In this Radar entry, technology does not refer only to generative artificial intelligence. It also includes digital dictionaries and corpora, employment platforms, recommendation systems, telephones, alarms, and social media. AI occupies a more specific area within that wider field: models that classify, predict, or generate from previously organised data.

This distinction reveals a shared operation. Before a system can process a language, a work trajectory, or a tradition, it must turn them into something legible. Legibility does not mean understanding them in full. It means translating their complexity into orthographies, categories, fields, images, sounds, and labels that an infrastructure can store and compare.

Today’s Radar follows three different situations. In Central America, Mayangna and Tawahka representatives are working on linguistic varieties, curricula, and dictionaries. In Ethiopia, research shows how prolonged unemployment suspends transitions into adulthood while digital platforms promise to improve connections between people and vacancies. In Indonesia, a religious practice loses part of its original function as personal alarms spread, then reappears as festival, heritage, and social-media content.

The three cases allow us to follow the same sequence without erasing their differences. A representation is produced first. Technology capable of classifying, recommending, or recreating it arrives afterwards. The story does not end there, however. Once deployed, systems also reorganise what supplied their data: they reinforce certain linguistic varieties, make particular skills visible, and circulate some images of the past more widely than others.

Before the Model, Someone Must Decide How a Language Is Written

From 13 to 17 July, the Sixth Honduras–Nicaragua Binational Meeting on the Revitalisation of the Mayangna Language is bringing together Mayangna specialists, teachers, linguists, and technical staff from Nicaragua with Tawahka representatives from Honduras. Its working groups are examining linguistic varieties, updating curriculum documents, and preparing dictionaries for four varieties. The report was published by the Fund for the Development of the Indigenous Peoples of Latin America and the Caribbean.

The meeting is not announcing an automated translator or an artificial intelligence model. Its technological relevance lies at an earlier stage.

Before a language can be incorporated into a keyboard, search engine, spellchecker, educational application, or natural language processing system, it requires some form of structured documentation. Orthographies must be agreed, variations described, words related to meanings, and materials organised.

These tasks are not neutral. Standardisation may facilitate teaching and intergenerational transmission, but it may also turn one variety into the reference point for all the others. A dictionary can strengthen a language while fixing boundaries around situated uses, overlapping forms, or different territorial agreements.

The question of legibility also appears in a paper published in July at AmericasNLP. Its authors describe the gap between specialist tools — focused on annotation, morphological analysis, and archiving — and community-facing applications designed around accessibility and learning. They also identify barriers including the cost of computational expertise, single-user workflows, and limited data governance.

In response, they present langlit, a collaborative platform with a searchable corpus, editable dictionary, annotation provenance, configurable access controls, and optional language-model integration with transparent data handling. The paper is available through the ACL Anthology and its DOI.

Who decides which form of a language becomes legible to the machine?

AI does not receive a language intact. It receives a representation produced through meetings, schools, dictionaries, categories, and territorial negotiations. By the time training begins, many cultural decisions have already been made.

The model may then act back upon them. If one variety has more examples, better tools, and a stronger presence in applications, its technical advantage may become linguistic authority. A choice originally made to support documentation may eventually influence how younger generations learn, write, and recognise the language.

A Platform Can Organise Opportunities, but It Cannot Create the Future

Research published on 5 June examines the experiences of 21 young people living with long-term unemployment in Hosanna, Ethiopia. The authors, based at the universities of Gondar and Bahir Dar, show that prolonged unemployment produces consequences extending far beyond lost income. It disrupts socially recognised transitions into adulthood, erodes aspirations, increases dependency, weakens social cohesion, and contributes to political disillusionment and risky migration. The article is available from Discover Sustainability and through its DOI.

The study does not investigate artificial intelligence or employment platforms. It is included in the Radar precisely as a contrast to promises about the digitalisation of employment. It helps distinguish which part of the problem a matching infrastructure can address and which part remains beyond its reach.

In Ethiopia, HaHuJobs brings previously scattered vacancies together on a digital platform. A World Bank analysis reports that this aggregation has improved job-finding rates. The same article describes a wider international ecosystem in which algorithms match profiles with vacancies, recommend alternative occupations, or help translate experience into CVs that employers can read.

The African Union’s Continental Artificial Intelligence Strategy argues that AI may stimulate new industries, innovation, and higher-value employment through an Africa-centred, inclusive, and responsible approach.

The contrast with Hosanna requires two dimensions to be kept separate. A platform may reduce the distance between an existing vacancy and a person seeking work. It can organise advertisements, facilitate applications, and improve the circulation of information. On its own, it cannot diversify a local economy, create enough dignified jobs, or repair the social effects accumulated during years of waiting.

Employment legibility also produces its own exclusions. To enter a system, a work trajectory must be converted into recognisable fields: education, years of experience, skills, location, and availability. Agricultural work, care, informal activities, or knowledge acquired outside institutions may be represented poorly. People with unstable connectivity, limited command of the interface language, or difficulty fitting their experience into prescribed categories are also placed at a disadvantage.

Looking for work is not simply a matter of matching skills with vacancies. It also organises dignity, recognition, family projects, urban belonging, and expectations about the passage of time.

An algorithm may accelerate a search. It cannot return the years during which a person was suspended between youth and a socially deferred adulthood.

The Ethiopian case is not direct evidence about the effects of AI. It provides an empirical limit to technological promises. Efficient matching is not the same as job creation, and improved intermediation cannot solve a structural scarcity of opportunities.

Platforms may also do more than represent the labour market. Their categories, recommendations, and metrics can influence which courses are offered, which skills people attempt to acquire, and which profiles employers come to regard as desirable. The initial representation therefore acts back upon the world it was meant to describe.

When an Alarm Is No Longer Necessary, Heritage May Begin

Across different parts of Indonesia, communities maintain practices intended to wake people before sahur, the meal eaten before the daily Ramadan fast. These take local forms including ronda sahur in Java, sahur bagarak in South Kalimantan, arakan sahur in Jambi, and keliling sahur in Langkat.

Traditionally, people used instruments such as the kentongan, bedug, and rebana to mark the time collectively. An article published on 30 June explains that telephones, clocks, and personal alarms have displaced this practical function. The custom has not disappeared. It has been reconstructed as festival, Ramadan procession, artistic performance, community soundscape, and social-media documentation. The study was published by Heritage of Nusantara and has the DOI 10.31291/hn.v15i1.887.

Technology works paradoxically here. It first makes the collective act of waking people less necessary. It then supplies the devices through which that same act can be recorded, edited, shared, and recognised as heritage.

The practice no longer serves only to indicate a time. It begins to represent local identity, religious solidarity, and continuity between generations. It is preserved not by remaining unchanged, but by acquiring another function.

Between the personal alarm and generative AI lies a decisive intermediate stage: digital documentation. Cameras and platforms turn routes, instruments, sounds, and participants into materials that can circulate. This circulation does not include every version equally. It tends to favour those that are more visible, spectacular, or compatible with the logic of each platform.

An article published on 20 May in Culture Unbound allows us to follow the next stage. It analyses public responses to AI-generated images of Indonesian historical figures and sites circulated by the AI Nusantara account on TikTok. Comments are organised around accuracy, technological curiosity, and shared knowledge. Audiences compare the reconstructions with academic references, cultural expectations, and local knowledge. The article is available from Culture Unbound and through its DOI.

AI does not simply reproduce a heritage that has already been defined. It produces new images that participate in how the past is imagined, discussed, and recognised. Authenticity no longer depends only on resemblance to a historical object. It also matters who selected the sources, which absences are present in the model, which authority validates the reconstruction, and how communities connected with what is represented are able to intervene.

The digital documentation of sahur is not the training dataset used in that study, and there is no evidence that it is currently being used to generate new images. The relationship is prospective. It shows how material selected for social media today may become a reference, archive, or dataset for other systems tomorrow.

Technology does not archive a tradition without transforming it into a particular version of itself.

Once produced, that version may return to the practice. The most widely circulated images can influence later festivals, tourist expectations, institutional reconstructions, or new generations of automated content. Digital memory does not merely preserve. It also proposes a canon.

Legibility Is Never the End

The three cases do not describe the same process. The Mayangna meeting is not developing an AI system. The Ethiopian research does not evaluate digital platforms. The study of sahur is primarily concerned with modernisation, heritage, and social media.

Their relationship lies in the social infrastructures that precede models.

In the first case, a language becomes orthographies, documented varieties, curricula, and dictionaries. In the second, a work trajectory becomes a profile composed of skills, experience, and availability. In the third, a practice becomes images, sounds, and descriptions able to circulate beyond their original context.

AI operates on those representations. Its results depend on who established the categories, what was left out, who may correct the data, and which institution acquires authority to speak in the name of a language, a work trajectory, or a tradition.

Saying that technology arrives afterwards does not mean that everything has already been determined. Once deployed, it acts back upon previous representations, ordering them and amplifying some above others. Its outputs become new data, expectations, and criteria for subsequent decisions.

Technology may strengthen linguistic transmission, widen access to opportunities, and keep a practice alive. It may also fix one variety, reduce a life to a profile, or turn the most spectacular version into the dominant representation of heritage.

It does not preserve without selecting.

It does not connect without classifying.

It does not represent without producing authority.

The anthropological question begins before the model and continues after its deployment:

Who can decide how a form of life becomes information, and what happens when that information begins to make decisions about it?

Lines to Follow

  • Which languages, forms of work, and practices remain outside because they never become legible data?
  • What capacity do communities and individuals retain to correct, limit, or withdraw the representations that feed a system?
  • How do original practices change when classifications, recommendations, and generated images return to them?

References

African Union. (2024). Continental Artificial Intelligence Strategy.

Ali, M. N., Lestari, R. and Suhara, A. I. (2026). Waking the Community for Sahur: Ramadan Heritage, Social Solidarity, and Cultural Transformation in Indonesia. Heritage of Nusantara, 15(1), 156–177. https://doi.org/10.31291/hn.v15i1.887

Atmadiredja, G., Wikayanto, A., Ocktaviana, S., Hartanto, A. and Cahyani, D. A. (2026). The Use of Artificial Intelligence in Visualizing Historical and Cultural Objects on Social Media: A Sentiment Analysis of Public Reactions to AI Generated Images of Indonesian Heritage. Culture Unbound, 18(1). https://doi.org/10.3384/cu.5537

Erkie, A. T., Moges, M. D. and Eshete, B. M. (2026). Sustainable futures at risk among long-term unemployed youth in Hosanna City, Ethiopia. Discover Sustainability. https://doi.org/10.1007/s43621-026-03596-w

FILAC. (2026, 15 July). Honduras y Nicaragua fortalecen acciones conjuntas para revitalizar la lengua mayangna.

Haberland, C., Crowther, C., Qu, J. and Centellas, A. (2026). Bridging Digital Tools for Linguistic Documentation and Revitalization. In Proceedings of the Sixth Workshop on NLP for Indigenous Languages of the Americas, 22–32. Association for Computational Linguistics. https://doi.org/10.18653/v1/2026.americasnlp-6.3

Meyer, C., Proffen, C., Madhvani, S. and Marguerie, A. (2026, 17 February). Can digital solutions enable more inclusive labor markets? World Bank Blogs.