When Knowing Too Much Can Reduce Autonomy

If an AI is always right about you, is it still helping you?

In the first four texts of this series, we built a path.

Data must return to those who produced it. It must return in the language of each Body-Territory. The representation produced by AI cannot take the place of the Being. And what we once were should not prevent the emergence of what we may still become.

Now another problem appears, perhaps a less obvious one:

what if Personalissima AI works too well?

Imagine an intelligence living alongside you for years.

It knows your schedules, movements, habits, texts, purchases, music, relationships, previous choices, and ways of reacting to different situations.

Over time, it begins to anticipate what you will probably do.

Before you search, it finds.

Before you ask, it suggests.

Before you choose, it organizes the options it considers most appropriate for you.

It seems like the perfect realization of personalization.

But there is a paradox:

the more AI becomes able to decide what is relevant to me, the fewer differences may reach me so that I can decide for myself.

A technology designed to expand autonomy could, if poorly designed, produce cognitive dependency.

Information can expand autonomy

This problem finds an especially important dialogue in the work of Maria Eunice Quilici Gonzalez.

Throughout her trajectory, Gonzalez has brought together philosophy of information, autonomy, self-organization, complex systems, and information technologies. In 2023, together with Mariana Broens, José Artur Quilici-Gonzalez, and Guiou Kobayashi, she directly examined the relationship between habits, rationality, autonomy, and Big Data.

The problem cannot be reduced to:

information versus autonomy.

Information can also expand autonomy.

If I discover that five paths exist where I previously perceived only two, my space of choice may increase.

If I receive information about consequences I did not know, I may reorganize a decision.

The problem begins when the same infrastructure that provides information also starts silently selecting which possibilities deserve to reach me.

The question changes:

at what point does helping organize information begin to organize the space within which we choose?

When helping begins to choose

Imagine that my Personalissima AI knows my food preferences.

I am traveling and ask:

“Where can I have dinner?”

It finds dozens of places, removes those that historically do not fit my preferences, and presents three options.

Excellent.

It saved time.

But there is a fourth restaurant serving a cuisine I have never tried.

The AI calculates that my probability of choosing it is low.

So it does not show it.

Nothing was prohibited.

No freedom was formally removed.

And yet, one possibility disappeared before entering my field of choice.

This example reveals a fundamental difference between:

helping someone choose

and

pre-organizing the environment so that certain choices become progressively less likely.

The better personalization becomes, the harder this difference may be to notice.

Habit makes life easier — and can also close possibilities

In the article Hábitos e racionalidade: um estudo filosófico-interdisciplinar sobre autonomia na era dos Big Data, Gonzalez and colleagues begin from a dilemma close to ours.

Information and communication technologies facilitate countless everyday activities, but they also participate in environments where opinions and decisions can be influenced by insufficient or distorted information, previously acquired habits, and emotional dispositions.

Habit is not the enemy of autonomy.

We need habits.

It would be impossible to consciously reconsider every gesture before performing it.

The problem appears when technological infrastructure learns our habits and begins to reinforce them before other possibilities can compete with them.

What I did yesterday helps select what I will see today.

What I see today participates in the conditions of what I will do tomorrow.

A loop emerges:

habit → data → prediction → selection → exposure → reinforcement of habit.

The AI appears to have discovered my preference.

But perhaps it is also participating in its repetition.

Predicting someone is not the same as preserving their autonomy

This is where an important distinction appears for the CEPID project.

A conventional system may be considered better when it increases predictive performance.

If it correctly predicted 70% of my choices and now predicts 95%, it apparently improved.

For a Personalissima AI oriented toward autonomy, however, this may not be enough.

Imagine prediction approaching 100% because the system has begun organizing my environment so efficiently that almost nothing different reaches me.

Prediction improved.

But my space of possibilities may have decreased.

We therefore need to ask not only:

“How much can AI predict about me?”

but also:

“How many possibilities remain available to me despite everything the AI already knows?”

Perhaps a good Personalissima AI must know a person deeply without turning that knowledge into an increasingly narrow corridor.

Self-organization is not being organized from outside

Self-organization offers another way into this problem.

In complex systems, patterns may emerge from interactions among components without an external controller defining every final state in advance.

We do not want to simply transfer this concept to people as if they were mathematical models. We want to extract a question from it.

Should Personalissima AI try to organize the Body-Territory?

“I know you. This is the best decision.”

Or should it provide informational conditions through which the Body-Territory continues participating in its own organization?

“Based on what I know about you, these options seem more compatible. There are also alternatives outside your usual pattern. Would you like to see them?”

In the first architecture, intelligence concentrates decision-making.

In the second, it returns differences to the one who decides.

The difference may look small in the interface.

But it is enormous for the concept of autonomy.

Perhaps AI needs to contradict its own personalization

Imagine a simple function:

“Show me something outside my pattern.”

Perhaps we would not even need to press a button.

The AI itself could recognize when it is narrowing the presented universe too much.

It could say:

“These options best match your history. There is another one that differs from the pattern I learned about you.”

This would preserve something fundamental:

the encounter with difference.

A choice does not depend only on the ability to select among alternatives.

It also depends on which alternatives manage to become perceptible.

A Personalissima AI oriented toward autonomy may need to know our pattern well enough to recognize when it should not reinforce it.

Information is not only quantity

In 2025, Valdirene Aparecida Pascoal and Maria Eunice Gonzalez published Notas para uma teoria integrativa da informação, seeking to build a conceptual map of the ontological, epistemological, and social dimensions of information.

This is especially relevant to our project.

A Personalissima AI should not assume that autonomy increases simply because the amount of information made available increases.

We may receive thousands of pieces of information and remain trapped within the same space of possibilities.

We may also receive a single relevant difference capable of reorganizing what we are able to perceive.

In the BrainLatam hypothesis, therefore, we need to ask not only:

“How much information did the AI return?”

but:

“What new differences did this information allow the Body-Territory to perceive?”

Body-Territory needs space and movement

Here we return to a principle that crosses our work:

the Body-Territory needs space and movement in order to signal and regulate itself.

As a BrainLatam hypothesis, we can extend this idea to the informational environment.

A Body-Territory surrounded only by content compatible with what its previous data predicts may receive an enormous quantity of information and, paradoxically, have very little room to change.

There is an abundance of content.

But little difference.

Personalissima AI could do precisely the opposite.

Know the pattern without turning it into destiny.

Preserve memory without closing possibilities.

Help without becoming the only way to decide.

Cognitive independence is not isolation

Preserving autonomy does not mean imagining isolated individuals.

We are formed through relationships.

We learn languages with others.

We receive recommendations.

We change our minds.

We are affected by institutions, technologies, and territories.

The question is not how to eliminate influence.

It is how to preserve our ability to participate in the reorganization produced by that influence.

I can receive a recommendation.

I can accept it.

I can reject it.

I can ask why it appeared.

I can request other possibilities.

I can change.

And I can surprise my own model.

By cognitive independence, we do not mean isolation of the Being, but the possibility of continuing to participate in the production of its next states.

BrainLatam Hypothesis: preserving the production of alternatives

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

a good Personalissima AI should not be evaluated only by its ability to correctly predict the Body-Territory, but also by its ability to preserve the production of alternatives by that Body-Territory.

Perhaps this requires new metrics.

Not only:

accuracy, prediction, engagement, retention.

But also:

diversity of preserved alternatives;

capacity for contestation;

reversibility of inferences;

voluntary exposure to difference;

capacity to surprise one’s own model.

Perhaps Personalissima AI should not aim to become indispensable.

One of its best measures of success may appear when a person says:

“I understand your recommendation. But I want to try something different.”

And the AI is able to respond:

“Then I need to learn again.”

It is precisely here that we are ready to move beyond the individual.

Because the same problem reappears at the collective level.

If an intelligence can know a person so well that it silently restricts their alternatives, collective intelligence can do the same with a population.

How can we learn from millions of Body-Territories without transforming regularities into obligations?

How can we belong without surrendering our cognitive independence?

It is at this point that the next concept must move to the center:

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

Commented References

GONZALEZ, Maria Eunice Quilici; BROENS, Mariana C.; QUILICI-GONZALEZ, José Artur; KOBAYASHI, Guiou. Hábitos e racionalidade: um estudo filosófico-interdisciplinar sobre autonomia na era dos Big Data. Trans/Form/Ação, vol. 46, special issue 1, pp. 367–386, 2023.
The main reference for Blog 5. The authors analyze the apparent conflict between relatively autonomous action and the influence that information technologies, habits, insufficient or distorted information, and emotional dispositions may exert on opinions and decisions.

PASCOAL, Valdirene Aparecida; GONZALEZ, Maria Eunice Quilici. Notas para uma teoria integrativa da informação. In: Estudos pluridisciplinares da informação: filosofia, tecnologia e semiótica. Oficina Universitária, 2025, pp. 19–44.
A recent work that seeks to map ontological, epistemological, and social perspectives on information. For our hypothesis, it helps avoid reducing “information” to the mere volume of data available to a person or system.

SOUZA, Edna Alves de; GONZALEZ, Maria Eunice Quilici. Big Data e Autonomia: continuidade ou revolução? In: Informação, conhecimento, ação autônoma e Big Data: continuidade ou revolução? Oficina Universitária/Cultura Acadêmica, 2019, pp. 25–46.
This reference predates the post-2021 scope of the series and is retained exceptionally as a conceptual antecedent of this line of research. The chapter directly examines Big Data, autonomous action, and ethical implications of the manipulation of large datasets.

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

BrainLatam Hypothesis: the more a Personalissima AI knows the Body-Territory, the greater its responsibility should be to preserve what its own model still cannot predict. Autonomy is not merely the ability to choose among the options AI considers appropriate; it is continuing to be able to produce a difference capable of forcing the AI itself to learn again.