OPINION | With AI, where should parents draw the line?

A generation of schoolchildren grew up with a guidebook tucked beside the textbook. Every question had an answer. Some children copied it without a second thought. Others reached for it only after working through the problem themselves. The book was identical in both cases. What differed was whether the child had done the thinking first.

Artificial intelligence has put a guidebook in every pocket, one that writes its own answers, and the old rule holds. It serves the child who knows the subject and misleads the one who does not.

In April 2026, the Union Education Minister launched a CBSE curriculum on computational thinking and AI for Classes 3 to 8, to be implemented in the 2026–27 academic session. Learning modules, teacher handbooks and an assessment framework come with it, and the classroom work is built around puzzles, games and group tasks.

Assessment shifts from rote recall to continuous, competency-based evaluation, with teachers keeping observation journals to track each child’s progress. For Classes 6 to 12, the skills ministry’s SOAR programme adds three 15-hour student modules and a 45-hour module for teachers, with ethical use among its themes.

Similar efforts to introduce AI learning in schools reflect a growing recognition that familiarity with the technology is not enough; children must also learn to assess its outputs and understand its limitations.

A curriculum, though, only sets the starting point. What matters is whether a child can tell a plausible answer from a correct one, and explain the reasoning without a machine doing it for them. That ability comes from knowing the subject. A chatbot can give a wrong date or skip a step in a proof, all in the same assured tone. Only a reader who knows the material catches the lapse and answers with a reason.

This is what “questioning the AI” means: expertise at work, the ability to say, “That step does not follow.” A child who pastes an answer that could not have been checked does more than borrow a result. The practice that builds the checker is skipped, and each skipped round makes the next dependence likelier.

The Ministry of Education’s UDISE+ report for 2024-25 found computers in 64.7 per cent of schools and internet access in 63.5 per cent. Roughly a third of schools lack computers, and a similar share lack internet. In those schools, a curriculum built on puzzles, games and worksheets does not depend on a device. Where devices are plentiful, the shortcut sits one tap away, and the discipline falls on parents and teachers.

Access, however, is only one part of the equation. Having access to AI does not mean a child knows how to use it well, just as limited access does not prevent a child from developing the reasoning skills needed to question its outputs. What matters is whether learning gives children the opportunity to test ideas, recognise errors and understand why an answer holds up—or does not. Practical exercises that make AI's workings visible can help bridge that gap.

One approach is to teach by building. Live online AI and coding classes for kids and teens, designed to help them become the innovators of tomorrow, have learners train small AI models on real data, test them and debug where they fail, with a teacher watching the reasoning. Building and breaking a model is the nearest classroom practice to questioning an output. The child sees that such systems find patterns and depend on their data.

The line, then, runs through the work itself. A parent can let the child attempt the problem or draft first, and bring in AI afterwards to explain, offer examples or check finished work. Asking “Is this right, and how would you check?” turns the child into the reviewer. A simple test at the dinner table helps: if the child cannot explain the answer step by step, the thinking happened elsewhere. Schools can set rules by assignment, marking AI as allowed, allowed with disclosure, or not allowed. They can ask for the working along with the answer, and use the observation journal already in the framework to record how a child reasons. At the system level, tasks that ask students to find the error in an AI output would build the same skill, and the 45-hour teacher module could teach it.

The sum still has to be worked first, whether the answer sits printed at the back of the book or appears on a screen. Only then, on the day that answer is wrong, is there a child who knows.

The author is the CEO of the learning platform Codingal.

The opinions expressed in this article are those of the author and do not purport to reflect the opinions or views of THE WEEK.