When the Lesson Adapts to the Child: How AI Could Change School

In a classroom, the same explanation can arrive too early for one child and too late for another. One pupil is still trying to understand the previous step. Another is ready for a harder problem. A third understands the idea but struggles with the language of the textbook.

A teacher has to respond to all of them while keeping the lesson moving.

This is where artificial intelligence offers a practical possibility: helping a lesson respond to the child. Software can select a more suitable exercise, offer another explanation or help a teacher recognise a recurring misunderstanding.

The ambition is familiar to anyone who has taught a class or helped with homework. Children need different kinds of support at different moments. AI could make that support easier to provide.

A lesson that can change course

“AI in education” covers several different technologies.

Adaptive learning software selects exercises according to a pupil’s responses. A generative chatbot creates explanations and answers. A purpose-built tutor adds teaching rules and curriculum materials to guide the conversation. A teacher’s assistant might prepare several versions of a worksheet before the lesson begins.

These differences matter because each tool performs a different educational job.

Singapore already brings several of these functions together in its national Student Learning Space. Its adaptive learning system supports mathematics at upper-primary and lower-secondary levels, alongside geography at upper-secondary level. A separate Learning Assistant guides pupils through questions, while teachers can set interaction limits and inspect conversations. Other tools help prepare activities and identify patterns in pupils’ responses. Source: Singapore Ministry of Education

The underlying idea is to make the next step more useful. A child who repeatedly struggles with fractions might return to a visual explanation. Another who has mastered the concept could move to a more demanding problem.

This can also change what a teacher sees. Instead of discovering a misunderstanding after marking an entire pile of work, the teacher can receive a clearer picture of which pupils need another explanation and where the class is ready to move forward.

What children have actually learned

Some of the strongest evidence comes from adaptive systems developed before today’s conversational chatbots.

In Delhi, a randomised study involving 619 pupils evaluated an after-school programme combining Mindspark’s adaptive exercises with small-group teaching. After four and a half months, pupils offered the programme performed better in independently assessed mathematics and Hindi.

The gains were substantial: 0.37 standard deviations in mathematics and 0.23 in Hindi, a statistical measure researchers use to compare results across assessments. The successful intervention combined personalised software with human instruction. Source: original study in the American Economic Review

A separate Indian trial offers a more targeted lesson. When pupils received different versions of the same platform, personalisation produced no significant average mathematics advantage, but initially low-performing pupils benefited. This suggests a particularly useful role for adaptation: helping children whose current understanding is below the level at which the class is working. Source: original study

Generative AI is also entering evaluated school programmes. In Nigeria’s Edo State, a six-week programme combined supervised Microsoft Copilot use with teacher-led after-school sessions. A World Bank randomised evaluation reported improved English performance and broader digital-learning outcomes.

The intervention brought together computers, additional learning time, teachers and conversational AI. Its practical value lies in that combination: pupils had a place to practise, an adult to guide them and a tool that could respond to their questions. Source: World Bank working paper

For schools and families, the most useful measure remains what a child can understand and do independently after the practice ends.

More time for the teacher

One of AI’s most immediate contributions happens before pupils enter the classroom.

An English trial involving 259 science teachers in 68 schools tested ChatGPT with a guide for lesson preparation. Teachers using the approach spent about 25 minutes less each week preparing the specified year-seven and year-eight lessons—a reduction of roughly 31%. An independent review found no detectable difference in the quality of sampled teaching resources. Source: NFER and the Education Endowment Foundation

That saving concerned a defined preparation task. Its significance becomes clear when translated into ordinary schoolwork: drafting questions, adjusting a text’s difficulty, preparing an alternative explanation or creating an additional activity for pupils who finish early.

A 2025 Gallup–Walton survey found that US teachers using AI weekly estimated average savings of 5.9 hours a week. This is a self-reported estimate, distinct from the time savings established in the English trial, but it shows why teachers are interested in these tools. Source: Gallup–Walton survey

The educational opportunity is to turn saved preparation time into more attention: a conversation with a struggling pupil, closer feedback on a piece of writing or better communication with parents.

Schools have a role in making that happen. Time saved becomes an educational benefit when teachers can use it for teaching.

The same classroom, more ways to participate

Personalisation does not have to mean every child learning alone behind a screen. It can make a shared lesson accessible in different ways.

In Uruguay, UNICEF, the national education administration and Ceibal have worked on accessible digital textbooks whose production uses AI alongside human review. Features include read-aloud support, image descriptions and simplified text. Source: UNICEF Innocenti

UNICEF’s account of fifth-grader Julieta makes the idea tangible. Her mother describes growing confidence with audiobook support. Her teacher, Magalí, describes pupils working on the same topic through accessible materials, with classmates helping one another. Source: UNICEF’s classroom account

The value is participation. A child who struggles with a printed page can approach the same lesson through listening, clearer language or a more accessible presentation.

These tools can complement specialist support for children with disabilities or learning difficulties. They can also give teachers more ways to include pupils in the work of the whole class.

Multilingual assistance offers another opportunity. A pupil learning in an unfamiliar language could receive a simpler explanation or a translation while continuing to develop the language skills needed to participate independently.

The goal is a classroom with more entry points into a shared subject.

Help that leaves the thinking to the child

Good assistance still asks the learner to think.

A Turkish high-school experiment illustrates the importance of how an AI tutor is designed. Researchers compared ordinary resources with two GPT-4-based tools.

The less restricted GPT Base improved performance during supported practice by 48%, but pupils subsequently performed 17% worse when working without AI. A guarded GPT Tutor improved practice performance by 127% and largely mitigated the later harm. Source: original study in PNAS

The contrast points to a central design choice. A tool can help a pupil obtain an answer, or it can help the pupil work through the reasoning needed to reach one.

A useful tutor might ask the child to identify the first step, explain a choice or reconsider an error. It can offer a hint and then leave space for another attempt.

This is also a useful principle for families. When a child asks for homework help, the productive question becomes: does this response help the child take the next step themselves?

Schools can reinforce that approach with short independent exercises, explanations in the pupil’s own words and unfamiliar problems that require applying the same idea.

Homework becomes more about the process

When AI can produce a polished answer, the route to that answer becomes more important.

Homework can include showing how a draft changed, explaining why a source was selected or identifying where an initial solution went wrong. A teacher can follow an AI-assisted activity with a related classroom problem or a short discussion.

These approaches make thinking more visible. They also give pupils a practical reason to understand the work they submit.

Teacher training is central to this shift. Educators need time to examine how tools respond, recognise inaccurate feedback and decide when an automated suggestion fits the lesson. Professional judgement remains part of the process from preparation to assessment.

Families benefit from equally clear expectations: which activities permit AI, what children should attempt first and when they should seek help from an adult.

Age-appropriate design, responsible handling of personal data and understandable safeguards belong within that educational approach. UNICEF’s guidance places children’s rights and accountable design at the centre of AI use. Source: UNICEF Guidance on AI and Children

What this could mean for Georgia

Georgia already has teacher-training initiatives on which further work could build.

Alte University announced an AI-in-school training programme in 2024. In 2025, the Educators and Scientists Free Trade Union reported 54 AI training sessions involving 1,890 teachers from 235 schools. These activities offer a starting point for developing practical classroom expertise. Sources: Alte University, educators’ union

Access requires attention to devices as well as connectivity. Geostat’s September 2026 release puts rural household internet access at 89%, while computer access stands at 42.8%. School-based access could therefore play an important role in making suitable tools available to children whose homes lack a computer. Source: Geostat

A practical next step would be a focused pilot: teacher-reviewed explanations for one mathematics unit, accessible Georgian reading materials or supervised language practice during school hours.

The programme would need accurate Georgian, alignment with the curriculum and support for the languages used by participating pupils. Teachers should help select and refine the material.

Success could be assessed through independent pupil work, later checks of what children retain, and the time teachers spend preparing and correcting resources. Including rural and urban schools would help reveal how the approach works under different conditions.

This would give Georgia a way to build experience around a specific educational need and expand approaches that prove useful locally.

A more attentive school

Over the next three to five years, adaptive practice, conversational tutoring and teacher assistants could become more closely connected.

A teacher might begin a lesson with a clearer picture of common mistakes. Pupils could spend a short period practising at different levels before returning to group discussion. A child could receive another explanation outside school hours, while the teacher retains a view of where further support is needed.

For families, well-designed school provision could make homework less dependent on their ability to purchase extra help. For teachers, it could make individual attention easier to organise within a busy classroom.

Achieving this will require reliable tools, teacher training, sustainable costs and regular assessment of what children actually learn.

The most valuable change may be quieter than the technology suggests: a pupil who receives the right explanation at the right moment, gains confidence and can then take the next step independently.