A policy for community media in Namibia and two studies of collective healing in the Colombian Amazon appear to belong to different fields.

In Namibia, the scene involves broadcasters, local languages, regulation and funding structures. In Colombia, women who survived the armed conflict paint, embroider, cultivate medicinal plants and rebuild relationships damaged by war.

Both scenes concern infrastructures that sustain memory. One allows voices displaced from national media to remain in public life. The other turns artistic work, mutual support and territorial care into forms of repair.

Artificial intelligence introduces a third scene. Radio broadcasts can become linguistic corpora. Embroidery, testimony and plant knowledge can become digital archives. Yet AI does not merely learn from them. It classifies and recombines them, and it may reproduce them incorrectly. The cloud that stores them also requires electricity, water, minerals, chips, networks and data centres.

Digital memory occupies territory too.

Namibia and the resources required to speak

In April 2026, Namibia approved its Community Media Policy for 2026–2030. The framework was developed with support from UNESCO, the Ministry of Information and Communication Technology, the Communications Regulatory Authority of Namibia and the national community broadcasters’ network.

The policy seeks to provide the institutional foundation the sector had demanded for years. Community radio carries local languages, links rural areas to public debate and creates spaces for people who are often absent from national conversations.

During the drafting process, proposed reforms included access to government advertising revenue, the creation of a representative national body and the expansion of the Universal Access and Service Fund to cover community media equipment and infrastructure. UNESCO’s public account confirms that these measures were proposed, although it does not explain in detail how they were incorporated into final implementation.

Ned Sibeya of the National Planning Commission described community media as the heartbeat of Namibian society. The metaphor is revealing. A local broadcaster is not simply a channel through which news passes. It may be a place where a community hears itself, recognises a language and decides which events deserve attention.

Formalisation nevertheless introduces a tension. Institutional language emphasises viability, capacity and coherent governance structures. These are necessary objectives in a sector frequently constrained by scarce resources. Without stable funding, equipment and decent working conditions, legal recognition may remain declarative.

Yet capacity-building is not neutral either. Someone must decide which form of speech is professional, which broadcasting style is considered reliable and which administrative structure is sufficiently stable. A policy designed to protect community media could end up normalising the practices it was intended to sustain.

Precarity should not be romanticised as an automatic form of autonomy. A lack of resources limits continuity and increases dependence on governments, advertisers or donors. At the same time, some informal forms of organisation have preserved local ways of narrating, participating and making decisions that do not fit easily within corporate models.

The material challenge is to provide resources without turning assistance into a technique of domestication.

AIthropology Lab key point — Media sovereignty is not secured merely by obtaining a frequency. It also requires the capacity to sustain a voice without reshaping it until it becomes recognisable to those who fund, regulate or assess it.

The Colombian Amazon and wounds made visible through embroidery

In Medical Anthropology, “Survivors, Artists, and Healers: Collective Healings of Ecological and War-Related Wounds in the Colombian Amazon”, by César E. Abadía-Barrero and colleagues, studies a collective creative process involving women who survived the armed conflict in the Colombian Amazon.

Painting, embroidery and installation made it possible to express losses, pain and expectations for the future that could not easily be reduced to conventional clinical protocols. The process culminated in Casa Común, an artwork bringing together personal and collective memories.

Families and friends appear in the textiles, but so do plants, animals, rivers and Amazonian landscapes. Healing is not represented as an exclusively individual recovery. It takes place through relationships with other people and with a territory that has also endured war.

Territory is not an inert backdrop against which conflict occurred. It forms part of the wound. Displacement, environmental destruction, the loss of crops and the breaking of community relations belong to the same process.

A related article by Vanesa Giraldo-Gärtner, Sandra González Sanabria and Edna Rocío Cortés Guzmán, “Plantas medicinales que sanan heridas de guerra: una experiencia de reconciliación interespecie”, examines the cultivation of medicinal plants in Caquetá as a practice of peacebuilding.

Plants act as mediators around which victims, former combatants and farming communities can rebuild trust and everyday forms of care. Cultivating, healing and remembering cease to be separate activities. A plant is not merely a store of biochemical properties. Its social efficacy also depends on the knowledge, affects, rhythms and obligations formed around it.

The value of these practices does not arise from their recognition as professional art or their incorporation into a cultural market. It emerges from shared time, repeated gestures and the possibility of remaining alongside other people.

An embroidered piece may be an artwork, but it is also a relationship.

AIthropology Lab key point — Repair does not occur only when an institution recognises harm. It is also produced through the slow practices by which a community reconnects bodies, memories, plants and territory.

Project time and care time

Namibia’s policy is organised around a defined period from 2026 to 2030. It requires budgets, objectives, indicators and evaluation dates. This is the linear time of public planning.

The territorial healing described in Colombia follows another temporality. It depends on cultivation, returning memories, life cycles and the gradual rebuilding of trust. It cannot be guaranteed by a delivery schedule.

Both forms of time may be necessary, but they are not equivalent.

An institution needs to plan. A community needs to continue after a budget, a cooperation project or a research programme has ended. The contradiction arises when administrative time attempts to replace the time of life.

Institutions tend to measure completed activities, participant numbers, materials produced and outcomes achieved. Communities may instead value the recovery of a language, the continuity of a relationship or the possibility of returning to a place without fear.

When the indicator replaces the process, whatever does not fit the timetable begins to look like failure.

The words that disappear

In Anthropology News, Sarah Duignan examines linguistic restrictions affecting environmental health assessments in the United States in “Erasing Words, Erasing Health”.

The list discussed in the article includes terms such as Indigenous community, climate change, gender-based violence, women, pollution, safe drinking water, sociocultural and accessible. When these words disappear, the vocabulary of reports is not the only thing that changes. It also becomes harder to ask particular questions and document specific experiences.

An assessment that cannot name an Indigenous community, polluted water or gender-based violence does not become neutral. It becomes unable to record the relationships that produce harm.

Every removed word closes a line of inquiry. What can no longer be named becomes harder to recognise, measure and defend. The epistemic loss described by Duignan belongs to a wider logic. Administrative and digital systems decide what can appear as a legible reality.

This scene helps connect Namibia and Colombia. A language that does not reach national radio loses public presence. An affective, spiritual or medicinal relationship with territory that does not enter a form may remain outside reparation policy.

Invisibility is also produced through classification.

When voices become data

Artificial intelligence could extend the reach of community media. Transcription and translation systems may help programmes circulate in local languages, improve accessibility and contribute to the preservation of oral records.

They may also help organise archives concerned with memory, biodiversity or community health.

These possibilities nevertheless depend on who designs the tool, who controls the data and which benefits return to those who produced the knowledge.

An Associated Press and Grist report on the 2026 United Nations Permanent Forum on Indigenous Issues captures this ambivalence. AI may support language revitalisation and territorial monitoring, while also extracting traditional stories, medicinal knowledge, genetic information and land data without consent.

The CARE Principles for Indigenous Data Governance identify four central dimensions — collective benefit, authority to control, responsibility and ethics. Their starting point is that making data findable, accessible and reusable is not enough. Historical inequalities, the purpose of collection and the authority of peoples over their knowledge must also be considered.

This discussion bears directly on both cases.

Broadcasts produced by a Namibian community radio station may be treated as training data. A model could learn vocabulary, pronunciation and local narratives from years of community work. Technical availability, however, does not amount to permission for reuse.

In the Colombian Amazon, embroidery, testimony and plant knowledge may also be digitised. Once separated from those who produced them, a system may turn them into visual patterns, emotional categories or reusable information.

Digital extraction begins when knowledge can be copied without retaining the obligations that accompany it.

The cloud is thirsty

The word cloud produces an illusion of weightlessness. It seems to describe a space without soil, borders or matter.

There is, however, no artificial intelligence without data centres and no data centres without energy. The International Energy Agency’s Energy and AI examines the electricity required by the expansion of AI and its implications for energy security, emissions and affordability.

The United Nations Environment Programme adds further layers. Data centres consume water to cool equipment, generate electronic waste and depend on critical minerals and rare elements whose extraction may cause environmental harm. Manufacturing chips and devices also requires large quantities of raw material.

The paradox is difficult to ignore.

Communities in Namibia and Colombia may turn to digitisation to protect a language, preserve a memory or defend a territory. By uploading those records to commercial AI infrastructures, however, they also participate indirectly in a material system that consumes energy, water and minerals.

AI does not merely learn from territories. It also requires territories in order to exist.

There is no direct line that allows each radio file to be attributed to a mine, or each digitised embroidery to a specific data centre. The infrastructure operates through global chains that are difficult to trace. That is precisely why the abstraction of the cloud is politically effective. It separates the apparently immaterial act of storing or processing information from the material conditions that make it possible.

The same technology presented as a means of preserving memory may depend on extractive models that weaken the ecological conditions required to keep that memory alive.

A policy of digital sovereignty should therefore consider not only control over content, but also the source of energy, water use, equipment lifespans and the mineral chains on which processing depends.

AIthropology Lab key point — Data sovereignty remains incomplete if it does not also ask about the energy, water and mineral sovereignty of the infrastructure that processes those data.

The machine invents too

The danger does not end with appropriation.

Generative systems do not retrieve memories in the manner of an archive. They produce responses through statistical relationships between fragments of information. They may therefore fill gaps, mix varieties or present as authentic a formulation that has never existed.

The studies Multilingual Hallucination Gaps in Large Language Models and Investigating Hallucination in Conversations for Low Resource Languages show that errors and hallucinations are not distributed evenly across languages. Performance varies according to language, resource availability and task.

This pattern points towards a particular risk. In a poorly documented language, a fluent-looking answer may mix varieties, introduce structures from a dominant language or invent plausible forms. The result may appear correct to people outside the linguistic community.

The machine may not only appropriate memory. It may produce a simulation of memory.

If that output circulates more widely than community sources, error may acquire authority through repetition, visibility and ease of access. A language with little digital presence risks being publicly represented by the probabilistic version produced by a model.

Technological preservation may therefore become an unintended form of replacement.

The right not to produce data

Data sovereignty is often expressed as the right to control who gathers information, where it is stored and for what purpose it is used.

There is also a prior right — the right not to produce the data.

Not all knowledge needs to be preserved through permanent records. Silence, secrecy, limited transmission and conditional access may operate as technologies of care. Some stories are shared only at particular times, with specific people or within relationships carrying responsibilities.

Local Contexts offers a practical translation of this principle. Its Traditional Knowledge Labels allow communities to specify conditions of use, including seasonal restrictions, secret or sacred materials, limited-access content, community-only uses and non-commercial conditions.

Recording a ceremony changes the ceremony. Archiving secret knowledge alters the conditions of its existence.

A Namibian community should be able to decide that a broadcast will not be recorded or that a narrative may not enter a linguistic corpus. A woman in Caquetá should be able to embroider without her work entering an image bank used to train emotion-recognition systems.

Authority over data includes the possibility of interrupting their circulation.

We might call the right to digital escape the collective capacity to withdraw from circuits of capture, prevent further uses, demand the deletion of records and keep certain practices outside algorithmic representation.

This is not merely a matter of individual privacy. It is a form of cultural and territorial sovereignty.

Opacity is not ignorance or the absence of knowledge. It may be a condition under which knowledge continues to belong to the relationships that sustain it.

Can AI help communities unlearn?

A more ambivalent possibility remains.

Rather than using AI only to preserve tradition, a community might use it to examine the discourses introduced into its archives by colonialism, war or bureaucracy. It might compare decades of broadcasting, administrative documents and educational materials to identify when particular vocabularies displaced other ways of naming territory.

A locally governed tool might also locate stereotypes, reveal absences or show which voices have dominated an archive.

In this limited sense, AI might assist processes of unlearning.

An algorithm cannot, however, decide which part of a language is authentic or separate on its own what is local from what has been imposed. Languages always contain exchanges, conflicts and transformations. Turning decolonisation into an automated cleaning of vocabulary could produce new forms of purism and surveillance.

Such a tool would make sense only if it remained subordinate to community decisions, acknowledged uncertainty and allowed its results to be rejected.

A possible cultural detox would not involve asking a machine to purify a language. It would involve using it provisionally to make relations of power visible before they are interpreted collectively.

AI might help to find the layers. It should not decide which ones must be torn away.

From the right to speak to the right to withdraw

In Namibia, the principal mediation takes the form of a public policy supported by an international organisation. In Colombia, it passes through bodies, needles, textiles, seeds and plants.

Artificial intelligence adds another translation. It turns voices, images and relationships into processable units.

Every translation allows something to circulate, but it may also remove part of its context. The right to communication and the right to territory therefore converge in a wider set of rights.

The right to decide what becomes data.

The right to know the material infrastructure that processes it.

The right to correct representations fabricated by a machine.

The right to delete, remain silent and withdraw.

A radio station is not merely a collection of audio files. An embroidered piece is not simply an image. A medicinal plant is not a list of compounds.

They are relationships sustained over time.

AIthropology Lab reading

Namibia reminds us that a language requires media, resources and infrastructure in order to remain public.

Colombia shows that memory and healing cannot be separated from bodies, the rhythms of care and territories.

The materiality of artificial intelligence requires us to add that digitisation does not take place outside those territories either. Every processed file enters an infrastructure that consumes electricity, water, minerals and devices.

Technological sovereignty cannot be reduced to placing more languages inside a model. It must include the capacity to govern data, assess material costs, challenge fabrications and keep certain forms of knowledge beyond the machine’s reach.

An AI system able to listen to more languages will not necessarily be more plural.

It must also accept that it has no right to hear everything.

Sources