From answer machine to thinking partner
AI becomes a thinking partner when it asks questions and gives hints instead of finishing the work. The child still does the reasoning, and teachers can see and override what the AI does.
The problem with instant answers
Generative AI is very good at giving answers. For a child who is learning, that can be a problem, because working through a question is often where the understanding comes from. The homework gets done, but the lesson does not.
Changing what the AI is asked to do
A better approach is to change the job the AI has. Instead of finishing the work, it can ask a question back, offer a hint, or ask the child to explain their reasoning. The child still does the thinking, with support when they get stuck.
Success is then measured differently too. What matters is the quality of the child's steps and explanations, not only whether the final answer is right.
Keeping adults in the picture
Teachers should be able to see what the AI is doing and override it. UNESCO's guidance on generative AI in education gives a useful standard here, including an age limit of 13 for independent use.
This is the idea behind the framework we are developing at IcarusX: treating generative AI as a partner in cognitive development rather than an answer machine.