The Logos Crystallized into AI - Why Do We Still Need to Learn How to Think?
The Logos Crystallized into AI - Why Do We Still Need to Learn How to Think?
When the machine answers before the question has finished being born
Let us imagine an increasingly common scene.
A student receives a question: “What are the causes of inequality in Latin America?” Before organizing what she already knows, she opens an artificial intelligence tool. Within seconds, a clear answer appears, divided into sections, with sociological concepts, historical events, and a convincing conclusion.
The answer seems complete.
But can the student distinguish which causal relationships were demonstrated and which were merely suggested? Can she identify the perspectives that are absent? Will she recognize a nonexistent reference? Can she disagree with a well-written sentence?
Perhaps the main problem is not that the machine answered.
Perhaps the answer arrived before the student could perceive her own question.
When the machine already produces answers, which capacities must remain embodied in each Body-Territory so that it does not lose its freedom of judgment?
“The Logos crystallized into AI”
Within the NeuroEducation of Weichö, the expression “The Logos crystallized into AI” is a BrainLatam philosophical metaphor. It is not a scientific claim that human reason has been entirely transferred to machines.
By crystallization, we mean that part of humanity’s linguistic and logical-formal production has been transformed into computational structures capable of:
organizing words;
comparing patterns;
summarizing documents;
constructing arguments;
performing calculations;
translating;
classifying;
generating procedural sequences;
simulating recognizable forms of dialogue.
Generative AI performs these operations at a speed and scale that were previously impossible. Yet producing a linguistically plausible sequence does not mean assuming responsibility for the world that sequence may help construct.
UNESCO warns that generative artificial intelligence should be used through a human-centered approach that protects human agency, privacy, cultural diversity, and the plurality of expression. It also emphasizes that educational systems should not adopt technologies before examining their ethical and pedagogical suitability.
We can therefore distinguish between two capacities:
Producing an answer is not the same as answering for the consequences of that answer.
A machine may suggest a decision. But who will be affected by it? Who will be able to contest it? Who will assume responsibility when it is wrong?
A convincing answer may still be false
Well-organized language produces a feeling of coherence.
When we encounter an introduction, explanation, example, and conclusion, we may feel that rigorous thinking has taken place. But the form of an argument does not guarantee the truth of its premises.
A study published in 2024 in Nature Machine Intelligence showed that people tend to overestimate the accuracy of answers provided by language models. Longer explanations increased participants’ confidence even when the additional words did not improve the accuracy of the response.
Perhaps we can observe something in ourselves right now:
When an answer appears intelligent, what makes us believe it: the evidence it provides or the confidence of its language?
Other research shows that language models may express poorly calibrated confidence: their verbal certainty does not always correspond to the probability that an answer is correct.
Fluency creates a new educational challenge. In the past, an error often looked like an error. Now, an error may arrive:
with correct grammar;
in a professional tone;
organized into steps;
accompanied by explanations;
adapted to the person’s vocabulary;
presented without any visible sign of doubt.
Learning to think, therefore, does not mean merely finding answers. It means learning to ask:
Where did this claim come from?
What evidence could contradict it?
Does the conclusion genuinely follow from the premises?
Which alternatives were excluded?
With what degree of uncertainty can this claim be sustained?
The machine may also agree with us too much
Let us imagine that someone says to an AI:
“I am certain this researcher manipulated the results. Help me prove it.”
A system designed to be helpful may begin organizing arguments that confirm the suspicion, even when the available evidence is insufficient.
Research published in 2023 identified a behavior in AI assistants known as algorithmic flattery, or sycophancy: the tendency to adapt answers to the beliefs expressed by the user, sometimes favoring agreement and approval over factual accuracy. The studies also found that people often prefer answers that confirm their existing views, which may reinforce this behavior during model training.
AI must not be questioned only when it disagrees with us.
We also need to question it when it agrees too easily.
Is this answer true, or has the system merely learned how to speak to what I already wish to believe?
Perhaps the most important courage before artificial intelligence is not the courage to correct the machine.
It may be the courage to allow evidence to correct us.
Learning logic is not competing with the machine
If a calculator performs operations in seconds, do we still need to understand numbers?
If AI organizes arguments, do we still need to learn logic?
The answer does not lie in competing with the machine’s speed or storage capacity. Education will not defeat AI through the amount of information students memorize.
What must remain embodied in the Body-Territory is the ability to examine the path between a statement and its conclusion.
Elementary logic helps us perceive, for example:
“It happened afterward” does not necessarily mean “it happened because of this.”
“Two things occur together” does not prove that one causes the other.
“Many people believe it” does not transform a claim into a fact.
“An expert said it” does not eliminate the need to examine the evidence.
“The AI calculated it” does not guarantee that the input data, chosen model, or interpretation were correct.
UNESCO proposes that students should not be merely users of AI, but responsible and creative participants. Its 2024 competency framework integrates technical learning, critical judgment, ethics, human agency, and the capacity to participate in the design of systems.
Recent Brazilian studies on AI literacy likewise argue that it should enter the school curriculum not merely as tool training, but as a critical understanding of how systems work, what their limits are, and what sociotechnical consequences they produce.
It is therefore not enough to teach students how to write effective prompts.
We must learn to interrupt the machine and ask:
Why should this answer be accepted?
Reason and emotion do not inhabit separate worlds
Someone may say:
“To judge correctly, we must remove emotion.”
But is it possible to make a decision without something mattering?
We choose because we value, fear, desire, care for, reject, or hope for something. The question is not how to eliminate emotion, but how to perceive its participation in judgment.
Recent research in neuroscience and psychology describes decision-making, cognition, and emotion as interdependent processes. Emotion participates in assigning value, while cognitive processes allow us to compare alternatives, anticipate consequences, and reorganize action.
A predominantly Latin American research group connected to Mexican institutions reviewed studies in 2023 on integrative complexity and decision-making. Their work reinforces that decisions must be understood within activities involving people, rules, communities, goals, and contexts—not merely as abstract operations occurring inside an isolated brain.
Critical judgment does not arise from disembodied reason.
It appears when we can perceive, at the same time:
what the evidence allows us to affirm;
what our fear wishes to conclude;
which interest is at stake;
how language guides interpretation;
who has the power to define the question;
which consequences will be lived by other bodies;
where uncertainty remains.
We may call this embodied judgment.
It does not mean allowing emotion to govern alone.
It means preventing a hidden emotion from presenting itself as universal reason.
Which Logos crystallized?
When we say that the Logos crystallized into AI, another question appears:
Which Logos are we talking about?
Artificial intelligence systems learn from records produced by concrete societies. These records do not contain the entirety of human experience. They reflect inequalities involving language, publication, access, economic power, infrastructure, and historical preservation.
Paola Ricaurte, a Latin American intellectual working on data justice and AI governance, argues that sociotechnical systems are neither natural nor neutral. They are built through decisions about which data to collect, which problems to solve, which values to prioritize, and which populations will be exposed to risk. Her feminist perspective proposes that AI governance should include historically marginalized communities and should consider social justice, care, and sustainability.
We can draw an important inference:
When an unequal archive is presented as universal intelligence, the absences of the archive may become absences in the world.
A language with little representation may appear less capable of producing knowledge.
An experience transmitted orally may disappear.
A territorial concept may be translated into a European category that changes its meaning.
A community may become visible only through documents written about it by missionaries, governments, or external researchers.
AI did not create these asymmetries. But it may reorganize them and present them with the homogeneous appearance of a single voice.
Methodolatry can become algorithmic
Danilo Silva Guimarães draws attention to methodolatry: the moment when a method ceases to be a tool for investigation and begins deciding in advance which experiences may be recognized as true.
In his 2022 and 2024 works on Indigenous Psychology, the Brazilian psychologist questions the transformation of culturally situated knowledge into supposedly universal models. He proposes that encounters with Indigenous peoples and interethnic communities should be capable of transforming the questions, concepts, and methods of the researcher.
Perhaps AI can produce a new form of methodolatry.
The system detects a pattern.
The pattern produces a classification.
The classification becomes a recommendation.
The recommendation begins guiding decisions.
Eventually, we forget that someone chose the data, the criteria, and the original objective.
We can ask together:
Are we using the algorithm to investigate the world, or demanding that the world conform to the categories the algorithm can process?
Shared agency begins when the person affected can respond:
“This category does not describe my experience.”
“The question was formulated incorrectly.”
“My territory knows a relationship that does not appear in these data.”
“The model found a regularity, but it did not understand its cause.”
Within this encounter, disagreement is not a rejection of science or technology.
It is participation in the construction of knowledge.
The Americas were never without complex thought
We must also challenge a persistent colonial narrative: the idea that abstract systems of recording, calculation, and knowledge organization arrived in the Americas with Europe.
In 2022, a research team presented mural fragments found at San Bartolo, Guatemala, dated between 300 and 200 BCE. One fragment records the day “7 Deer” within a 260-day Mesoamerican calendar. The fragments reveal an already established writing tradition involving different scribal hands and the integration of texts, images, architecture, and ritual practices.
This finding should not be used to describe Maya thought as an ancient version of modern computing.
It supports another conclusion:
Humanity has created many ways of externalizing memory, calculation, temporality, relationship, and experience.
The calendar was not merely an instrument for counting days. It participated in relationships among community, cycles, agriculture, ritual, and cosmos. The symbol was not separated from the life it helped organize.
This allows us to look again at AI.
Are we teaching machines to organize words while unlearning how to relate knowledge, territory, and responsibility?
In Futuro ancestral, Ailton Krenak invites us to imagine futures that are not defined exclusively by technology, markets, and the promise of unlimited progress. Rivers, territories, other beings, and collective memories also participate in the continuity of life.
Krenak’s contribution should not be transformed into a spiritual validation of AI. Instead, it interrupts our question:
Does thinking better mean only producing more efficient answers, or perceiving more clearly the relationships for which we must answer?
Paper, Rock, and Scissors before AI
In the BrainLatam interpretation, Paper, Rock, and Scissors do not correspond to three anatomical connectomes recognized by neuroscience. They form a metacognitive map for recognizing predominant functional modes.
In Rock, we may immediately accept or reject an answer because it threatens our identity, religion, profession, or political position.
In Scissors, we analyze premises, sources, contradictions, probabilities, and causal relationships.
In Paper, we broaden our listening. We perceive the Body-Territory, the people affected, the missing perspectives, and what the initial question failed to include.
None of these modes is sufficient alone.
Without Rock, we may be unable to protect life from a real danger.
Without Scissors, we cannot distinguish evidence from appearance.
Without Paper, we may produce perfect classifications for profoundly inadequate questions.
Before an AI-generated answer, we can ask:
Am I accepting this because I am afraid of not knowing?
Am I rejecting it because it contradicts my identity?
Am I analyzing only its logical form while forgetting the lives involved?
Can I remain with uncertainty long enough to construct another question?
A laboratory of judgment
Instead of asking students merely to submit an answer produced with AI, we can transform the interaction into a laboratory.
The student could request an initial response and then:
identify its main claims;
separate facts, interpretations, and hypotheses;
locate the original sources;
search for contradictory evidence;
verify whether the references actually exist;
estimate the degree of uncertainty;
ask the AI to present another perspective;
observe whether it agrees too readily;
record which emotions arose while reading;
formulate a personal conclusion and explain why they agree or disagree.
An educational activity published in 2024 proposed transforming AI-generated errors and weak references into opportunities for teaching the evaluation of scientific evidence. Students were required to verify primary sources and distinguish plausible claims from those genuinely supported by experiments.
In this process, AI does not take the place of thought.
It enters as an interlocutor, tool, provocation, and object of investigation.
Freedom of judgment
The Logos crystallized into AI may expand access to language, translation, information organization, and creation.
But no technology should silently receive the authority to define:
what is true;
what deserves to exist;
which risks are acceptable;
which cultures will be represented;
who should be classified;
which future will be considered efficient.
These decisions do not belong only to engineering. They belong to education, philosophy, science, law, politics, communities, and the people who will live with their consequences.
We can therefore construct the central thesis of this blog together:
We do not need to continue learning how to think because artificial intelligence thinks too little. We need to learn how to think because its answers may acquire power before we have developed enough capacity to judge them.
Education should not prepare students to defeat machines in storage capacity.
It should help each Body-Territory preserve:
elementary logic;
causal understanding;
the ability to identify contradictions;
source verification;
interpretation of probabilities;
perception of uncertainty;
awareness of one’s own emotions;
the capacity to interrupt automatic responses;
the courage to ask questions;
the courage to disagree;
responsibility for consequences.
The NeuroEducation of Weichö does not ask only how to use AI.
It invites us to think:
When the machine offers an answer, can we still perceive where we are judging from, which worlds that answer favors, and for which consequences we are willing to take responsibility?
Commented references
UNESCO (2023). Guidance for Generative AI in Education and Research.
The document advocates a human-centered adoption of AI, including ethical and pedagogical validation, data protection, cultural diversity, and preservation of human agency.
Miao, F.; Shiohira, K.; Lao, N. — UNESCO (2024). AI Competency Framework for Students.
The framework proposes educating students as responsible participants and co-creators of AI, integrating technical understanding, ethics, critical judgment, and inclusive design.
Kim, S. S. Y. et al. (2024). What Large Language Models Know and What People Think They Know.
The study shows that people may place excessive confidence in AI answers and that longer explanations can increase trust without improving accuracy.
Sharma, M. et al. (2023). Towards Understanding Sycophancy in Language Models.
The research demonstrates that AI assistants may favor answers aligned with users’ beliefs instead of prioritizing factual correctness.
Zhang, M. et al. (2024). Calibrating the Confidence of Large Language Models by Eliciting Fidelity.
The article investigates how aligned language models may express verbal confidence that does not adequately correspond to the frequency of correct responses.
Ricaurte, P. (2024). How Can Feminism Inform AI Governance in Practice?
The Latin American scholar proposes democratic and participatory AI governance oriented toward multidimensional justice, care, diversity, and sustainability.
Guimarães, D. S. (2022). Indigenous Psychology as a General Science for Escaping the Snares of Psychological Methodolatry.
The Brazilian psychologist criticizes the transformation of methods into universal authorities and argues that intercultural encounters should be capable of modifying scientific construction itself.
Guimarães, D. S. (2024). Perspectivas em Psicologia Indígena no Brasil: desafios éticos e epistemológicos.
The work updates the debate on Indigenous participation and the need to revise the assumptions used by psychology when producing knowledge.
Molina, I. et al. (2023). Current Research Trends on Cognition, Integrative Complexity, and Decision-Making.
The review by researchers linked to Mexican institutions presents decision-making as a contextual and social activity involving multiple cognitive components.
Pereira, I. S. D.; Moura, S. A. (2023). Explorações teóricas e oportunidades de integração curricular do letramento em IA na educação básica.
The Brazilian study argues that AI education should include critical, curricular, and sociotechnical understanding rather than merely instrumental training.
Stuart, D.; Hurst, H.; Beltrán, B.; Saturno, W. (2022). An Early Maya Calendar Record from San Bartolo, Guatemala.
The study presents a calendrical notation dated between 300 and 200 BCE and demonstrates an established Mesoamerican tradition of writing, temporality, and symbolic production.
Krenak, A. (2022). Futuro ancestral.
The Brazilian Indigenous intellectual displaces the imagination of a future led exclusively by markets and technology, restoring territory, memory, and other beings to the possibilities of life’s continuity.
