Artificial intelligence does not begin when a model produces an answer. It begins earlier, when an institution decides which conflict merits a case file, which research merits recognition and which experience becomes data at all.

That decision rarely presents itself as technological. It is distributed across forms, fees, assessment criteria, publishing infrastructures and administrative records. This is why digitalisation can open a door while, at the same time, reinforcing a pre-existing boundary.

This radar brings together three scenes. An unpaid debt in England, a debate on open science in Chile and a representation gap in Chilean health records. They do not describe the same problem, but they illuminate a shared question: who has the material and institutional conditions to claim, to count and to be taken into account.

When a debt becomes claimable

In May 2026, the self-employed consultant Tamires Camal Taquidir obtained a favourable ruling in England over an unpaid debt of £7,000. To prepare her case, she turned to Garfield.Law, a regulated firm that uses automation to organise evidence, prepare documents and support small-claims litigation.

The story matters not because a machine replaced justice or legal practice. Oral advocacy remained the work of a human barrister. What matters is that part of the documentary preparation became more affordable. The roughly £400 reported in the press relates to Garfield’s service for the legal letter and the start of proceedings, not to the total cost of litigation. Court fees and other possible costs still apply.

Even with that limitation, the case raises a material question. When preparing a claim costs more than the amount likely to be recovered, many debts cease to be enforceable rights and become losses people simply absorb. A tool can reduce that barrier. But it also selects which disputes fit its forms, which evidence it can process and which people remain outside because their case is too complex, urgent or difficult to document.

What knowledge counts

At the 12th Congress of University and Specialist Libraries at the University of Chile, the Argentine sociologist María Fernanda Beigel called for a shift in the centre of scientific evaluation. Against a system that privileges productivity, citations and impact factor, she argued for ways of valuing the social impact of knowledge, its linguistic diversity and its relationship to public needs.

This is not an opposition between numbers and rigour. Metrics can help us observe part of scientific activity. The problem arises when a figure designed to measure the circulation of a journal becomes a substitute for the quality, relevance or value of research. The San Francisco Declaration on Research Assessment has challenged this automatic use of publication metrics for years.

From Latin America, the debate also concerns infrastructure. Who controls databases, who can afford to publish, which languages are recognised as vehicles of knowledge and which problems achieve international visibility. The proposal for democratic and socially relevant open science is not merely about widening access to articles. It seeks to contest the conditions through which particular kinds of knowledge come to count.

An AI system that classifies, recommends or evaluates research does not enter an empty field. It learns from catalogues, repositories and prestige systems that have already been ordered. The question is not only what it answers, but which hierarchies it inherits as its point of departure.

When the record falls short

Chile’s 2024 Census found that 11.5% of the population self-identifies as belonging to an Indigenous or native people. Yet the academic Sandra Flores Alvarado has warned that, in many health records, this presence may fall to 2–3%. This is not a uniform figure for the entire health system, but a warning sign of a representation gap that must be examined carefully.

A blank field does not have a single meaning. It may signal a barrier to access, a question that was never asked, a poorly designed form, a database without interoperability or the reasonable choice not to disclose an identity in an institutional setting marked by experiences of discrimination. As Flores explains at the University of Chile, treating all these situations as the same turns a complex social history into a straightforward technical problem of incomplete data.

The issue becomes especially significant in digital health. Systems supporting management and clinical decision-making draw on the records available. When a group appears only in fragments, the model does not independently discover what is missing. It may treat that absence as adequate representation and reproduce a partial view of needs, risks and patterns of health-service use.

The answer is not to extract more information without conditions. It requires discussion with the Indigenous peoples and communities concerned about which data are collected, who holds them, what they are used for and under what conditions of consent, transparency and collective control.

Exclusion before the model

These three scenes do not show that all digitalisation excludes. They show something more precise. A tool can lower the cost of making a claim, broaden the circulation of research or improve the organisation of a service. But it can also make operational the categories, absences and hierarchies that were already embedded in institutions.

This is why inclusion is not achieved simply by adding more people to a database or by offering a cheaper interface. It requires asking who defines the categories, who sustains the infrastructure, who can challenge a decision and which forms of knowledge or experience remain outside the record.

AI does not begin in the model. It begins at the earlier moment when it is decided who can claim, who can count and who is left out.

Sources