“Rise up women, ignite your flame… dance your mad, naked dance… and awaken your power to burn the world.”
— Kazi Nazrul Islam, 1920s
Kali does not mourn what is already dead. She counts the skulls, names them, and asks what deserves to live. The information ecosystem is not in decline. It is being systematically replaced. That is a different thing. The Devourer names what is being consumed at industrial scale and at what cost.
There are now hundreds of US counties where the only “local” news published is machine-generated. No reporter attended the council meeting. No one knocked on a door. The AI assembled a story from public records and a wire feed. It is indistinguishable from journalism. That is the point.
A coordinated disinformation campaign requiring a team of 40 and six months in 2018 can now be run by three people in a week. The cost of synthetic content, synthetic distribution, and synthetic credibility has collapsed. Democratic institutions were not designed for this economics.
AI systems now mediating information were trained overwhelmingly on English-language, Western, post-Enlightenment sources. They don’t just have gaps. They have an epistemology — one that structurally renders other ways of knowing inadmissible. This is not a bug to be fixed. It is a design choice that was never consciously made.
What gets seen is determined by engagement optimisation, not epistemic value. Outrage outperforms accuracy. Novelty outperforms significance. The algorithm is the editor. The editor has no name, no address, and no professional obligations. There is no recourse mechanism. There is no public interest test.
The ten largest US journalism foundations account for an estimated 34% of global philanthropic journalism funding. Their editorial priorities — which issues get funded, which geographies, which formats — structurally shape what independent journalism looks like globally. This is rarely named directly.
AWS, Azure, and Google Cloud host most of the world’s media infrastructure. The same companies are content producers, distributors, and AI training data consumers. Amazon hosts the Washington Post, which it owns. No historical precedent exists for this level of vertical integration in information infrastructure.
An unsentimental audit. Dead routes named first.
Automated content at scale. AI-synthesised journalism indistinguishable from human-written copy. This route exists. It is already being walked. The question is not whether it can work as a business. It can. The question is what it does to democratic accountability. Nobody is asking that with sufficient urgency.
Current trajectory. No structural intervention. Philanthropic funding fills gaps selectively. Platform regulation arrives too late. Local accountability journalism continues to collapse. Information deserts expand. Democratic institutions operate without scrutiny. This is not a prediction. It is the baseline.
Permanent capital, insulated from political and market cycles. Modelled on university endowments and cultural trusts. The structural argument: journalism that depends on annual renewal is not independent. Full stop. Norway exceeds €400M. Eight jurisdictions in active development.
Mandatory levy on platform advertising revenue redistributed to verified public interest publishers. Canada’s C-18 attempted this. Meta pulled out. The test proved the dependency — and proved the leverage. A robust version requires international coordination or unilateral regulation with teeth.
Reader-owned, member-governed, structurally independent. Slow to build. Extremely hard to kill. The Guardian’s Scott Trust is the scaled example. Dozens of smaller cooperatives across Europe are proving the model works at community scale. The question is whether it can scale fast enough.
Publishers licensing content to AI training pipelines for revenue share, attribution, or traffic. Currently a race to the bottom between desperate publishers and platforms holding all leverage. A viable version requires collective negotiation — which requires collective organisation that does not yet exist.
The Witness operates like LIDAR — active, pulsed, revealing structure through the dark. Power does not hide itself by being absent. It hides itself by becoming infrastructure. The Witness maps the topography beneath the surface: who owns the pipes, who decides what counts as knowledge, who benefits when certain things remain invisible.
Google’s index is a theory of what knowledge is worth finding. SGE is a theory of what knowledge is worth synthesising. Both are proprietary. Neither is accountable. The people who designed these systems are not elected. The criteria they use are not disclosed. They are, collectively, the largest editorial force in human history.
Algorithmic curation is editorial. The absence of a named editor does not mean the absence of editorial decisions. The myth of neutrality does not describe how platforms work — it protects the power of the curator by making that power invisible. There is no neutral algorithm. There is no neutral index.
Decisions about what to include in foundation models were made by a small number of people and are largely undisclosed. The inclusions and exclusions constitute a theory of knowledge at infrastructure scale. The training data is not documented. The curation choices are not public. The epistemological assumptions are not named.
“Funders keep asking us to prove impact. Meanwhile the thing destroying the ecosystem is growing at 40% per year and nobody is asking it to prove anything.” The asymmetry is structural: accountability is applied to the small and public-funded. It is not applied to the large and privately funded.
The CARE principles (Collective Benefit, Authority to Control, Responsibility, Ethics) offer a fundamentally different theory of data governance than the dominant FAIR framework. Active in 12+ countries. Adopted by 14 national data frameworks. Not one major AI company’s governance document references them. This is not ignorance.
The local reporter with twenty years in a community holds embodied knowledge: presence, relational trust, tacit expertise, somatic knowledge of place. AI can simulate the output. It cannot replicate the knowledge. When we replace the reporter we are not replacing the article. We are replacing the accountability relationship that produced it.
Who benefits from each route succeeding or failing. The Witness names the beneficiaries.
AI-generated content at scale replacing human accountability journalism. Already operational in 200+ US markets. Output indistinguishable from journalism.
Current trajectory. No structural intervention. The baseline that benefits from being unnamed and unexamined.
Permanent capital insulated from political and market cycles. Norway exceeds €400M. Eight jurisdictions in active development.
Mandatory levy on platform ad revenue redistributed to public interest publishers. C-18 proved the leverage. The will was not there.
Reader-owned, member-governed, structurally independent. Proven. Under-used. The Scott Trust at scale is the model.
Publishers collectively negotiating with AI training pipelines. Requires collective organisation that does not yet exist at the necessary scale.
“What must be defended? What is already emerging?”
People from the periphery are not subjects of journalism — they are its producers. As Periferias operates from and for the urban margins of São Paulo. Reaches millions. Zero institutional funding dependency. A demonstration of what journalism looks like when communities own it entirely.
Distributed oral archive spanning 800+ years. Resilience by design. An active model for distributed knowledge infrastructure that has never needed a platform, never needed a server, never needed institutional backing — and has outlasted every institution that tried to contain it.
Te Mana Raraunga, Local Contexts, Global Indigenous Data Alliance. Living frameworks for collective knowledge governance — operational now in multiple jurisdictions. Not waiting for AI regulation to catch up. Already doing what AI governance frameworks haven’t yet imagined.
People trust those near them, known to them, physically proximate — over credentialed intermediaries. These networks were always there. They become visible when the institutional layer is removed. LIEM research finding, UK, 2024. The infrastructure that survives is relational, not institutional.
Permanent capital insulated from political and market cycles. Norway exceeds €400M. Eight jurisdictions in active development. Journalism that depends on annual renewal is not independent journalism. Full stop. The endowment is not a nice-to-have — it is the structural precondition for independence.
“Every community that built something independent when the funding collapsed is a proof of concept. Not a miracle. Not an exception. A demonstration that the model the market said was impossible was simply unprofitable for the market.”
Viable and emerging routes foregrounded. These exist now.
Permanent capital outside market and political cycles. Norway exceeds €400M. Eight jurisdictions developing. Requires political will but infrastructure exists and is building.
Communities own their infrastructure and their data. Te Mana Raraunga operational across 12 countries. Doesn’t wait for institutional permission. Already operational.
Griot-model distributed knowledge infrastructure. No single point of failure. Technology now enables digital equivalents of 800-year oral tradition resilience.
Relational trust infrastructure at community scale. Already exists — becomes visible when institutional layer removed. LIEM evidence base established, 2024.
Wait for market incentives to align with public interest journalism. The market has had 30 years. The market has answered.
Annual platform or philanthropic renewal. Creates incentive structures incompatible with accountability journalism. Documented pattern of editorial compromise.
“Where are we in the cycle? Who has seen this before?”
“India is an endless cremation ground… Meditate on great power in the vira pose.”
The nation summoning Kali against colonial information control. A political movement weaponising the iconography of destruction as liberation. The printing press and the pamphlet as epistemic rupture weapons — in both directions simultaneously.
The Annals were written so that no reigning king could read them until after his death. Structural independence built into the architecture of the record itself — not into the law, which can be rewritten. The most radical press freedom mechanism in history. Not a protection of speech. A structural impossibility of interference.
A civilisation that built 5,125-year cycles into its knowledge infrastructure as a matter of cosmological necessity. The current generative AI disruption is approximately four years old. Where does that sit in a long count? The dominant rupture narrative is Western and linear. It cannot see cycles it does not measure.
The printing press ruptured the manuscript economy in 1440. By 1557 — 87 years later — a guild had formed, concentrated power in London, enabled censorship, and created a stable model lasting 150 years. The window for new institutional forms had closed. New concentration had replaced old concentration. The question for now: when does our window close?
Every epistemic rupture creates a window when new institutional forms become possible. Power reconcentrates faster than reformers expect. The printing press gave us both the Reformation and the Index Librorum Prohibitorum. The internet gave us both Wikipedia and Cambridge Analytica. The pattern is not a prediction. It is a pressure. The window is open. It will close.
800+ years of distributed oral knowledge infrastructure. No server. No algorithm. No platform. No single point of failure. Still operating through every rupture the dominant framework has experienced. The griot tradition is not a historical artefact. It is a living model — and a proof that distributed, relational knowledge infrastructure has already solved what the current rupture makes urgent.
Every rupture creates a window. The question is what forms get built in it, and by whom.
The window is open. Eight endowment jurisdictions developing. Sovereign data frameworks operational. Community models scaling. This is what the window looks like when something fills it.
Joseon Sillok principle: independence built into architecture, not law. Decentralised infrastructure, distributed archives, structural impossibility of single-point capture.
GDPR as precedent: 26-year lag, imperfect, but operational. AI regulatory frameworks forming across multiple jurisdictions. Slow, but the printing press took 87 years.
Distributed, relational, no single point of failure. Community-owned. 800-year proof of concept already exists. Technologically implementable now. The model is not new — only the substrate.
Historical default. Power fills the vacuum faster than reform. The Stationers’ Company route. Already underway: six platforms control 80% of AI training data. The window is also being closed from the inside.
No structural intervention. The window closes. New institutions form around existing power concentrations. The Index Librorum Prohibitorum rather than the Reformation. This is the baseline.