In Each Person’s Own Language

When returning the data does not mean that it has truly returned

In the first text of this series, we proposed:

the data must return to those who produced it.

But how do we know whether it has truly returned?

Imagine a person receiving years of information produced about them: health, education, consumption, movements, digital behavior, relationships with institutions and, in the future, perhaps data organized by their own Personalissima AI.

Tables, probabilities, graphs, and classifications appear on the screen.

Technically, the data has returned.

But what if that person cannot make meaning from it?

Perhaps restitution is not simply about returning information. We need to ask:

in what language must the data return in order to encounter again the Body-Territory that produced it?

It was precisely this expression — “in each person’s own language” — that appeared in the question posed years ago to Roberto Lent and now becomes the second problem of W41/2026.

Speaking the same language does not mean making meaning in the same way

Let us begin by eliminating a misunderstanding.

“In each person’s own language” does not simply mean translation.

It is not enough for an AI to identify whether someone speaks Portuguese, Spanish, Aymara, or an Indigenous language and change the interface accordingly.

Two people who speak Portuguese may understand the same information in profoundly different ways.

One may prefer numbers. Another, images. One immediately understands a probability. Another needs to relate it to an everyday situation. One is familiar with a particular scientific vocabulary. Another possesses extremely sophisticated knowledge about their territory but organizes differences through categories other than those used by academic science.

Language, therefore, is not merely a container into which we place information.

Language participates in the way differences acquire meaning.

This is where the work of Brazilian psycholinguist Marcus Maia becomes particularly important to our question.

When psycholinguistics encounters other territories

Marcus Maia and his collaborators have developed research on language processing that includes Brazilian Indigenous languages.

This encounter also exposes a problem within science itself.

A substantial part of cognitive psychology and experimental psycholinguistics has been constructed from populations frequently described by the acronym WEIRD — Western, Educated, Industrialized, Rich, and Democratic.

When findings obtained from this particular population are automatically transformed into universal models of human cognition, we may fail to perceive other forms of linguistic and cognitive organization.

In recent work with the Karajá, Marcus Maia, Daniela Cid de Garcia, and Juliana Novo Gomes discuss precisely such Non-WEIRD psycholinguistic initiatives, in which Indigenous participants do not appear merely as providers of experimental responses. There is also a metacognitive dimension, in which speakers can reflect on their own language and on the phenomena being investigated.

This creates an important connection with our hypothesis.

Perhaps the problem is not merely knowing the other better.

Perhaps the person being known should also participate in the categories used to understand them.

What if our categories arrive first?

Imagine an AI trained on enormous volumes of data.

It learns categories, associations, and statistically efficient ways of organizing differences.

Then it encounters a Body-Territory.

There is a tendency:

to fit that Body-Territory into the categories the machine already possesses.

But what if differences that are fundamental to that person do not even exist within the system’s classificatory architecture?

An AI may speak a language perfectly while still imposing categories produced in another context.

This does not mean that every Body-Territory inhabits an incommunicable universe. If that were the case, shared knowledge itself would be impossible.

The challenge lies precisely between two extremes:

not assuming that everyone makes meaning in the same way, while also not assuming that no one can share meaning.

Within this interval, an important characteristic of Personalissima AI begins to emerge.

Perhaps the machine needs to learn how to ask

We usually think of personalization as the ability to provide better answers.

What if it also meant learning to ask better questions?

A system reports:

“Your risk has increased by 17%.”

But it could ask:

“Would you prefer to understand this as a probability, a historical comparison, or an everyday situation?”

It detects:

“Your behavior has changed.”

And asks:

“Does this change correspond to something that happened in your life?”

Or it infers:

“You usually prefer X.”

But keeps another possibility open:

“Does this description still make sense to you?”

Language is no longer merely the channel through which AI delivers a conclusion.

It becomes part of the mechanism through which that conclusion remains open to meaning-making.

We are not proposing that facts become negotiable.

We are allowing interaction with interpretations.

When the person who was observed can look at the data

There is a particularly interesting image in Maia’s experiences with Karajá teachers.

In activities involving linguistic phenomena and eye tracking, the data produced could subsequently participate in reflections on reading and on the language itself.

Consider the movement.

The equipment observes.

It produces data.

The researcher analyzes it.

But the data can return.

And the person who was being observed can look at a representation produced from their own activity and once again participate in the process.

We are not claiming that Maia or the Karajá participants proposed Personalissima AI. That extension is BrainLatam’s.

What this experience helps us see is a methodological principle:

the participant does not need to remain merely the place from which data is extracted; they can participate again in interpreting what was produced from them.

This brings us closer to what we called restitution in Blog 1.

Latin America does not fit into a single interface

The question becomes even larger when we consider our territory.

The International Decade of Indigenous Languages, promoted by the United Nations from 2022 to 2032, draws attention to enormous linguistic diversity and to the risks faced by many of these languages.

In Latin America, recent initiatives have also addressed digital presence, language technologies, and the participation of communities themselves.

This suggests that it may be insufficient to follow the model:

first we build the technology; then we translate the interface.

We need to ask beforehand:

which differences is this technology capable of perceiving?

A language is not merely a different collection of words used to name the same things.

It participates in histories, relationships, and culturally situated ways of organizing experience.

Nor should this be romanticized as something immutable.

Body-Territories change.

Languages change.

Cultures change.

Relationships with technologies also create new ways of making meaning.

For this reason, Personalissima AI should not attempt to discover someone’s “definitive language.”

It would need to continue learning with that person.

The data needs to encounter the person who produced it

In Blog 1, we proposed:

Body-Territory → data → processing → Personalissima AI → restitution → interpretation → contestation/transformation → new data.

Now we realize that something was missing between restitution and interpretation:

meaning-making.

We can expand the sequence:

Body-Territory → data → processing → Personalissima AI → restitution → language/meaning-making → interpretation by the Body-Territory → contestation/transformation → new data.

If there is no bridge of meaning, data can return physically without returning cognitively.

It reached the screen.

It did not necessarily reach the person.

BrainLatam Hypothesis: language is also part of restitution

We therefore arrive at the second hypothesis of W41/2026:

A Personalissima AI should not merely learn about a Body-Territory. It should continuously learn how to return its representations in ways that allow that Body-Territory to participate in producing meaning from them.

This changes the question.

Instead of asking only:

“How can we explain this more simply?”

we can ask:

“Which differences are relevant for this Body-Territory to interact with this representation?”

Perhaps words are needed.

Perhaps images.

Numbers.

Sounds.

Spatial relationships.

Examples connected to the territory.

Or combinations we do not yet know.

Personalissima AI should not decide in advance which language is correct.

It would need to discover this with the Body-Territory.

And perhaps it is precisely this with that begins to prepare our passage from the individual to the collective.

Because language never happens entirely alone.

We learn signs with others. We disagree. We reorganize meanings. We create new habits.

If different Body-Territories make meaning in different ways, how can they build something in common?

How can they share without requiring everyone to interpret in the same way?

How can one belong without giving someone else the power to define meaning alone?

This is where we slowly approach Jiwasa — “We / Us”: to feel that we belong without having to reduce the Being.

We have not yet reached the collective Jiwasa.

Before that, we need to confront another fundamental distinction.

An AI can learn our language, our habits, and our signs. It can construct increasingly accurate representations of us.

But what it can say about me is still not me.

That is why our next essay begins with a simple statement:

The data is not you.

Commented References — Latin America, post-2021

MAIA, Marcus; CID DE GARCIA, Daniela; GOMES, Juliana Novo. Non-WEIRD (psycho)linguistic endeavors: theoretical and metacognitive research with the Karajá of Central Brazil. Revista Linguíʃtica, 2023/2024.
The central reference for this essay. It discusses the limitations of WEIRD predominance in psycholinguistic research and research experiences with the Karajá that include a metacognitive dimension and greater participation by the speakers themselves.

MAIA, Marcus. Focalizando e Topicalizando na língua Karajá: brincando epilinguisticamente. UFRJ, 2022.
Presents experiences with Karajá teachers involving linguistic reflection and experimental data. For the BrainLatam hypothesis, it helps us think about the transition from the participant as a source of data to the participant as someone involved in its interpretation.

MAIA, Marcus; GOMES, Juliana Novo. Cultural attitudes and linguistic processes in Karajá. Linguistic Approaches to Bilingualism, 2023.
Contributes to understanding how linguistic processes, bilingualism, and cultural attitudes can participate in the same linguistic ecology, avoiding the treatment of language and context as independent dimensions.

LLANES-ORTIZ, Genner. Digital Initiatives for Indigenous Languages: Creating Digital Futures for Indigenous Languages. UNESCO, 2023.
Brings together strategies and experiences related to Indigenous languages in digital environments and highlights the participation of Indigenous peoples themselves in constructing technological futures for their languages.

UNESCO. Empowering Indigenous Languages in the Digital Age: A Toolkit for Action. 2023.
Helps move the discussion beyond the simple digitization of languages toward questions of participation, digital presence, technological creation, and communities’ relationships with their own languages.

GONZÁLEZ ZEPEDA, Luz Elena; MARTÍNEZ PINTO, Cristina Elena. Inteligencia artificial centrada en los pueblos indígenas: perspectivas desde América Latina y el Caribe. UNESCO, 2023.
Important for thinking about a Latin American AI that does not treat cultural diversity as a later interface-adaptation problem, but as a relevant dimension from the very conception of the technology.

W41/2026 — CEPID: From Personalissima AI to the Collective Future

BrainLatam Hypothesis: returning data does not simply mean making it reach the person again. Data returns when it can encounter the Body-Territory in a language through which that Body-Territory can participate in the production of its meaning.