The Data Is Not You

When a representation begins to take the place of the Being

In the first two texts of this series, we made two movements.

First:

the data must return to those who produced it.

Then:

it must return in a language through which the Body-Territory can participate in its meaning-making.

But now an even more difficult problem appears.

Imagine that all of this works.

A Personalissima AI accompanies someone for years. It learns their language, habits, movements, relationships, choices, and changes. It combines thousands of variables and begins to produce extraordinarily accurate representations of that person.

At some point, we might look at this model and say:

“This is you.”

That is precisely where we need to stop.

Because it is not.

The data is not you.

The map can improve without becoming the territory

A location recorded by a smartphone can tell us where we have been.

But by itself, it does not tell us what that place meant.

A purchase can record what we acquired.

But not necessarily why we bought it.

An EEG can record electrical potential differences associated with neural activity.

But it is not the thought.

An fNIRS measurement can track hemodynamic changes related to brain activity.

But it is not what someone felt.

The distinction seems obvious when presented this way.

It becomes less obvious when thousands or millions of these variables are combined by systems capable of identifying patterns that we ourselves might never perceive.

The better the model becomes, the greater the temptation to confuse accuracy of representation with the identity of what is being represented.

A photograph can gain resolution.

It remains a photograph.

A map can become extraordinarily detailed.

It still does not become the territory.

Between the object, the sign, and what happens next

This is where the semiotics of Charles Sanders Peirce, as developed in contemporary work by Vinicius Romanini, offers an important distinction for our proposal.

In his 2025 article Semiose, inteligência e inferência ativa, Romanini brings Peircean semiosis into dialogue with active inference and cognitive sciences.

In a semiotic relation, there is not simply “a thing” and its copy.

There is a triadic relationship involving what functions as a sign, what that sign refers to, and the interpretant — the effect produced by the representation within a semiotic process.

This changes how we think about data.

Data does not arrive alone carrying a complete meaning within itself.

It participates in relationships.

It can become a sign of something for a particular system or organism under particular conditions.

Therefore:

data ≠ meaning.

And meaning should not automatically be confused with truth either.

When AI infers

Suppose an AI repeatedly observes that a person leaves home at 7 a.m., follows a particular route, and remains for eight hours at the same address.

The data records movement.

The system may infer:

“workplace.”

It will probably be correct.

But that information is no longer simply the primary data.

It is an inference produced from it.

Now imagine another sequence.

The person reduces their movements, changes their schedule, sleeps less, and begins searching for a particular subject.

The system identifies a correlation and produces a classification.

The farther we move from primary data, the more important it becomes to recognize how we arrived at the final representation.

In Blog 1, we proposed distinguishing:

primary data → processing → inference → interpretation.

Now we can add:

none of these stages should disappear inside the word “data.”

An inference about me should not return as though it were simply a fact about me.

A habit is powerful precisely because it can change

Romanini proposes a particularly interesting connection between intelligent semiosis and active inference: organisms and systems construct habits and models that help reduce surprise and navigate dynamic environments.

For Personalissima AI, this creates both a possibility and a risk.

The possibility is evident.

An AI can learn regularities within a Body-Territory and progressively become more useful.

The risk appears when regularity becomes identity.

“You have done this many times” can become:

“This is who you are.”

“You demonstrated this preference” can become:

“This is your preference.”

“Your behavior is statistically similar to this group” can become:

“You belong to this profile.”

But a habit is interesting precisely because it has a history.

It can stabilize.

It can be reinforced.

And it can change.

A Personalissima AI needs to learn habits without transforming habits into essence.

A Cognitive Twin is still the other

Computation makes this distinction even more important.

In 2025, Wandemberg Gibaut and Ricardo Gudwin presented a method for building a Cognitive Twin capable of learning approximations of the interaction behavior of an external agent.

The work is important because it concretely demonstrates how distributed cognitive systems can model behavioral patterns from inputs and outputs.

But it also gives us a powerful image.

Suppose the model becomes extraordinarily accurate.

It anticipates almost every choice I make within that environment.

Does that make it me?

No.

It makes it an increasingly accurate model of certain observable aspects of my interactions.

This distinction does not diminish the technology.

It establishes both its power and its limit.

The better the twin becomes, the more important it is to remember that there are still two.

Information does not exhaust sentience

This limit finds another formulation in the work of Alfredo Pereira Jr.

In Triple-Aspect Monism, Pereira Jr. works with Matter, Information, and Sentience as necessary aspects for understanding consciousness, derived, in his ontological formulation, from a basis he calls Energy.

In recent work on Qualiomics, he also draws attention to the first-person perspective and the problem of singular qualitative experiences.

For our project, we do not need to claim that this ontology resolves the problem of consciousness.

We need to recognize the challenge it places before Personalissima AI:

an informational description of an experience is not automatically the felt experience itself.

We can record that someone remained in front of a landscape for ten minutes.

We can measure respiration, cardiac activity, EEG, and fNIRS.

We can ask what they felt.

We can correlate everything.

Even then, each measurement corresponds to a different cut.

In BrainLatam language:

feeling confirms consciousness, not truth.

The fact that I feel something confirms the reality of that felt experience for me. It does not automatically transform my interpretation of its cause into truth about the world.

Likewise, measuring something confirms the recording of that variable under particular conditions.

It does not automatically transform that measurement into the totality of the Being.

The danger of a representation that does not know it is a representation

We can now identify one of the greatest risks of Personalissima AI.

It is not simply that it might be wrong.

A system that recognizes uncertainty can correct an error.

The deeper problem appears when a representation stops presenting itself as a representation.

When it says:

“You are.”

instead of:

“Based on these data, during this period, using this model, there is evidence supporting this inference.”

This linguistic difference reveals an epistemological difference.

The first statement closes.

The second preserves space.

Space for context.

Space for contestation.

Space for new data.

Space for change.

An AI that shows where its statement came from

We can therefore imagine an important characteristic of Personalissima AI.

It should not merely return conclusions.

It should be able to show, in comprehensible language:

what was observed;

what was calculated;

what was inferred;

the degree of uncertainty;

and what remains interpretation.

If it says:

“You are becoming less social,”

the Body-Territory should be able to ask:

“Why?”

And perhaps receive:

“I observed a 38% reduction in movement and fewer recorded interactions over the last four weeks. I am inferring reduced sociability, but other explanations are also possible.”

Now there is an opening.

The person can respond:

“I am working from home.”

The representation changes.

The Body-Territory did not alter the primary data.

It added context to the semiotic process.

BrainLatam Hypothesis: the right to a distance between myself and my model

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

Every Personalissima AI should explicitly preserve a distance between the Body-Territory and the computational representations produced about it.

This distance is not a defect that we need to eliminate.

It may be precisely what protects the possibility of change.

The model says something about me.

I can recognize it.

I can contest it.

I can add context.

I can learn from it.

But I must remain able to say:

“This speaks about me without completely taking the place of who I am.”

Perhaps cognitive independence begins precisely within this distance.

And here the problem of the next essay emerges.

Because a representation does not need to be perfect to produce real consequences.

A profile can determine which content someone receives.

A classification can alter an opportunity.

A prediction can influence a future decision.

And, little by little, what began as a description may begin to participate in producing the behavior it claimed merely to observe.

The question is therefore no longer only:

“What does AI know about me?”

It becomes:

“What happens to me when I begin to live inside what AI has learned about me?”

That is why the next text will be:

The right to change the data about me.

Commented References — Latin America, post-2021

ROMANINI, Vinicius. Semiose, inteligência e inferência ativa. deSignis, no. 43, 2025.
The central reference for Blog 3. Romanini brings Peircean semiosis into dialogue with active inference, discussing representation, habits, surprise reduction, and meaning-making in organisms and artificial systems. For the BrainLatam hypothesis, it provides a foundation for distinguishing data, sign, inference, and interpretive process.

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.
Helps explain why the material existence of information does not, by itself, resolve questions of meaning, interpretation, and belief formation in environments mediated by AI.

CALDAS, Pedro; ROMANINI, Vinicius. Public Opinion and New Communication Technologies: the impacts of big data on public opinion studies from the pragmatism perspective. International Journal of Communication, 2025.
Expands the discussion from the individual to informational environments mediated by Big Data, preparing the later transition from Personalissima AI toward collective processes and Jiwasa.

PEREIRA JR., Alfredo. Qualiomics: The Metaphysics of Consciousness. Forum for Philosophical Studies, 2024.
Discusses the singularity of first-person qualitative experiences and their relationship to sentience. It is used here to support a careful distinction: the informational description of an experience should not automatically be identified with the felt experience itself.

GIBAUT, Wandemberg; GUDWIN, Ricardo R. Building a Cognitive Twin Using a Distributed Cognitive System and an Evolution Strategy. Cognitive Systems Research, 2025.
Presents an architecture capable of producing computational approximations of an agent's interactive behavior. For our discussion, it simultaneously demonstrates the power of modeling and the need to distinguish the computational model from the agent it models.

SEMIA — Research Group on Semiotics, Communication and Artificial Intelligence, ECA-USP. Academic production, 2022–2026.
The group coordinated by Romanini brings together recent research on semiotics, AI, disinformation, Big Data, public opinion, algorithmic agents, and technological sovereignty, constituting a particularly relevant Latin American research nucleus for the discussion proposed in this series.

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

BrainLatam Hypothesis: the better an AI becomes at representing a Body-Territory, the more important it becomes to preserve the distance between the representation and the Being. The model may learn with me, but it should not acquire the right to close off who I may still become.