Make Them Practice Anyway

Why should the child bother?

An Indian mother told ThePrint in 2025 that her son, once an 80 percent student, began using ChatGPT for all his homework. According to her, he stopped trying to learn. His grades fell, and he grew afraid of exams. The homework was done every night. The boy had not learned to do it.

That story holds the question every parent now faces. Why should a child spend half an hour on a problem AI solves in three seconds?

The usual answers, learning, discipline, thinking for oneself, are reasonable. I think they miss something more basic. When a child struggles and finally succeeds, the result is not the only thing that changes. The child does. Yesterday she could not. Today she can.

The psychologist Erik Erikson saw this as one of the central tasks of the school years, the stage he called industry versus inferiority. Children learn that effort produces things that matter, and when someone they respect notices, a lasting belief forms: I am someone who can. It is not the whole of a person; relationships, play, and belonging matter just as much. But it is one of the foundations, and adolescence builds an identity on it.

That foundation is being laid just as machines make the struggle optional. Adults already feel it. A young professional in a study of ChatGPT users asked: "If the machine is as good as me, then what use am I?" Now imagine that question at nine.

The result is not the achievement

Three things are easy to confuse: producing something, knowing something, and becoming capable of something. For generations, finished work (a drawing, a solved problem, a piece played without stopping) was imperfect but useful evidence of the third. It mattered partly because it was hard to produce.

AI weakens that link. A child can now produce something impressive without the ability to produce it alone. A child who has learned the piano still knows she can play; nothing takes that away. What fades is the incentive to begin, when the output comes without the learning. Curiosity and play get children started, but the long middle of learning anything, the drills, the stretch of being bad at it, rarely runs on joy. That is where the machine offers a way out. The danger is not what machines can do. It is that children may learn to confuse having an answer with knowing how to reach one.

AI is not only a threat. It can teach, patiently and at any hour, and learning to question, direct, and judge an AI is a real skill of its own. What matters is whether it builds ability or replaces it, and the tutor's design decides much of that. In one experiment, nearly a thousand students at a Turkish high school got unrestricted GPT-4, a version that gave hints instead of answers, or no AI. With the AI taken away, the first group scored 17 percent worse on an unassisted exam than students who never had it. The hint group largely avoided the harm. Same machine, different demands on the child. Chess offers a model: play the game yourself, then run the engine to see where you went wrong. The engine becomes a critic of your effort, not a substitute for it.

Even a well-designed tutor needs a reason to be used. When schools offered Khan Academy's AI tutor for two years, most students barely used it productively, and Sal Khan said the study showed "how much motivation matters." The machine can explain. Something else has to bring the child back.

Form, person, audience

I am from India and live in Japan. Both cultures have old ways of carrying a beginner through the long stretch before practice pays off. As mechanisms rather than heritage, they come down to three things.

Form makes practice manageable. Japanese tea and martial arts call mastery shu-ha-ri: follow the form exactly, then break from it, then leave it behind. A beginner does not need to see the whole path. The form gives them the next step.

A person provides accountability and recognition. In Indian classical music, riyaz, daily practice, goes on for years under a guru who notices a skipped day. In Japan, the sensei. The teacher believes in the student before the student does.

An audience makes effort socially meaningful. Japanese lessons build toward the happyōkai recital. In Tamil tradition, years of training culminate in the arangetram, a first performance on stage. Work done for people weighs more than work done for a grade.

These traditions are not always gentle. A master can nurture or coerce, and many families in both countries never had access to them. But the mechanisms need neither tradition nor wealth. A form can be a printed sequence of exercises; a person, a grandparent, a cousin, or a neighbor who mastered something; an audience, the family at dinner.

Two childhoods, one machine

The same machine enters very different childhoods. I grew up in India in modest circumstances; my children are growing up in Japan with opportunities I did not have. I have stood close enough to both to see that AI poses different risks to each.

In a home with plenty, a child often has what a good education provides: teachers, books, time, a quiet room, adults who notice progress. AI arrives as one more convenience, and the risk is that it replaces effort.

In a home with little, a child may have ambition and nobody to explain a hard problem. Many such children are asked to build wealth their parents never had, and education is the route: practice for marks, marks for a way out. There, AI may be the first patient tutor the family has ever had.

One child needs reasons to struggle. The other needs someone to make the struggle possible.

The evidence cuts both ways. In ASER's 2023 survey of rural India, about nine in ten households had a smartphone, but a household phone is not a child's phone; among young people who could use one, young men were more than twice as likely as young women to own it. Fewer than half of 14- to 18-year-olds could do third or fourth grade division, so a tutor that explains algebra may find little to build on. UNESCO's 2021 report counted more than a million vacant teaching posts. Human help is scarcest where it is needed most. The rich will keep the humans.

Yet children in poor homes now reach much of the same underlying capability as children in rich ones, even if devices, connections, and languages still differ. In a World Bank pilot in Nigeria, teachers guided secondary students through a free AI tutor after school. In six weeks they gained the equivalent of one and a half to two years of typical schooling, and girls gained more than boys. The machine was powerful. The person beside it was part of the design.

I suspect AI will also loosen the old promise of practice for marks and marks for escape, as employers come to value different skills, though exams and credentials still guard the gates for now. But the capacity to struggle, improve, and act independently may become more valuable, not less. When output is cheap, what stays scarce is judgment about what is worth making, the initiative to start unprompted, and the stamina to stay with a problem that has no script.

Japan, short of children and long on elders, is a reminder that the people who can sit with a child are not only parents and paid teachers. Human attention can be shared well beyond the immediate family, if someone arranges it. Japan's local governments do. Through their Family Support Centers, about 50,000 people aged 60 or older have signed up to look after neighbors' children. There is another inheritance worth passing on: someone willing to sit with a child until the child can do it alone.

What practice is really for

Practice has usually had an external payoff: we learned to write to communicate, to calculate to solve problems, to perform to be paid or applauded. AI weakens some of those incentives, but not the deeper one. Practice builds agency: the ability to choose a hard goal, tolerate being bad at it, improve through effort, and know the improvement is yours. A person with agency can act without being continually directed by someone, or something, else.

Not all practice does this. A child drilled under relentless pressure can become highly skilled and still have little confidence in choosing a goal of their own. Agency grows from practice that is chosen and supported, not merely imposed. That is why the child picks the instrument, and the adult's job is to keep it from quietly disappearing, not to decide what it is.

It is also the difference between a good tutor and a good teacher. A machine can measure how much a child has improved. A person can care about who the child is becoming.

None of this requires adults to control what a child chooses to master. It requires them to protect the chance to struggle.

• Let the child choose one hard thing, and do not let it quietly disappear.
• Make sure there is a person who notices, hired or borrowed.
• Require the first attempt before the machine.
• Choose AI that gives hints, not answers, and check that the child can work without it.
• Give the work an audience of people.
• Count days, not scores.

When your child tells you the AI can play it, write it, or solve it for them, agree. It can. Then hand them the instrument anyway. The song was never the point. The child who can play it is.

Appendix: A parent's checklist for practice in the age of AI

I face this question with my own children. This list is my synthesis so far, and a checklist I keep for myself.

For every child

1. Pick one hard thing, together. An instrument, a sport, a craft, a language. The child chooses. Quitting is allowed at a natural stopping point, never on a bad day.
2. Find a human. Hire one if you can. Borrow one if you cannot: a grandparent, a cousin, a neighbor who mastered something.
3. Copy before you create. Trace, imitate, repeat. Originality is the third stage, not the first.
4. No AI on the first attempt. The child's own draft first, then the machine as critic, the way chess players run the engine after the game.
5. Let them be stuck. Agree in advance how long they wrestle with a problem before anyone helps, you or the machine.
6. Book the recital. A performance in the living room, a meal for the family, a story read to grandparents.
7. Count days, not scores. In the Nigeria pilot, the students who attended most gained most.
8. Be a beginner in front of them. Children copy what parents do more than what parents say.
9. Tell them what the machine can and cannot do. It can measure their progress. It does not care who they become.
10. Celebrate the first rebellion, after the form is mastered. Breaking a form you know well is how it becomes yours. Abandoning one you never learned, to escape the discomfort, is only a shortcut.

If a child is starting with less

For these children, struggle is not scarce. It is constant, and rarely chosen. What is scarce is margin (a second attempt, a buffer when a parent falls ill), a map (someone who knows which exam, which form, which scholarship), and the safety to say "I don't understand" without shame. The child carries the family's hope, and hope that heavy can feel like fear.

11. Let them ask the machine. A child afraid to ask in class can ask a machine at midnight, in Tamil, as often as needed, and no one laughs. Keep it a tutor, not an answer key: hints first, then the child's own work.
12. Guard the marks by guarding the effort. Here marks are not vanity: for now they decide scholarships and seats. Homework the machine did is a debt the exam collects. In the Turkish study, it cost 17 percent.
13. Be the map. Many of these children fall short for lack of information, not effort. Someone who has been through it can be worth more than a tutor. The machine can help too, reading a form or explaining a rule in the family's language.
14. Pay for the margin. If you have more, the most useful help is often small: an exam fee, a data pack, bus fare, a quiet room in exam month. Ask the family what is in the way; do not guess.
15. Go back for one child. If you started with less and made it through, you are the map, the margin, and proof that people like them can. Sit with one child until the child can do it alone. In an Indian experiment, young women from the community, not teachers, raised the scores of children falling behind.

@misc{palaniappan2026practiceanyway,
  author       = {Palaniappan, Sucheendra Kumar},
  title        = {Make Them Practice Anyway},
  year         = {2026},
  howpublished = {\url{https://suchee.org/musings/make-them-practice-anyway/}},
  note         = {Essay, suchee.org}
}