AI beside the teacher, not in place of the teacher

Debate about AI in education usually starts with the wrong question: “will it replace the teacher?” In early childhood the answer is not technical but principled. A three-year-old's development advances in the presence of an adult they are attached to. No system can take that on. The right question is: which tasks that steal the teacher's time can be delegated?

What can be delegated

A significant share of a teacher's day goes to non-pedagogical work: compiling scattered notes, formatting reports, drafting plans, writing the same information in three places. These are tasks AI is genuinely good at. Turning a term's accumulated observations into a readable draft gives a teacher their evenings back.

What cannot

• Judgement about a child: deciding what a behaviour means is a pedagogical interpretation, and it belongs to the teacher.
• Relationship: only someone who knows a child notices why they are quiet today.
• Communication with families: predicting how a sentence will land is human work.
• Decision: which child to work with, on what, and when, is the teacher's call — informed by data.

In practice this line is called human approval. No AI output should reach a child or family without a person approving it. That is not a slowdown; it is the system's most important safety mechanism — the point where a flawed interpretation stops before it spreads.

The second principle is transparency. Teachers and parents must know where the text they are reading came from. An unlabelled AI text reads like something a person wrote, and goes unquestioned — which is exactly where trust erodes invisibly.

AI's legitimate role in early childhood is modest, and that is precisely what makes it valuable: save the teacher time, make patterns visible, prepare drafts. Leaving decision and relationship with the teacher is not a technical limitation but a pedagogical choice.