bam Bam” (1982) is one of the most sampled records ever made; she saw no royalties from it for over thirty years. That is what following the money is for.

A room in a record shop where the lineage is named, the originals are for sale and the door money goes somewhere honest is a small radiating thing — the only kind that has ever beaten the other sort.

Rabbit holes

Decolonial thought — Linda Tuhiwai Smith, Decolonizing Methodologies · Sylvia Wynter, “Unsettling the Coloniality of Being” (2003) · Boaventura de Sousa Santos, Epistemologies of the South · Gloria Anzaldúa, Borderlands/La Frontera · Abeba Birhane, “Algorithmic Colonization of Africa” (2020)

Extraction and its critics — Shoshana Zuboff, The Age of Surveillance Capitalism · Kate Crawford, Atlas of AI · Nick Couldry & Ulises Mejias, The Costs of Connection · Safiya Umoja Noble, Algorithms of Oppression · Mariana Mazzucato, The Value of Everything · Bender, Gebru et al., “On the Dangers of Stochastic Parrots” (2021)

The lineage — Naomi Klein, The Shock Doctrine · Rees-Mogg & Davidson, Blood in the Streets and The Sovereign Individual · F. A. Hayek, “The Use of Knowledge in Society” (1945) · Milton Friedman, preface to Capitalism and Freedom (1982 ed.) · Rosa Luxemburg, The Accumulation of Capital · Silvia Federici, Caliban and the Witch · Donna Haraway, “Situated Knowledges” (1988)

Procedure and performance — György Ligeti on Structures Ia (Die Reihe 7, 1960) · John Cage, Musicircus (1967) · Pauline Oliveros, Sonic Meditations (1971) and Deep Listening · Lucy Suchman, Plans and Situated Actions (1987) · Valentino Braitenberg, Vehicles: Experiments in Synthetic Psychology (1984)

Fences — Statute of Anne (1710) · Donaldson v Becket (1774) · Jessica Litman, Digital Copyright · Lawrence Lessig, Free Culture · Pamela Samuelson’s writing on Eldred and fair use · Ruth Okediji on copyright and development · Open Future / IViR study on share-alike and machine learning (2024) · Creative Commons, “Update on CC signals” (April 2026) and “Guidance on Using CC Licenses in an AI Ecosystem” (September 2026) · Software Freedom Conservancy · Tim Berners-Lee, “Charlie Works” (W3C Design Issues, 2025)

Acorns — Elinor Ostrom, Governing the Commons · Robin Wall Kimmerer, Braiding Sweetgrass · Suzanne Simard, Finding the Mother Tree · Rebecca Solnit, A Paradise Built in Hell · Anna Tsing, The Mushroom at the End of the World · Octavia Butler, Parable of the Sower · adrienne maree brown, Emergent Strategy · Ursula K. Le Guin, “The Carrier Bag Theory of Fiction”

The music — Totó la Momposina · Petrona Martínez · La Yegros · ZZK Records · Discos Fuentes · Sister Nancy, “Bam Bam” · Bristol Archive Records (for the St Pauls sound-system history)