Interview | Rukmini Banerji on AI, ASER and Fixing India's Learning Crisis

There are many reasons why we as an education system are not able to ensure that children acquire foundational learning in the early years of primary school. The National Education Policy [NEP 2020] acknowledges some of these reasons and recommends solutions. For example, children enter standard 1 with varied background experiences. Some may have had pre-schooling, others not, yet, all are expected to navigate the same standard 1 curriculum. NEP 2020 makes a compelling case for universal, quality early childhood education.

Another challenge is that teaching in Indian classrooms is often done for the whole class, without recognising that children are at different levels. Nobel laureate Abhijit Banerjee, in his book Poor Economics, calls this practice ‘teaching to the top of the class’. NEP 2020, especially for the early grades, calls for play-and activity-based pedagogy through which participation and inclusion can be ensured.

Our school systems expect teachers to complete a fixed syllabus each year. Cohorts of children move through the system in a linear way, with curricular expectations getting progressively harder at each grade. Children who start school at a disadvantage fall further and further behind—a pattern economists Lant Pritchett and Amanda Beatty have termed the “negative consequences of an over-ambitious curriculum”. While families understand the importance of schooling, many don’t realise their child has ‘fallen behind’. All of these factors combine to hold children back. If NEP 2020’s goal of ensuring foundational learning in primary school is to be achieved, ‘catch-up’ or remediation measures will need to be built into the primary grades early on.

A lot will depend on how AI is used and by whom. Handling variation in learning levels within a classroom is the biggest challenge teachers face. Teachers’ training, however, is periodic and largely ‘one size fits all’. It doesn’t equip teachers to handle the specific composition of their own classrooms.

If AI can support teachers with differentiated instruction, that could be one way to change the status quo. Pratham’s well-tested ‘teaching at the right level’ approach has already proven effective in helping children in standard 3 and above catch up on foundational learning. We are experimenting with how AI can help teachers group children by their current learning level and support each group with appropriate activities; the goal being to improve the effectiveness of ‘catch-up’ solutions at scale.

Before developing any strategy, approach, service or product, it’s important to start by understanding the specific problem. Does a solution already exist? Can it be improved? Or is a new approach needed altogether?

First, a one-size-fits-all approach doesn’t work for ensuring every child learns. What is needed is a targeted instructional approach. Second, it’s critical to think about who the child is and the context she lives in. Take children who migrate with their families. They may need institutional arrangements suited to their circumstances, seasonal teaching-learning packages, or instruction that can travel with them. If so, what human input is required? And what kind of technology or AI can support that human effort appropriately for children on the move?

Technology can free our traditional, linear education system from its constraints. Imagine children entering middle school not only with strong foundational skills but also with real curiosity to explore the world. Rather than being confined to grade-level textbooks, digital devices, digital skills and digital resources can open infinite pathways for a child to explore subjects of their choice. The real ‘revolution’ in school education will happen when we let children discover the world around them and beyond. Access to digital devices, good-quality accessible content and the capability to use them well will help a child acquire the skill of ‘learning to learn’. Rather than thinking of AI simply as a personalised tutor, consider it a friendly guide and companion for the child’s learning journey.

There’s a tendency to frame AI as something that steps in to do what a teacher finds difficult. Current personalised AI tutors help children travel curricular pathways at their own pace, but the evidence on the effectiveness and durability of learning gained this way is still accumulating.

That said, this is a narrow view of what technology can do. We need to rethink, in fundamental ways, how children can fuel their own learning through choice and curiosity.

The potential of using AI for learning across languages is still largely undiscovered, particularly for young children. In many Indian states, English is introduced early in primary grades, but teachers are often not comfortable teaching it, and children may have little exposure to English in their environment. Given this, how can AI help teachers, children and parents not only strengthen the language they already use, but also use proficiency in a first language as a bridge to building capability in others? Can oral mastery of one language support learning another well?

No doubt there is research underway somewhere, in a lab or a tech start-up. But at scale and at the right price, such pathways are not yet available to the public, and certainly not for primary school populations.

Ten years from now, India’s foundational learning levels will be far above where they are today. National survey data from 2024 [Performance Assessment, Review and Analysis of Knowledge for Holistic Development and ASER] shows significant post-Covid learning gains, especially in the early grades. Over the same period, preschool coverage has expanded substantially and the age at which children enter formal schooling has shifted. Far fewer children are now enrolling under-age. Many of these changes align with NEP 2020’s recommendations. There is still a long way to go, but the evidence of the past few years shows that substantial change in foundational learning is possible in a short period of time. Whether AI will contribute to that change remains to be seen.