I was invited to address nearly 500 students during their first week at GITAM University in Hyderabad. I was hosted by Dr. Motahar Reza, Director of the GITAM School of Science. A native of Murshidabad in West Bengal, Dr. Reza trained at IIT Kharagpur, worked across computational fluid mechanics, machine learning and data science, and joined GITAM in 2020. His journey from fluid dynamics to artificial intelligence made him an ideal host for a conversation about how learning itself is changing.

GITAM began in Visakhapatnam in 1980 as the Gandhi Institute of Technology and Management. The acronym has since become an educational brand, with campuses in Visakhapatnam, Hyderabad and Bengaluru. This year, the university is home to 28,000 full-time students—young, energetic, curious and ambitious. Yet scale alone does not define a university. What struck me was its carefully curated environment: orderly without being oppressive, ambitious without appearing anxious, and imbued with the Tagorean aspiration, ‘Where the mind is without fear, and the head is held high.’

I chose ‘The Three Habits in the AI Era’ as the subject of my address. My message was simple: do not fear artificial intelligence; form habits that help it amplify your intelligence. The three habits were: Ask, Speak and Write.

Ask AI one meaningful question every day. Once a week, select one idea that arises from those questions and speak about it for three minutes. Let AI transcribe the recording and polish the language. Then study the polished version: What became clearer? Did the machine improve the grammar, or alter the thought? Finally, write the improved version by hand in a dated notebook.

Asking cultivates curiosity; speaking reveals gaps in thought; writing consolidates learning and strengthens memory. The student also learns to inspect rather than passively accept an AI output. Sustained through four years, the notebook can become a personal intellectual autobiography—perhaps a 200-page book authored before graduation.

The students received the idea warmly. Afterwards, Dr. Reza hosted lunch with faculty colleagues and Dr. S. Chandrasekhar, one of the finest green chemists of our time and a former Secretary of the Government of India’s Department of Science and Technology. Despite his eminence, he possesses the warmth and humility of a true scholar. A university is revealed as much by conversations around a table as by lectures in an auditorium.

During lunch, my thoughts returned to mathematics. Mathematics has always fascinated me, although I could never claim mastery of it. During my master’s programme at G. B. Pant University, I chose Tensor Methods in Continuum Mechanics as an elective and only just secured a B grade. Yet the course gave me a sufficient grounding in matrix algebra to prove invaluable later in operational research and mechanical vibrations. During discussions at DRDL on six-degree-of-freedom modelling—the three translational and three rotational motions of a missile—I found myself unexpectedly confident and vocal.

This experience taught me that mathematical education cannot be judged only by marks secured at the moment of instruction. A concept imperfectly grasped today may become the instrument through which one understands a real-life system years later. Education deposits structures in the mind; application activates them.

I therefore asked Dr. Reza, “How do young students see mathematics today, and how is AI affecting its teaching and learning? Mathematics has always required a heady combination of rigour and intuition. Does AI strengthen that combination—or quietly undermine it?”

The answer lies not in the intelligence of the machine, but in the wisdom with which we choose to wield it.

Mathematics contains four interacting layers. The conceptual layer concerns meaning: What is a derivative, a vector or an eigenvalue? The procedural layer concerns operations such as differentiation and integration. The representational layer connects equations with graphs, tables, simulations and physical phenomena. The metacognitive layer concerns thinking about one’s own thinking. It asks: Why did I choose this method? Where could it fail? Is the answer logically, dimensionally and physically plausible?

Traditional classrooms often overemphasise procedure because it is easiest to demonstrate and examine. AI can perform many procedures rapidly. That does not make mathematics obsolete; it reveals that procedure was never the whole of mathematics. Done till here

A large language model is most useful at the conceptual and linguistic interface. It can explain an idea through an analogy, a graph, a physical example or simpler language. It can help students whose English is functional rather than fluent, generate graded examples, compare alternative methods, and ask Socratic questions. A computer algebra system manipulates symbols; a numerical solver approximates solutions that resist closed-form expression; a visualisation engine reveals parameter sensitivity; and a proof assistant verifies that deductions follow from stated axioms. Mathematical literacy now includes knowing which technology to trust for which task.

This is a neuro-symbolic partnership. Neural models excel at pattern recognition, language and analogy; symbolic systems are stronger at exact manipulation and formal verification. The human must still choose the assumptions, formulate the model, interpret the results and decide whether the answer is relevant.

The distinction matters because a language model generates plausible sequences, while mathematics demands valid relationships. A convincing explanation is not a mathematical proof. A confident derivation may contain a sign error, an unstated assumption or an invalid cancellation. One recent evaluation found that although 90 per cent of AI tutoring dialogues appeared instructionally strong, only just over half were mathematically correct throughout. OECD analyses likewise warn that generative AI may improve task performance without producing genuine learning when students outsource the intellectual work.

The pedagogical principle should therefore be:

AI may assist the struggle, but it must not abolish productive struggle.

A sound mathematics lesson in the AI era might follow five stages: Conjecture – Attempt – Dialogue – Verification – Reconstruction.

First, the student predicts what should happen. Second, the student attempts the problem unaided. Third, AI offers a hint, questions an assumption, presents an alternative representation or constructs a counterexample—without immediately supplying the answer. Fourth, the result is checked through substitution, limiting cases, dimensional analysis, numerical testing or formal proof. Fifth, the student reconstructs the argument in their own words and, where appropriate, by hand.

This is an AI sandwich: human effort before AI, critical interaction with AI, and human synthesis afterwards.

Teachers therefore cease to be mere transmitters of worked examples. They become designers of mathematical encounters, diagnosticians of misconceptions and guardians of epistemic standards. Instead of asking only whether the student obtained the answer, they ask whether the student can explain the method, identify the weakest step, test a boundary case, distinguish an exact result from an approximation, and justify the conclusion.

Assessment should evolve accordingly. Students can critique an AI-generated solution, locate a deliberately inserted error, compare methods, defend their assumptions orally, annotate a proof or maintain a handwritten reasoning journal. The trace of thought becomes more important than the final answer.

My three habits fit mathematics particularly well. Ask becomes the habit of forming a precise mathematical question. Speak becomes the ability to articulate a chain of reasoning and hear where it breaks. Write becomes the disciplined conversion of intuition into symbols, definitions and proof. AI can accompany all three, but it cannot assume responsibility for any of them.

The real danger is not that AI will make students weak at mathematics. It is that institutions will continue teaching mathematics as the production of answers when machines have become excellent answer-producing partners. The opportunity is to restore mathematics to its deeper purpose: modelling reality, detecting structure, reasoning under uncertainty and learning to recognise when one is wrong.

The classroom board will not disappear. Nor will the notebook. They will stand beside the bot. The university’s task is to ensure that the machine supplies speed and range while the human being retains curiosity, judgement and the courage to ask:

“Is this merely a convincing answer—or is it mathematically true?”

That question stayed with me as I made my way home. In my mind’s eye, Dr. Reza’s infectious smile returned, seeming to carry the simplest possible reply:

“It can be both.”