How to use AI for learning without handing over the teaching
Use AI as a planning assistant: start with a specific skill, review the evidence and activity, protect learner information, and check the result.
Let AI help narrow the teaching options while an adult checks the fit, chooses the activity, and reviews what happened.
AI can be useful when it helps an adult move from a specific observation to a manageable teaching choice. “This learner needs help with math” is too broad. “The learner can build equal groups but subtracts one object when using tens to solve a nine-times fact” gives a planning assistant a much clearer job.
The important design decision is where the judgment sits. In Glow's M.I.R.E.R. approach, the adult reviews the suggestion and chooses what to do. The learner's experience does not depend on an open-ended conversation with a chatbot.
Begin with a teachable observation
Look for a pattern across comparable attempts. Separate a checked answer from a learner's self-report and an independent attempt from one completed after help. Add relevant context you observed yourself.
Do not translate a pause into a claim that a child is distracted or a missed item into a diagnosis. A planning tool works better when the evidence is described precisely and its interpretation stays open to adult judgment.
Ask for a bounded next step
A useful planning task is choosing between a small number of appropriate strategies: model first, explain first, or revisit a prerequisite. A broad request to invent an entire curriculum from a few answers gives you far more to inspect.
With permission, Claude can assist M.I.R.E.R. by selecting from Glow's teaching strategies. This keeps the suggestion connected to authored activities and a known learning goal. The adult still checks whether the approach fits the learner and the lesson sequence.
Review before the learner uses it
Check the mathematics, spelling demands, instructions, and required materials. Ask whether the activity addresses the observed difficulty or merely shares the topic. Make sure any “independent check” is actually possible without an answer being visible.
For a reader, inspect every word the child is expected to decode. For math, work through the model and answer yourself. A polished explanation can still contain a mismatch between the intended skill and the task.
Close the loop
After using the activity, observe a fresh response and plan a later return. If the learner still needs help, change the support or revisit a prerequisite. Do not keep assigning a recommendation just because it came from AI.
UNESCO's guidance places human agency at the center of educational AI. In daily practice, that means adults remain able to understand, adjust, and decline a suggestion. The value of the assistant is a clearer next teaching step, followed by evidence that helps you decide whether it was useful.
Sources & further reading
Research background and Glow’s approach, alongside the original activity ideas in this guide.
