Who Governs What Becomes Visible

Three initiatives concerning Indigenous language data, audiovisual heritage, and Kiswahili show that visibility is never mere exposure. It depends on rules governing access, format, quality, and authority, and those rules also shape the artificial intelligence that may be built afterwards.

This Radar examines artificial intelligence alongside the social infrastructures that produce its data, categories, and representations. Some of the cases are directly concerned with AI systems. Others show the prior work through which a language or cultural practice becomes recordable, comparable, and reusable material.

Making something visible may mean recognising a language, widening the circulation of a tradition, or improving public access to knowledge. It also involves selection. Someone must decide who may record, which version is retained, the format in which it circulates, and the conditions under which another institution may use it.

Three recent initiatives make that process visible. UNESCO has opened an incubator for Indigenous-led governance of language and cultural data. In Indonesia and Timor-Leste, an ichLinks update presents the results of a 2024 programme that turned living heritage practices into short films. In East Africa, the Bujumbura Resolutions position Kiswahili as strategic infrastructure for AI and announce the development of a regional certification system.

Across all three cases, visibility appears less as a condition than as a relationship of power.

Data Do Not Become a Commons Simply by Being Open

On 27 July, UNESCO opened applications for the Indigenous Language Data Commons Incubator, an initiative for Indigenous-led teams seeking to develop community governance systems for language and cultural data in the age of AI. Initial concept notes may be submitted until 14 August 2026.

The call was co-designed with a global steering committee of 19 Indigenous language-data experts and will be implemented with The GovLab and Microsoft. Selected teams will receive mentoring, technical guidance, and capacity-building support. At present, this is an incubator rather than an operating infrastructure whose outcomes have been evaluated. The available information describes its intentions, but does not yet tell us which communities will take part, which projects will receive support, or how concrete disagreements over data control will be resolved.

UNESCO’s framing is significant. The problem is not presented only as a shortage of data for training models. The page identifies risks of extraction, misuse, and loss of control over cultural knowledge. A data commons therefore appears as a framework for collective decisions about how materials are collected, accessed, and used, and about how benefits may return to communities.

This approach is close to the CARE Principles for Indigenous Data Governance: Collective Benefit, Authority to Control, Responsibility, and Ethics. CARE emerged as a counterweight to open-data models primarily concerned with making information findable, accessible, interoperable, and reusable. Technical openness does not by itself resolve the historical inequalities between those who produce knowledge, those who store it, and those able to turn it into scientific or commercial products.

The distinction matters for AI. A corpus may be accessible while having been assembled without adequate consent. It may be technically well structured while including knowledge that a community does not consider public. It may help build a translator or speech model while transferring authority over the language to the institution that maintains the servers, licences, or tools.

The word ‘commons’ does not remove internal disagreement either. Communities are not homogeneous actors. Generations, territories, organisations, families, and authorities may disagree about what should be shared, who may represent the collective, and how long consent remains valid.

The partnership with Microsoft introduces another tension. Access to technical capabilities and funding may enable projects that would otherwise struggle to proceed. At the same time, an infrastructure described as community-governed may depend on standards, services, and specialist knowledge controlled elsewhere. The call is not evidence of corporate appropriation, but neither does it demonstrate that this asymmetry has been resolved.

Sharing data is not the same as surrendering the authority to decide over them.

The meaningful measure will not be the quantity of material entering a repository, but the capacity retained by the Indigenous Peoples involved to establish permissions, restrict reuse, correct representations, withdraw material, and participate in the value those data produce.

A Practice Enters the Frame

On 29 July, with an update the following day, UNESCO published a detailed account of the ichLinks video competition in Indonesia and Timor-Leste. The publication date may make it appear to be a new initiative, but the page itself identifies the project period as 2024. It should be read as a retrospective update on its results, not as a programme launched this week.

The call received 59 proposals: 37 from Indonesia and 22 from Timor-Leste. A joint jury selected ten finalist teams. They took part in workshops, received microgrants and technical mentoring, produced short films, and presented their work at public screenings in Yogyakarta and Dili.

The programme began from an inequality of recognition. In Yogyakarta, living practices associated with heritage are often overshadowed by the monumental visibility of the Cosmological Axis. In Timor-Leste, tais weaving attracts much of the international attention while other cultural expressions have a weaker public presence.

The ichLinks response was to train young people, students, cultural practitioners, and content creators to document this heritage through video. The project presents digital storytelling as a form of safeguarding and intergenerational transmission. A practice does not remain intact when it enters a camera, however.

Producing a short film requires a beginning and an ending, the choice of protagonists, an ordering of actions, the selection of sounds, and an explanation for an audience that may not know the context. A ritual relationship, craft, or celebration becomes a narrative unit with a duration and a point of view.

This process may open space for self-representation. Young people cease to appear solely as recipients of an inherited tradition and become editors and mediators of its public circulation. Yet the visibility they gain also depends on learning which images, rhythms, and stories the audiovisual format recognises.

Gaining visibility also means learning how to become visible within a format.

The UNESCO page explains selection, training, mentoring, and final dissemination. It does not specify who retains copyright in each recording, how the consent of everyone filmed was documented, or what capacity communities retain to restrict later reuse. The absence of this information from the page does not establish that protocols were absent. It identifies an issue left outside the institutional account.

This case is not directly about artificial intelligence. Its relevance to the Radar lies in the audiovisual infrastructure that precedes many digital systems. Before a model can classify, recommend, or generate heritage images, someone must produce archives, descriptions, and selections. The visible collection does not represent everything that exists. It represents what was chosen, recorded, edited, and circulated.

The camera expands the archive, but it also participates in its form.

A Language Becomes Strategic Infrastructure

Between 5 and 7 July, public officials, specialists, academics, technology practitioners, young people, publishers, and regional organisations gathered in Bujumbura for the Third EAKC International Kiswahili Conference and the East African Community’s fifth World Kiswahili Language Day celebration. The East African Community published the outcomes on 17 July under an explicit heading: placing Kiswahili at the centre of the artificial intelligence era.

The Bujumbura Resolutions on Kiswahili, Multilingualism and AI propose treating the language as strategic digital infrastructure. Partner States called for a regional roadmap to digitise East African languages, shared corpora and open language datasets, sustained mechanisms for standardising AI terminology, expanded scholarships and university collaboration, and African ownership of the language data powering AI systems.

The institution estimates that Kiswahili is spoken by more than 250 million people. Such reach does not eliminate digital inequality. A language may have extensive oral, educational, and political circulation while remaining underrepresented in speech tools, translation, search, or text generation.

One of the most concrete announcements was the beginning of work on the Kiswahili Trust Mark, or Alama ya Uthibitisho wa Kiswahili. The proposed certification is intended to assess independently how well language models understand, speak, and serve Kiswahili. According to the East African Community, criteria will include linguistic quality, terminological accuracy, cultural context, and safety. A public register of certified systems is also envisaged.

The Trust Mark is not yet an operating mechanism. Consultations with regional institutions, Kiswahili councils, and technology companies are expected to shape the framework over the coming year. Presenting it as an established standard would confuse a political commitment with a technical outcome.

The proposal changes what counts as model evaluation. It would not be enough for a tool to produce grammatical sentences or achieve an aggregate score. It would also need to answer to regional criteria concerning terminology, context, and social usefulness.

Certifying a language nevertheless requires decisions about who may speak in its name. Kiswahili contains varieties, territorial histories, urban registers, and cross-border uses. Standardisation may support education, public services, and technological development. It may also turn particular institutions or varieties into the measure of correctness for all others.

This dilemma does not invalidate certification. It reveals its political dimension. A trust mark does not merely measure a machine; it distributes authority among regional bodies, universities, language councils, companies, and communities of speakers.

The initiative also announced a new compendium of AI terminology in Kiswahili and a peer-reviewed academic journal, Jarida la KAKAMA. Producing technical vocabulary and scientific knowledge in the language itself prevents the AI debate from depending entirely on later translation from English. At the same time, each term that becomes established participates in constructing what technology will come to mean in the region.

A language does not enter AI only as data. It also enters as an institution, a norm, and a political project.

The regional ambition matters because it is not limited to requesting inclusion in models developed elsewhere. It seeks to construct its own criteria of quality and control. Its practical reach will depend on funding, effective participation by diverse speakers, openness in the certification process, and control over the technical infrastructure.

Governing Visibility

The three cases operate at different scales. The data incubator is a global call without results yet. ichLinks is presenting a 2024 audiovisual project in 2026. The Bujumbura Resolutions announce a regional agenda and a trust mark that remains to be developed.

Their relationship does not lie in all three using artificial intelligence. It lies in the infrastructures through which something becomes visible and usable by digital systems.

In the first case, language data become a governed resource organised through permissions, responsibilities, and expectations of collective benefit. In the second, a cultural practice becomes an audiovisual sequence prepared for circulation. In the third, a language becomes strategic infrastructure and an object of technological assessment.

Visibility may correct historical exclusions. It may also produce new ones. Knowledge that is not digitised may remain outside a model, while knowledge digitised without community authority may become available for unwanted uses. A little-known practice may reach new audiences, while its filmed version overshadows less spectacular variations. A language may gain tools and recognition while a technical norm reduces its internal diversity.

Governing what becomes visible therefore involves more than deciding what is shown. It includes establishing:

who authorises recording;
who controls access;
which formats organise representation;
which criteria determine quality;
who may correct it;
which benefits and responsibilities follow from reuse.

Artificial intelligence appears at the end of some of these chains and in the middle of others. It uses materials already organised, but its outputs later return to the social world. A certified model may reinforce a linguistic norm. A recommendation system may make particular videos more visible. A database may shape funding and language-revitalisation policy.

Who establishes the conditions under which a language or practice may be seen, used, and recognised as valid?

Lines to Follow

Which teams are selected by the incubator and which concrete powers they retain over licences, access, and withdrawal of data.
How rights, consent, and reuse are governed for videos produced by digital-heritage initiatives.
Who participates in designing the Kiswahili Trust Mark and how it represents different varieties, registers, and linguistic contexts.

References

Carroll, S. R., Garba, I., Figueroa-Rodríguez, O. L., Holbrook, J., Lovett, R., Materechera, S., Parsons, M., Raseroka, K., Rodriguez-Lonebear, D., Rowe, R., Sara, R., Walker, J. D., Anderson, J. and Hudson, M. (2020). The CARE Principles for Indigenous Data Governance. Data Science Journal, 19, 43.

East African Community. (2026, 17 July). East Africa Puts Kiswahili at the Heart of the Artificial Intelligence Era as Bujumbura Hosts the 5th EAC World Kiswahili Language Day Celebrations.

Global Indigenous Data Alliance. (n.d.). CARE Principles for Indigenous Data Governance.

UNESCO. (2026, 27 July). Call for Applications: Indigenous Language Data Commons Incubator.

UNESCO. (2026, 29 July; updated 30 July). ichLinks Video Competition: Capturing the Intangible Cultural Heritage.