Before a model classifies the world, other institutions have already classified it.

The school selects content, languages and legitimate ways of learning. The archive decides what deserves to be preserved and under which description it can later be recovered. A sensor network determines which dimensions of a territory become variables. By the time artificial intelligence enters the scene, much of that prior work has already been done.

This radar connects three signals that have already materialised and one anticipatory signal from Chile and Uruguay. They do not form a homogeneous technological trend. They share a deeper concern: which institutions allow certain experiences to become knowledge, evidence and data, while leaving others outside the record.

The selection criterion is not the appearance of a new technology. These signals make visible the educational, documentary and material infrastructures that produce the data on which algorithms subsequently operate.

An anticipatory signal from schooling

On 20 August 2026, the Faculty of Philosophy and Humanities at the University of Chile will host the international seminar “Educación, Estado y pueblos indígenas en la Baja Amazonía y los Andes”. It is organised within EDGES —Entangling Indigenous Knowledges in Universities—, a network funded by Horizon Europe through the Marie Skłodowska-Curie Actions.

The programme does not treat the school as a neutral container through which knowledge simply circulates. It places schooling within historical processes of state formation, social discipline, cultural homogenisation and territorial dispute.

The papers will examine the Chileanisation of the frontier during the nineteenth century; the relations between missions, schools and colonisation in La Araucanía; Mapuche teacher training; state narratives during the dictatorship; Indigenous women’s access to legal culture; and the incorporation of intercultural approaches into medical education.

Pirjo Kristiina Virtanen, an academic at the University of Helsinki, will address learning through territory, “multibeings” and deep time along the Purus River. The closing lecture will be delivered by Elisa Loncón Antileo, an academic at the University of Santiago de Chile, on education, coloniality and cognitive justice from Mapuche philosophy.

The anthropological significance of the meeting extends beyond the history of education. It allows us to see how a state does not merely administer populations and territories. It also tries to stabilise a shared reality through official languages, curricula, calendars, maps, examinations and authorised forms of knowledge.

Schooling, however, never operates in only one direction. Communities negotiate, translate, resist and reappropriate its devices. The school can function as a mechanism of deterritorialisation and, at the same time, as a space from which memories, political identities and demands for recognition are reconstructed.

This discussion bears directly on educational artificial intelligence. Systems that process curricula, assessments or educational trajectories do not encounter neutral categories. They inherit historical decisions about which languages, capacities and temporalities an institution is able to read.

Educating without conquering the world

A second signal comes from an interview with Tim Ingold published by Revista Santiago on 21 July 2026.

Ingold contrasts techno-optimism and eco-pessimism with a form of hope that does not regard the future as a technical problem awaiting a solution. His assertion that the future is a life to be lived condenses a wider critique of promises of total automation and of the idea that accumulating information allows uncertainty to be mastered.

In education, following the philosopher of education Gert Biesta, he takes up the distinction between strong and weak education. “Weak” does not mean inadequate. It describes a practice capable of exposing itself to the unexpected, paying attention and responding to what was not anticipated by the programme.

The interview introduces another decisive distinction. Knowledge can accumulate information in order to act upon the world. Wisdom requires learning how to correspond with it.

This matters for educational artificial intelligence. Many platforms turn learning into a sequence of objectives, metrics, recommendations and predictions. They can provide valuable support, but they can also reinforce a strong conception of education in which every meaningful outcome must be specified in advance and made measurable.

A system may recommend a reading on the basis of a person’s previous performance. It is far harder for it to determine when that person needs to encounter an uncomfortable idea, an unforeseen experience or a perspective that does not fit their profile.

The point is not to deny the usefulness of automation. It is to recognise that every AI-assisted educational system contains a conception of what learning means and of which kind of subject can be recognised as someone who knows.

That conception is not confined to the curriculum or the software. It is also inscribed in the buildings, sensors and energy systems that make school life possible.

When the school is also a sensor

The third signal shifts attention from the curriculum to physical infrastructure.

On 24 July 2026, the University of Chile announced the creation of the first High-Andean Living Lab for public infrastructure in the Arica and Parinacota Region.

The project will be carried out by the Andean Geothermal Centre of Excellence, part of the Faculty of Physical and Mathematical Sciences, and is financed through the 2026 Regional Fund for Productivity and Development of the Regional Government of Arica and Parinacota. The University of Chile’s official source identifies Linda Daniele, an academic in the Department of Geology, as director of the initiative and Bárbara Bravo as its general coordinator.

The Living Lab will intervene in the boarding facilities of the Liceo Granaderos de Putre and the Visviri School, which have geothermal heating systems previously developed by the University of Chile. Improvements to the buildings’ thermal envelopes will be accompanied by permanent monitoring infrastructure.

The sensor network will record the buildings’ thermal behaviour, the performance of the energy systems and environmental conditions on the Altiplano. The schools will retain their educational and care functions, but they will also begin to operate as platforms for the continuous production of scientific evidence.

The initiative responds to an immediate material need. Improving thermal conditions for pupils, teachers and other members of educational communities at high altitude is not an abstract matter. The project also aims to generate models capable of informing future public investment in infrastructure adapted to extreme conditions.

Precisely because of its value, the Living Lab raises questions that should be incorporated from the design stage. Who defines the relevant variables? How do educational communities participate in interpreting the data? Which forms of territorial knowledge cannot be accommodated by the sensor network? Under what conditions may the information produced be reused?

A monitored territory is not merely observed with greater precision. It also begins to be governed through new categories of evidence. Before a predictive model exists, the infrastructure will already have decided what counts as cold, comfort, efficiency and data.

The legible archive is not the whole archive

The fourth signal comes from Uruguay.

On 16 July 2026, the University of the Republic hosted the round table “Humanidades digitales: experiencias y oportunidades” as part of the International Workshop on Reverse Typography. The National Library of Uruguay, which helped organise the event and hosted its working sessions, later published a specific account of the meeting.

The event brought together specialists from different disciplines to discuss the use of machine learning in the study and preservation of documentary heritage. Reverse typography works from large collections of images of printed documents to extract information about content, authorship and provenance.

The materials discussed make the problem concrete. Jean-Michel Morel presented the technique as applied to historical Chinese books printed from woodblocks, where the aim is not merely to recognise characters but to reconstruct their forms without allowing the system to invent plausible solutions. Another line of work, RETIPIA, applies the automatic recognition of early typefaces to individual theatrical prints from the Spanish Golden Age, a corpus marked by editions that are difficult to identify and, in some cases, lack reliable bibliographical information.

The National Library also demonstrated the material diversity that any digitisation policy must confront. Its holdings include manuscripts, writers’ personal archives, early printed books, a Chinese bibliographical collection and historical recordings on magnetic tape. The latter are the subject of a project with the Faculty of Engineering to develop machine-learning tools for their restoration.

Digitisation is often presented as an act of rescue. A fragile document can be copied, distributed and consulted without handling the original. Yet turning an archive into a computable corpus requires materials to be selected and described, metadata to be established, characters to be recognised, errors to be corrected and decisions to be made about which relationships will become searchable.

Each of these operations expands certain possibilities for reading while narrowing others. Documents that are not digitised, receive inadequate descriptions or produce too many errors can remain technically available and at the same time become practically invisible to automated search.

Artificial intelligence does not gain access to the past in its entirety. It gains access to those portions of the past that have been preserved, catalogued, digitised and converted into processable formats.

A model’s forgetting does not therefore necessarily begin during training. It may start much earlier, in a preservation policy, an archival description or a typeface that the system fails to recognise.

Before the model

The four signals allow the same sequence to be reconstructed.

The school produces categories, competencies and legitimate narratives. The archive converts part of memory into a searchable corpus. Sensors translate territorial conditions into data series. Models then learn from those materials. In this sense, classifications are not merely administrative arrangements: as Geoffrey Bowker and Susan Leigh Star showed in Sorting Things Out, they distribute visibility, memory and the capacity to act, even though their decisions may eventually come to appear natural.

This perspective requires us to broaden algorithmic governance. Auditing a system once it has been trained is necessary, but insufficient. We must also examine the institutions that produced its data, the criteria that organised its categories and the absences that remained outside the record.

Data provenance should not be reduced to a technical datasheet. It needs a social history.

That history includes curricular decisions, language policies, archival practices, colonial relations, funding priorities, measurement standards and forms of community participation. It also includes everything that never became a document or a variable.

As Boaventura de Sousa Santos argues, cognitive justice is not merely a matter of adding Indigenous or community content to a database that has already been built. It requires discussion about who defines the categories, who controls the infrastructures and who is able to contest the interpretations produced from them.

An infrastructure for attention

The EDGES seminar reminds us that the school has been a field of colonial dispute. Ingold and Biesta offer a compass for resisting the reduction of education to metrics. The Living Lab shows that buildings also produce data. The Uruguayan meeting makes visible the documentary selection that precedes every digital corpus.

Together, these signals outline a programme. Before auditing an algorithm, we need to study the institutions that feed it. Before designing an AI system, we need to decide which infrastructures will allow knowledge to be not only what a machine can measure, but also what deserves attention.

Every model has an institutional prehistory. Understanding it is a condition for imagining technologies capable of relating to the world without reducing it to what they already know how to classify.

What to keep watching

  1. What mechanisms for participation and data sovereignty will the High-Andean Living Lab incorporate for the educational communities of Putre and Visviri?

  2. Could the intercultural approaches discussed by EDGES offer criteria for recognising forms of learning based on attention and correspondence, in Ingold and Biesta’s sense, without immediately turning them into new metrics?

  3. Which collections, languages and media remain poorly legible to current technologies for digitisation and documentary recognition?

Sources

  1. Universidad de Chile. Seminario Internacional Proyecto EDGES: “Educación, Estado y pueblos indígenas en la Baja Amazonía y los Andes”. Event scheduled for 20 August 2026.

  2. Tapia, Patricio. “Tim Ingold: ‘El futuro es una vida por vivir, no un problema por resolver’”. Revista Santiago, 21 July 2026.

  3. Universidad de Chile / CEGA. “U. de Chile desarrollará el primer Living Lab Altoandino de infraestructura pública en Arica y Parinacota”. 24 July 2026.

  4. European Commission, CORDIS. Entangling Indigenous Knowledges in Universities — EDGES. Horizon Europe / Marie Skłodowska-Curie project, 2024–2027.

  5. Biblioteca Nacional del Uruguay. “Investigadores de referencia en humanidades digitales se reunirán en la Biblioteca Nacional”. 8 July 2026.

  6. Biblioteca Nacional del Uruguay. “La Biblioteca Nacional fue sede del International Workshop on Reverse Typography”. 23 July 2026.

  7. AISO 2026. Programme for the panel “Impresos sueltos teatrales: nuevas perspectivas ante un antiguo problema bibliográfico”. Presentation of the RETIPIA project.

  8. Universidad Internacional de La Rioja. “Impresos sueltos del teatro antiguo español: base de datos integrada del teatro clásico español”. Description of the bibliographical problem posed by editions with unreliable or false information.

  • Ingold, Tim. Llevando la vida: antropología y educación. Translated by Ana Stevenson. Ediciones Universidad Alberto Hurtado, 2022.
  • Bowker, Geoffrey C. and Susan Leigh Star. Sorting Things Out: Classification and Its Consequences. MIT Press, 1999.
  • De la Cadena, Marisol. Earth Beings: Ecologies of Practice across Andean Worlds. Duke University Press, 2015.
  • Escobar, Arturo. Designs for the Pluriverse. Duke University Press, 2018.
  • Santos, Boaventura de Sousa. Construyendo las epistemologías del Sur. CLACSO, 2018.