The Data Must Return to Those Who Produced It

From a question posed to Roberto Lent to the emergence of Personalissima AI

Perhaps a Personalissima AI does not begin with artificial intelligence.

Perhaps it begins with a much simpler question: where is the data we continuously produce about ourselves going?

Some years ago, during a Science for Education congress in Brazil, we asked Brazilian neuroscientist Roberto Lent a question that stayed with us:

If each person's data is continuously being mined, how can we make each person's data return to them, in their own language, so that they can interact with and change the data produced about them?

At the time, the question seemed to be about data.

Today, we realize that it was also about language, autonomy, plasticity, meaning, and participation.

It was about who gets to interpret whom.

And perhaps it already contained the first question of what we now call Personalissima AI.

Data leaves us

Every day, we produce traces.

What we buy. Where we go. What we search for. What we watch. How long we remain in front of a particular piece of content. How we write. When we sleep. How much we walk.

Sensors add other layers: heart rate, movement, sleep, location, and physical activity.

In scientific and clinical environments, we can observe even more: EEG, fNIRS, ECG, SpO₂, respiration, and countless other variables.

But there is a distinction that will accompany this entire series:

the data produced from me is not me.

An EEG is not the brain.

A measurement of oxygenated hemoglobin is not the experience.

A location is not the lived territory.

A purchase is not necessarily a preference.

An isolated word does not contain everything a person intended to mean.

Even thousands of these variables processed together remain representations produced from some dimension of a Being.

This does not diminish the value of data.

On the contrary.

The better our ability to measure and infer becomes, the more important it is to remember the difference between what we can represent and what is being represented.

When data begins to speak for us

An important transformation is taking place.

Data is no longer merely a record of what happened.

Artificial intelligence systems can use it to generate inferences:

“You prefer this.”

“Your profile is that.”

“Your risk is this.”

“You will probably do that.”

Perhaps the inference is correct.

Perhaps it is wrong.

But there is a third possibility that is particularly important:

perhaps it was correct yesterday and is no longer correct today.

This is where the question posed to Roberto Lent encounters neuroscience once again.

An organism has a history, but it also has plasticity. Learning, experience, development, and environment continuously participate in its possibilities for reorganization.

If we can change, a machine that learns our past must also preserve the possibility that we may cease to be what it has learned about us.

Otherwise, the better its memory becomes, the greater the risk of transforming history into destiny.

We will explore this question more deeply later in this series. For now, it establishes a principle:

a model of a person must continue to admit that the person can change.

Having data about someone is not the same as producing knowledge with someone

In Latin America, this question acquires a particularly important dimension.

Recent work by ECLAC on Indigenous peoples has drawn attention to the relationship between statistical visibility and participation. It is not enough for populations to appear in databases. The participation of Indigenous peoples themselves in the production, interpretation, and use of information is relevant to their presence in decisions that affect them.

The difference is enormous.

We can produce data about someone.

Or we can create conditions in which those being represented continue to participate in the meaning produced from that data.

In 2023, UNESCO reached a related issue when discussing artificial intelligence centered on Indigenous peoples in Latin America and the Caribbean, emphasizing diversity, identity, rights, and participation in technological development.

This leads us to an uncomfortable question:

who should learn whose language?

Must the Body-Territory adapt to the language of systems?

Or can systems learn the different ways in which Body-Territories understand, organize, and give meaning to information?

“In each person's own language”

This small part of the original question has become an entire problem in itself.

Returning data does not necessarily mean returning it in a comprehensible form.

We can hand someone a spreadsheet containing thousands of variables and technically state:

“Here is your data.”

But has it truly returned?

Marcus Maia, through his work in psycholinguistics and Brazilian Indigenous languages, helps open a question that we will develop in the next essay: language cannot be treated merely as packaging into which already-formed information is placed.

“In each person's own language” does not simply mean choosing Portuguese, Spanish, English, or an Indigenous language from a menu.

Two people who both speak Portuguese may require completely different forms of presentation in order to meaningfully interact with the same dataset.

The problem is not merely translation.

It is meaning-making.

Data must return in a form that allows the person who produced it to re-enter the interpretive process.

Body-Territory

This is why we prefer not to begin this proposal with the word “user.”

User is a useful category for systems.

Consumer is useful for markets.

Patient is useful for healthcare systems.

Voter is useful for political processes.

A tax identification number is useful for administration.

But none of these categories, in isolation, contains the Being.

We propose thinking through Body-Territory as a unit in which body, environment, history, culture, relationships, and possibilities for transformation are not artificially separated when we attempt to understand a person.

This does not mean rejecting categories or measurements.

It means recognizing that they are cuts through reality.

The screen is a transduction of the cut. It has light, but it does not illuminate everything.

The same is true of data.

From personalized AI to Personalissima AI

Digital technologies are already personalized.

Platforms select content, systems recommend products, and algorithms learn behaviors.

We can represent this movement in simplified form:

Body → data → platform → model → decision/recommendation → Body.

The Body provides data and later receives something produced from it.

The BrainLatam hypothesis adds another movement:

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

The difference may appear small.

But it changes who remains inside the circuit.

Personalissima AI would not simply be an AI that knows a great deal about someone.

It should allow that person to continuously participate in what is being produced about them.

Imagine an AI saying:

“You prefer staying at home.”

And the person being able to respond:

“I used to. My life has changed.”

Or:

“Your data shows a particular behavior.”

And the person being able to ask:

“What period are we talking about?”

Or:

“Your profile indicates a particular category.”

And the person asking:

“Which data produced that inference?”

The AI no longer merely presents a conclusion.

The process opens again.

The right to answer the algorithm

This does not mean allowing someone simply to erase facts because they dislike them.

A recorded event does not cease to have occurred because we disagree with it.

We need to distinguish:

primary data → processing → inference → interpretation.

It is especially in the later stages that different possibilities can emerge.

Vinicius Romanini and the semiotic tradition running through his work become important to this series precisely here: a representation must not be confused with what it represents.

Ricardo Gudwin, recommended to this project by researcher Alfredo Pereira Jr., adds another dimension through his work on cognitive architectures and Cognitive Twins: computational systems can learn increasingly sophisticated approximations of agents' behavior.

Yet the better the mirror becomes, the more important it is to remember:

the mirror is not what it reflects.

An AI can learn a great deal about me.

But it must preserve room to be wrong about me.

And it must preserve something even more important:

my room to change.

BrainLatam hypothesis: restitution

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

Data and inferences produced from a Body-Territory should be able to return to that Body-Territory in forms that it can understand, interpret, contest, and transform, without the representation produced by the system automatically acquiring the status of truth about the Being.

We call this movement restitution.

Restitution is not simply making a file available.

It is restoring the capacity to participate.

Perhaps this is precisely where Personalissima AI begins.

Not with a machine capable of saying:

“I know who you are.”

But with a relationship in which it remains possible to answer:

“This is what your data can say about me now. I also participate in that meaning — and I can still change.”

So far, we have spoken about one Body-Territory.

One.

But a State is constituted by millions of them.

Each producing data, creating meaning, changing, and carrying possibilities that no database can completely anticipate.

What would happen if these differences could become intelligible to one another without having to disappear?

This is where a word that will accompany this series begins to appear:

Jiwasa — “We / Us”: to feel that we belong without having to reduce the Being.

We do not yet need to speak about collective intelligence.

First, we need to return the data.

Because perhaps the collective future does not begin when artificial intelligence finally manages to know everyone.

Perhaps it begins when each person recovers the possibility of participating in what is being known about them.

The data must return to those who produced it.

Commented References — Latin America, post-2021

ACOSTA, Laura Débora; RIBOTTA, Bruno. Visibilidad estadística y mecanismos participativos de los pueblos indígenas en América Latina: avances y desafíos. ECLAC/CEPAL, 2022.
Helps distinguish the mere production of statistics about populations from the participation of those populations in constructing, interpreting, and using the information that represents them.

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 AI through Latin American cultural diversity, rights, identity, and the participation of Indigenous peoples in technological development.

MAIA, Marcus. Focalizando e Topicalizando na língua Karajá: brincando epilinguisticamente. UFRJ, 2022.
Contributes to the question that will be explored in the next blog: what does it really mean to return information “in each person's own language,” particularly when different linguistic and cultural structures are considered?

ROMANINI, Vinicius. Semiose, inteligência e inferência ativa. deSignis, 2025.
Provides a bridge between semiosis, cognition, and intelligence, helping distinguish representation, interpretation, and that which is being represented.

RIPOLL, Leonardo; ROMANINI, Vinicius. Considerações sobre a dimensão estética semiótica da desinformação no âmbito da pós-verdade e no desenvolvimento da inteligência artificial. Tríades, 2025.
Relevant to understanding why information and meaning are not equivalent and how interpretive processes participate in the relationship between humans, signs, and artificial systems.

GIBAUT, Wandemberg; GUDWIN, Ricardo R. Building a Cognitive Twin Using a Distributed Cognitive System and an Evolution Strategy. 2025.
The Cognitive Twin proposal demonstrates the possibility of constructing computational models that learn behavioral approximations of agents. For the BrainLatam hypothesis, it also provides a fundamental counterpoint: an increasingly accurate model of someone is still not that someone.

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

BrainLatam Hypothesis: data does not truly return to the Body-Territory merely when it is made available. It returns when the person who produced it can once again participate in the process through which that data acquires meaning.