Column | ChatGPT Killed Homework. Can AI Tutors Save Learning In India?

Walk into almost any middle-class home in India around eight in the evening. The child is bent over a notebook, a parent is hovering, and a phone lies nearby, supplying the sentences.

This is happening everywhere. In the United States, the share of teenagers using ChatGPT for schoolwork doubled from 13 to 26 per cent in a single year. By late 2025, more than half said they had used a chatbot for it. Indian homes are no different; our surveys are just slower. Over a cup of tea in the staff room, teachers will tell you what they would never say at a parents' meeting: homework has stopped telling us what a child knows. So what do we let go of, and what do we hold on to?

The first instinct of many schools is to buy an AI detector. It is a poor bet. A Stanford study found that seven popular detectors flagged 61 per cent of essays by non-native English speakers as machine-written. OpenAI withdrew its own detector in 2023 because it caught only about a quarter of AI text. A child can be labelled a cheat by software that is only guessing. No school should punish a student on that basis.

Yet the machine that broke homework may also fix something far older. In 1984, the American psychologist Benjamin Bloom found that a child taught one-on-one by a tutor does better than 98 per cent of children in an ordinary classroom. The catch was cost. No country could afford a tutor for every child.

AI could rewrite that arithmetic. In a World Bank trial in Nigeria's Edo State, students who used an AI tutor after school for just six weeks gained as much as 1.5 to 2 years of normal schooling. In a Harvard physics course, students who learnt with an AI tutor gained more than twice as much as those in an active-learning class, and took less time. In Ghana, a maths tutor running on WhatsApp, an app most families already have, lifted scores over eight months. These are early studies. Still, for the first time in history, Bloom's tutor costs about as much as a data pack.

Not every experiment has ended well. In an MIT study, still awaiting peer review, 83 per cent of students who wrote essays with ChatGPT could not recall a single line of what they had written only moments earlier. South Korea spent over 1.2 trillion won on AI textbooks; school adoption then fell from 37 to 19 per cent, and the books lost their official status in 2025.

The best teachers I have known never gave away the answer. They asked a question, dropped a hint, and then waited. Any AI we put in a child's hands should be built to do the same.

For India, this matters more than almost anywhere else. The world as a whole needs 44 million more teachers by 2030, UNESCO estimates, and no country can train them all in time. At home, ASER 2024 found that 55 per cent of Class 5 children in government schools could not read a Class 2 text. Ministry data show nearly 10 lakh teacher posts vacant in 2024-25, and about a lakh schools running with a single teacher.

The next leap could be a tutor that never loses patience: one that speaks every Indian language, works on the cheapest phone and costs a child nothing. Gujarat already uses an AI tool that listens to children read aloud. Imagine it working in Santali or Bhojpuri, telling every teacher by evening who needs help this week.

As district collector, I often walked into government schools unannounced. My test was simple: pick a child from the back bench and ask her to read a paragraph aloud. The enrolment figures always looked better than the reading. The latest ASER district data tell the same story. In Pakur, fewer than one in five children in Classes 3 to 5 can read a Class 2 text; in Hazaribag, more than half can. I served in both districts, and I came to one firm conclusion. These children were never short of intelligence. They were short of attention: one adult's patient attention, stretched across far too many young minds.

Part of the answer is surprisingly old-fashioned. When a teacher asks a child to explain her answer aloud, no chatbot can step in. Proof of learning can come back into the classroom, through oral questions, essays written in class and projects explained in front of classmates. The good old viva, once every student's nightmare, may yet make a comeback. The University of Sydney already splits its assessments into two kinds: some are supervised, and in others AI is openly allowed.

Homework can survive if we stop pretending the machine is not in the room. Children could use AI openly and hand in the chat along with the answer. A child who can question a machine well is learning a skill the job market already rewards. And no child should ever be branded a cheat because a detector said so.

Teachers can be trained before any software is bought, as Estonia did, and AI can lighten their load. In an English trial, science teachers who used ChatGPT to prepare lessons cut their preparation time by about a third.

Some states are already showing the way. Gujarat's Vidya Samiksha Kendra keeps watch over about 1.15 crore students in more than 54,000 schools, and uses AI to flag where help is needed. Andhra Pradesh used machine learning to spot children at risk of dropping out; in Visakhapatnam district alone, it flagged 19,500 in a single year. Kerala has set out to train 80,000 teachers in AI tools. With such systems, a district education officer can see, week by week, which school is slipping in which subject, long before the annual survey arrives.

Homework was always a stand-in for practice, and the machine has simply exposed it. Somewhere in Jharkhand tonight, a girl from the back bench is staring at a sum she cannot solve. Homework will not rescue her. A patient tutor that speaks her language might. AI may have killed homework. Used wisely, it can bring learning back to life.

(Views are personal.)