The Cognition Paradox

Bharat has crossed its pedagogical Rubicon. From the academic year 202G–27, Artificial Intelligence and Computational Thinking will become mandatory subjects from Class 3 onwards, as announced by the Ministry of Education in late 2025 and aligned with NEP 2020 and NCF SE 2023.

The stated vision is admirable: AI literacy as a basic universal skill, embedded early to advance innovation and “AI for Public Good.”

Yet beneath this ambition lies a question that has not been asked loudly enough:

What happens when children learn to query machines before they learn to question reality itself?

Richard Feynman famously noted that the first principle is that you must not fool yourself, and you are the easiest person to fool. In an AI-saturated classroom, this self-deception scales at the speed of light. Students mistake the fluency of the prompt for the mastery of the subject. The cognitive architecture we embed today may produce not empowered thinkers, but intellectually dependent ones.

This is not merelyspeculative; emerging empirical patterns warrant caution.

The Evidence We Cannot Ignore

Empirical research from systems that have already adopted AI in classrooms should give us pause.

At MIT’s Media Lab, researchers led by Nataliya Kosmyna tracked 54 adult participants (aged roughly 18-35) over several months, monitoring brain activity with EEG as they wrote SAT-style essays. Participants were divided into three groups: those using ChatGPT, those using Google Search, and those working unaided. The results were stark: those using ChatGPT showed the lowest neural engagement and connectivity during writing. More troubling, many struggled to recall key points from their own essays shortly afterwards. Kosmyna termed this buildup of deferred mental effort “Cognitive Debt” – the more one outsources thinking, the less one encodes into long-term memory. Teachers assessing the essays described them as “structurally perfect but intellectually empty” or “soulless” – a hallucination of competence where the student’s voice had been replaced by a statistical average.

A field experimentled by Wharton School researchers (University of Pennsylvania), involving nearly 1,000 high-school math students in Turkey, revealed a similar trap. Students using an unguarded ChatGPT-like interface (GPT Base) solved 48 per cent more practice problems correctly during sessions, working faster and feeling more confident. Yet when tested without AI access, they scored 17 per cent worse than peers who never used it – highlighting dependency rather than deepened understanding.

While these studies involve older students and specific tasks, they illustrate risks that could compound in younger learners whose cognitive habits are still forming.

The crucial distinction is simple but profound: whether AI does the thinking, or makes you think. The former creates dependency; the latter builds capacity. India is about to run both experiments simultaneously on 26 crore students, with a rollout timeline measured in months. That would be ambitious for a pilot programme. For a civilisational-scale intervention, it is breathtaking.

The Missing Pedagogy

Current discussions around AI in education focus on access, inclusion, and skills. They address what to teach and how to deliver it – teacher training, digital resources, and age-appropriate modules. What they rarely address is Cognitive Sequencing: when to introduce cognitive shortcuts, and when to withhold them so that thinking itself can mature.

AI does not merely assist learning; it reshapes habits of mind. It shifts students from reasoning to requesting, from exploration to optimisation, from grappling with uncertainty to seeking instant coherence. Treated incautiously, it risks eliminating the very struggle through which learning occurs.

What is missing is not more technology, but Cognitive Sovereignty – the ability to think independently in the presence of powerful cognitive aids.

This is the Cognitive Gap that purely technical curricula ignore. To bridge it, we do not need more code; we need a sturdier internal operating system. Bharat already possesses a conceptual framework capable of addressing this challenge, if we are willing to take it seriously.

Anvīkṣikī, Abhyāsa, Yog: A Cognitive Counterweight

Long before “critical thinking” entered modern pedagogical vocabulary, Indian epistemology articulated a triad that aligns remarkably well with what contemporary cognitive science now demands in an AI-saturated environment.

* Anvīkṣikī

It is the art of Strategic Scepticism. Described by Kautilya as the lamp of all knowledge, Anvīkṣikī is disciplined inquiry. It is not merely evaluating answers, but interrogating assumptions – examining how a conclusion is reached, what premises it rests on, and where one’s own perspective shapes the question itself. A student trained in Anvīkṣikī does not simply ask whether an AI-generated response is correct. They ask whether the question was well-formed, and what alternative framings might reveal.

* Abhyāsa

In the age of ‘Instant Everything,’ Abhyāsa reclaims friction as a feature, not a bug. Often reduced to repetition, Abhyāsa refers to sustained, effortful cultivation toward mastery. Learning consolidates through friction – through error, revision, and repeated engagement – not through convenience.

Where AI optimises for speed and fluency, Abhyāsa preserves slowness and struggle. It ensures that understanding is internalised, not merely accessed on demand.

* Yog

Not exercise or physical movements, but integration. Yog concerns attentional discipline, emotional regulation, and coherence between thought, action, and purpose. In an environment saturated with prompts, notifications, and algorithmic nudges, Yog restores the capacity to pause, focus, and choose deliberately. It is the antidote to the algorithmic nudge. It ensures that students are not merely capable of thinking, but capable of deciding when and how to think.

Together, this triad integrates inquiry, effort, and self-regulation into a single cognitive architecture. It does not resist artificial intelligence. It conditions the human mind to engage with AI without being displaced by it.

What This Looks Like in Practice

Consider a classroom in the early years.

A Class 3 student watches a leaf fall in the school playground. Before the digital abstraction of a screen intervenes, the teacher grounds them in direct observation: What did you notice? How did it move? Did all leaves fall the same way? The children mimic the motion with their hands. Someone mentions a feather. Another recalls a paper aeroplane. Only later does the teacher introduce AI – showing time-lapse videos of seasonal change or slow-motion footage of falling objects.

Here, Anvīkṣikī begins as observation before optimisation. Abhyāsa appears in repeated observation and articulation. Yog shows up in the patience to sit with uncertainty rather than rushing to explanation. AI amplifies curiosity; it does not replace it.

At later stages, the sameprinciples scale differently. Adolescents might use AI to generate arguments and counterarguments, then reconstruct the reasoning without it – defending positions orally and in writing. The learning lies not in the output, but in the comparison: between borrowed coherence and earned understanding.

What is at Stake – and Why This Moment Matters

This debate is not about rejecting modernity or romanticising tradition. It is about the kind of cognitive citizens Bharat intends to cultivate.

If AI becomes a crutch before students learn to walk, the employability gap will widen rather than narrow. More importantly, we risk producing generations fluent in prompts yet fragile in judgment – efficient operators of intelligence they do not possess.

This is not an abstract concern for some distant future; curriculum decisions are being made now.

The programme framework for AI education is still being shaped. That window matters. It allows us to embed Cognitive Sovereignty alongside technical fluency, discernment alongside access, and formation alongside skill.

We can teach our children to query AI efficiently, or we can teach them to question reality deeply. The curriculum we design in 202G will determine which cognitive architecture defines the next generation of Indians.

That choice cannot be outsourced – not even to the most sophisticated AI.

Views expressed are personal. The writer is an Honorary Visiting Fellow at Bharat Ki Soch