There is a moment I keep returning to. My child — eleven years old, homework
due the next morning — typed a question into ChatGPT, read the response once,
and copied it down. No pause. No verification. No curiosity about where the
answer came from or whether it could be wrong. Just a clean, confident
paragraph, and the laptop snapping shut.
I did not intervene that evening. I sat with the discomfort instead. Because what I
had just watched was not laziness. It was trust — a child placing complete,
uncritical faith in a tool they did not understand. And as both a teacher and a
parent, that bothered me in a way that took me a while to fully articulate.
• • •
Two Very Different Classrooms
I have spent a large part of my professional life teaching at the postgraduate level
— MBA students, research scholars, doctoral candidates. The conversations in
those classrooms are rigorous. Students are trained to interrogate sources, to
question assumptions, to hold conclusions lightly until the evidence is robust.
When I assign a case or a research paper, the expectation is not that students will
absorb them — it is that students will argue with them.
At home, I was watching something different. My children — sharp, curious,
perfectly capable of skepticism — were not applying any of that skepticism to AI.
They used it the way their generation use google maps or ride hailing apps:
confidently, without needing to understand how it works, and without much
thought about what happens when it fails.
• • •
What I Actually Believe About Teaching
I have one core conviction as a teacher, and it has not changed across fifteen
years and two very different kinds of classroom: a good teacher makes herself
redundant.
That sounds counterintuitive. Teaching is supposed to be about transferring
knowledge, building expertise, filling gaps. But I think the best teaching does
something else. It does not give students answers — it gives them the capacity to
find answers. It does not teach concepts — it teaches students how to learn. The
goal is a student who, when they encounter something unfamiliar, does not freeze
and look for an authority. They lean in. They start pulling at threads. They know
how to learn their way to understanding.
At the MBA level, this shows up in how I run case discussions. I do not resolve
the ambiguity in a case. I sit in it with the students and make them sit in it too.
The discomfort is the point. At home, I try to do the same thing: when my
children bring me a question, my first instinct is almost never to answer it. It is to
ask, “What do you already know? What would help you find out?”
Students — all students, at any age — are far more capable than we assume. They
just need the occasional nudge to stay on their own path of discovery, rather than
being handed a shortcut off it.
• • •
The ChatGPT Problem Is Not What Most People Think It Is
When educators and parents worry about AI tools in the classroom, the
conversation usually goes to one of two places: plagiarism, or screen time. Both
are real concerns. But neither is the one that worries me most.
What worries me is epistemological. It is what my child demonstrated that
evening: the unexamined belief that a fluent, confident answer is a correct one.
Large language models are very good at sounding authoritative. They are also
capable of being confidently, smoothly wrong — and of encoding, in their
outputs, the biases of the humans whose data trained them. A model trained
predominantly on one kind of writing, from one kind of source, in one kind of
cultural context will reflect that.
https://www.thehindu.com/opinion/op-ed/why-algorithmic-sovereignty-
should-be-indias-top-priority/article70723855.ece
The student who does not know this cannot account for it.
This is not the model’s fault. It is a design reality. But it is a reality that children
are currently navigating without a map.
Corporations will build guardrails. Governments, we hope, will regulate. Content
filters, output warnings, usage policies — these will all improve. But the most
durable guardrail is not external. It is internal. A student who understands, even
roughly, how an AI system is built — where its knowledge comes from, what it
optimizes for, where it reliably stumbles — is a student who can use these tools
safely and critically. That understanding cannot be outsourced to a policy or a
warning label.
How This Shaped the Course
When we sat down to design Train the Machine — an introductory AI and ML
course for learners from around Grade 7 upwards — these two streams of thought
were present throughout. The educator who wants students to learn how to learn.
The parent who watched her child place uncritical trust in a system she did not
understand.
The first chapter does not open with a definition. It opens with a
question: when you hear the words ‘artificial intelligence,’ what picture comes to
mind? A robot? A supercomputer? A mystery? Students are invited to name their
mental image before any content is introduced. The lesson’s explicit goal is to
‘replace the magic’ — not to make AI seem less impressive, but to make it
understandable. Because a mystery cannot be questioned. A system that is even
partially understood can be.
The Maya story — a recurring character used through the course to illustrate how
machines learn — was a deliberate choice rooted in the same instinct. Abstract
definitions do not build durable understanding. Stories do. When a student in Chapter 7, encountering clustering for the first time, is reminded that Maya once figured out who was a stranger and who was a known face — without anyone explicitly telling her — and developed stranger anxiety as a result, they are not meeting a new idea. They are formalizing something they already felt.
The ‘Pause and Reflect’ moments embedded throughout the reading
material are the direct classroom equivalent of my instinct not to answer my children’s questions directly. They are invitations to think before being told. They
surface intuitions before concepts are named. They are the nudge toward self-
discovery, built into the text itself.
And the 'Bias and Fairness' chapter — positioned not as a footnote or an ethics
elective, but as a full, structured chapter with the same rigor as the technical
ones — is perhaps the most direct expression of what motivated the course in the
first place. Students are asked to train a model on data where color incorrectly
predicts emotional expression. They observe the model making confident, wrong
predictions. Then they are asked one question: what is one thing a machine
learning model cannot do on its own, without humans?
The right answer is not technical. The right answer is that the model
cannot evaluate the quality of what it learned. It cannot recognize its
own blindspots. That is the human’s job.
• • •
What I Want for My Students — All of Them
I want my MBA students to use AI in their careers. The tools are extraordinary
and the managers who understand them will have a significant edge. But I want
them to understand enough to lead the humans and machines together, not
simply to defer to whichever output arrives fastest.
I want my children to use AI too. This is unambiguously where the immediate
future is headed, and resisting it on principle would be both futile and unfair to
them. But I want them to use it the way a skilled navigator uses a GPS: with an
understanding of the underlying geography, a healthy awareness that the map
can be wrong, and the confidence to take a different route when something feels
off.
The goal is not AI skepticism. It is AI literacy. And literacy, in any domain, begins
with the same thing it has always begun with: understanding enough about how
something works to know when to trust it, when to question it, and when to set it
aside and think for yourself.
• • •
A Note for Educators and Curriculum Designers
If you are designing AI literacy programs, or thinking about how to bring
critical thinking about AI into your classroom, I want to offer one observation
from both sides of this experience.
The technical content matters. Understanding that AI learns from data — not
from rules — is genuinely clarifying. Understanding that training data encodes
the assumptions of whoever collected it is the foundation of any meaningful
critique. Students who have this knowledge are better equipped than students
who do not.
But the technical content alone is not enough. The pedagogical approach matters
as much as the curriculum. If we teach AI literacy the way we often teach
technology — as a set of facts to be absorbed and definitions to be memorized —
we will produce students who can pass a quiz but still copy the first ChatGPT
answer without a second thought.
The deeper goal is to build the disposition to question. Not to distrust AI, but to
engage with it actively rather than passively. To ask, before accepting an answer:
where did this come from? What might it be missing? Who built this, and what
did they optimize for?
These are not AI-specific questions. They are the questions every good teacher
has always wanted students to ask of every source, every authority, every
confident-sounding answer. AI just makes them more urgent — and, if we teach
well, a little more practiced.
• • •
The best outcome I can imagine for a student who has worked through ‘Train the Machine’ is not that they only know what machine learning is. It is that the next time they receive a fluent, confident answer from an AI tool, something small shifts in how they receive it. A thread gets pulled. They do not simply accept — they question.
Because, as we have kept saying, understanding AI is not for engineers alone. Future doctors, lawyers, teachers, bank managers, lawmakers — everyone will rely heavily on AI. Understanding the intuition behind it is what will help them arrive at decisions that are genuinely their own.
That is the nudge we are trying to build — and in our experience, it is a meaningful one. The curiosity, the critical thinking, the capacity to learn independently: every student carries these. What the course does is give them shape and direction. It puts the right questions in front of you at the right moment, so that what was always in you has somewhere to grow. The nudge is the door. The path is theirs. It always is.