Colombia wants to build artificial intelligence of its own. In May 2026 it activated the governance framework for two Sovereign Foundation Models: Bachué, oriented towards health and well-being, and Chiminigagua, focused on sustainability, territory and the environment.

The announcement allows us to pose a question that comes before the model. Who decides what health is, what territory is, which forms of knowledge count, and which lives must be turned into data before a machine can recognise them?

This is not an abstract question.

On the Colombia–Venezuela border, DignifAI provides AI data-annotation services through a labour-inclusion initiative aimed at Venezuelan migrants. People there prepare, annotate and verify information used in AI-training processes. Carlos Andrés Arroyave Bernal and Oscar Javier Maldonado Castañeda studied precisely this relationship between labour, migration, dignity and artificial intelligence.

Between these two scenes —sovereign models and the people who perform data work— lies one of the questions we want to follow at AIthropology:

not only what an artificial intelligence can do, but what world it needs in order to exist and what world it helps to produce.

In Colombia, that question has a particularly fertile genealogy.

An intuition that began in Cyberia

In 1994, Colombian anthropologist Arturo Escobar published Welcome to Cyberia: Notes on the Anthropology of Cyberculture in Current Anthropology. Long before generative systems entered everyday life, he proposed studying information and biological technologies as cultural phenomena rather than simply as external tools arriving to transform society.

The intuition is decisive.

Technology does not appear only after a society has decided what a person, a territory, a disease or a job is. Technology also participates in producing those categories.

Escobar’s later work took this argument towards the pluriverse and ontological design. In Designs for the Pluriverse, he argues that design is not merely about making objects or solving problems. Design participates in the creation of particular forms of existence.

Applied to artificial intelligence, this becomes especially powerful.

A model selects variables, classifies phenomena, establishes relationships, optimises objectives and enables some actions while making others more difficult. Before it can calculate, it must first decide what can be made calculable.

Every AI system therefore contains a theory of the world.

The political question does not begin afterwards, when we try to regulate the algorithm. It is already inside its architecture.

What does well-being mean for Bachué?

What counts as territory for Chiminigagua?

Which forms of knowledge will enter their datasets?

Who will be able to intervene in those definitions?

The Ministry of Information and Communications Technologies presents the governance of these models as a process bringing together the state, academia, the health sector, companies and other actors to contribute use cases and specialist knowledge. Escobar’s question lets us go one step further: can an AI genuinely be sovereign if it reproduces only one way of turning the world into data?

Not a school, but a constellation

There is no “Escobar school” in any strict sense.

Myriam Jimeno, Diana Obregón, Derly Sánchez Vargas and Carlos Andrés Arroyave Bernal have developed independent trajectories across anthropology, history, sociology and science and technology studies.

Yet their work allows us to construct a Colombian constellation that is particularly useful for thinking about artificial intelligence.

Escobar asks what worlds technology produces.

Jimeno helps us ask what is lost when an experience becomes data.

Obregón investigates how a form of knowledge acquires the authority to define reality.

Sánchez Vargas shows how digital infrastructure can itself become a form of labour regulation.

Arroyave makes visible the human labour that narratives of automation tend to hide.

These are not variants of the same theory. They are different lenses on the same sociotechnical assemblage.

Myriam Jimeno and what does not fit inside a variable

Myriam Jimeno, professor emerita at the National University of Colombia, who completed her doctorate at the University of Brasília and twice served as director of the Colombian Institute of Anthropology and History, has devoted much of her career to violence, emotions, recognition and social reconstruction.

In Cultura y violencia: hacia una ética social del reconocimiento, Jimeno connects historical structures of violence with the subjective experiences of those who live through them.

Her work is not about artificial intelligence.

That is precisely why it is useful.

Algorithmic systems need to translate the world before they can process it. A medical history becomes fields in a database. Poverty becomes indicators. A working life becomes variables. A person considered vulnerable becomes a category.

Such translation may be necessary.

But it is never neutral.

Quantifying an experience is not the same thing as understanding it.

Jimeno helps introduce into AI debates what remains outside datafication: emotion, memory, recognition, social relationships and histories that continue to exist even when they cannot easily become variables.

The question is therefore no longer only which data a model needs.

It is also what must disappear from an experience before that experience can become data.

That problem becomes particularly significant when Colombia develops models intended for health, public services or territory.

From what data erase to who decides which data count

There is a second question.

Someone has to determine which information constitutes legitimate knowledge.

Here Diana Obregón Torres enters the picture, a Colombian historian of science with a doctorate in Science and Technology Studies from Virginia Tech.

In Batallas contra la lepra: Estado, medicina y ciencia en Colombia, Obregón reconstructed how medicine, bacteriology, health statistics and public institutions simultaneously shaped a disease and the forms of intervention regarded as legitimate for those who lived with it.

Science appears here as a historical practice.

Its categories stabilise through institutions, controversies, professions, technologies and relations of power.

This perspective is especially productive as Colombia develops Bachué as a foundation model linked to health and well-being.

A medical AI system does not acquire authority simply by reaching a particular level of accuracy.

Many other things must have happened first.

Someone decided what constitutes a disease.

Someone determined which observations should be recorded.

A health system accumulated some data rather than others.

Specific institutions legitimised certain forms of knowledge.

Others were excluded.

Obregón’s question, applied to AI, is simple and deeply political:

how does an algorithm acquire the authority to decide?

Carlos Andrés Arroyave and Oscar Javier Maldonado have taken this problem into contemporary artificial intelligence. In Promesas algorítmicas, they examine sociotechnical imaginaries surrounding AI in healthcare, shifting attention away from a tool’s performance alone and towards the expectations, discourses and institutions involved in legitimising it.

Algorithmic authority also has to be built.

When code becomes regulation

The third lens takes us from knowledge to labour.

Derly Yohanna Sánchez Vargas, from the Faculty of International, Political and Urban Studies at Universidad del Rosario, has researched platform economies and the sociotechnical relations of digital labour in Colombia.

Together with Oscar Javier Maldonado Castañeda and Mabel Rocío Hernández Díaz, she studied the regulatory conditions of digital food-delivery platforms in Technolegal Expulsions: Platform Food Delivery Workers and Work Regulations in Colombia.

Their concept of “technolegal expulsions” describes a fundamental displacement.

Platforms do not exist outside regulation.

They construct technical, contractual and legal architectures that can shift labour relations into spaces in which workers’ rights become harder to recognise.

Software therefore participates in organising work.

An application allocates tasks.

It records time.

It produces ratings.

It can condition future access to work.

What matters for an anthropology of AI is that such decisions can appear to the person receiving them as the impersonal operation of a system.

“Nobody decided this.”

“The algorithm decided it.”

But that separation is fictitious.

Code does not remove regulation. It can become one of its forms.

What appears on a screen as a technical decision can simultaneously be an economic and political decision.

The border where AI acquires a body again

And then we reach Carlos Andrés Arroyave Bernal.

A sociologist with a doctorate in Human and Social Sciences from the National University of Colombia and an institutional affiliation with Universidad Externado de Colombia, he works on science, medicine and artificial intelligence.

His research with Oscar Javier Maldonado on DignifAI turns this whole discussion into a material case.

Dignity as a commodity? Labour dynamics and partial inclusion among data annotators in artificial intelligence in Colombia, published in Globalizations in 2025, studies DignifAI, an AI annotation service provider on the Colombia–Venezuela border whose initiative seeks to offer dignified employment opportunities to Venezuelan migrants.

The study is based on interviews with workers, managers and institutional actors involved in the programme.

Here the fiction of immaterial AI disappears.

For certain systems to classify images, interpret language or recognise patterns, people have to prepare and annotate data.

Machine learning exists because human labour came first.

Yet Arroyave and Maldonado avoid reducing this relationship to a simple opposition between exploitation and emancipation.

They examine partial inclusion and the way dignity participates in the transformation of labour power into a commodity. Their analysis addresses tensions between humanitarian inclusion and vulnerability while also bringing autonomy, control and consent into the discussion.

That matters because it helps us escape two equally inadequate stories.

One says that AI will eliminate human labour.

The other answers that every form of data annotation is necessarily the same form of exploitation.

The research gives us something more uncomfortable: concrete labour relations shaped simultaneously by need, opportunity, dependence, aspiration and agency.

Artificial intelligence acquires a body again.

And a territory.

And a border.

The question is no longer only which jobs AI will replace, but which forms of human labour it needs to render invisible in order to appear intelligent.

Two meanings of sovereignty

This is where the Colombian constellation connects directly with the present.

Since February 2025, Colombia has had a National Artificial Intelligence Policy, CONPES 4144. Its roadmap establishes more than one hundred actions through 2030 and planned investment of 479 billion Colombian pesos. Its six axes include governance and ethics, infrastructure and data, research and innovation, talent, risk prevention, and AI adoption in institutions, companies and territories.

At the end of December 2025, the government presented its Ethical Guide for the Implementation, Development and Use of Artificial Intelligence Systems in Colombian Public Entities. MinTIC presented it again in early January 2026 as part of the implementation of CONPES 4144.

On 22 April 2026, MinTIC also published specific security and privacy guidelines for AI systems in public entities.

And on 22 May 2026, it presented the governance framework for the Sovereign Foundation Models Bachué and Chiminigagua.

The sequence matters.

Colombia is no longer discussing only how to use models built elsewhere. It is beginning to ask how to build and govern them.

That is precisely why anthropology becomes more necessary, not less.

There are at least two ways of understanding technological sovereignty.

One is having infrastructure, computing capacity, data, talent and models of one’s own.

The other is having the collective capacity to decide which forms of knowledge a technology incorporates, which categories it uses, which labour relations sustain it and which worlds it allows us to imagine.

The first can be measured in data centres, parameters or computing capacity.

The second requires politics, anthropology and social participation.

Five lenses for looking at the same machine

The constellation we have followed reveals different layers of the same system.

Escobar makes us look at the ontology embedded in design.

Jimeno, at what disappears when experience becomes data.

Obregón, at the institutions that turn particular forms of knowledge into authority.

Sánchez Vargas, at the regulation that can hide inside digital infrastructure.

Arroyave, at the human labour and social relations that materially make artificial intelligence possible.

The sequence shifts the debate:

from model to world; from data to experience; from accuracy to authority; from software to governance; from automation to labour.

That displacement is one of anthropology’s most productive possible contributions to contemporary debates about AI.

What this Radar leaves open

Bachué and Chiminigagua now offer an especially interesting laboratory.

Their technical capabilities will not be the only thing worth following.

We will need to ask who participates in their governance, which datasets they use, how they represent Colombia’s territorial diversity, which forms of knowledge they incorporate, which forms of labour sustain their development, and how people and communities will be able to contest their decisions.

We will also need to ask what it actually means for a model to be Colombian.

That it was trained in Colombia?

That it uses Colombian data?

That its infrastructure is under national control?

Or that the communities whose worlds it seeks to represent can intervene in the categories through which the system turns them into data?

The work examined here does not provide a recipe.

It offers something perhaps more useful: a set of tools for refusing to confuse technological sovereignty with computational sovereignty.

From Cyberia to the present

In 1994, Escobar asked anthropology to take digital technologies seriously as spaces of cultural production.

Thirty-two years later, the scale has changed.

Algorithmic systems now participate in medicine, labour, public administration, communication and the production of knowledge. Colombia is also beginning to build models explicitly intended to respond to its health, its territory and its needs.

The question posed by Cyberia no longer belongs only to the anthropology of technology.

It has become a political question.

The artificial intelligence Colombia builds will not be sovereign simply because the model is its own. It will be sovereign to the extent that the people, territories and forms of knowledge it seeks to represent can also participate in defining the world that the model makes computable.

That leads to the hypothesis this Radar proposes to follow:

the decisive struggle over artificial intelligence will not be only over who owns the models, but over who has the right to define the worlds those models learn to recognise.

References

Note on sources

The 2025–2026 public-policy milestones have been checked against official DNP and Ministry of Information and Communications Technologies sources. The connections drawn between Jimeno or Obregón and artificial intelligence are AIthropology Lab’s analytical reading: the works cited do not themselves study AI systems. For DignifAI, the article’s own categories —partial inclusion, dignity, autonomy, control and consent— are retained without adding figures or working conditions that the study does not document.