Epistemic foundations for AI literacy · a methodology inspired by Montessori education
Judgement first. Then AI.
Sense & Think is a developing methodology for ages 6-12. It first makes questions, predictions and evidence tangible through physical, screen-free work. AI becomes an organising concept later, followed by teacher-mediated work with real systems.
An answer is not evidence.
When a system sounds convincing, a child needs more than instructions for operating it. They need a way to remain the author of their judgement.
Education need not stop at the correct answer. It can show how to reach it and when not to use it.
The cost of a question.
An answer is now almost instant and cheap. In the first activities, a question therefore costs something for a while.
A child has a limited number of tokens. Before revealing evidence, they consider what an answer could really tell them and commit to both predictions.
The token is a temporary scaffold. We later remove it and look for whether planning an informative question remains.How it works
Five verbs. One judgement.
A small map of questions accompanies a child from first observation to a decision with real consequences.
Investigatecore 6-9 · current 8-9
COLLECT
What did it have available? What is missing?
A child notices which examples a collection shows and which it leaves outside the picture.
↘Investigate: core ages 6-9, current core 8-9. Act: ages 10-12. Five verbs are a practical map, not a complete taxonomy.
What is missing from the collection?
Five similar examples can create a very convincing and very narrow picture of the world.
The collection we saw
All dogs are small and brown.
Which missing example best shows that the collection was too narrow?
Illustrative photo cards created to explain the COLLECT principle digitally.
What does a weak question cost?
Three tokens temporarily slow the question so its information value becomes visible. Evidence is revealed only after both predictions are committed.
A digital explanation of the closed DISTINGUISH activity. It is not an AI simulation or evidence of impact.
What we already know
Two possible rules
Choose a test piece
What does a weak question cost?
Commit before the evidence
A + B
In this closed task, A or B is true. In a real system, both possibilities may be incomplete.
A citation is not yet support.
A claim, its origin and independent evidence are three different things. Trust grows when we can separate them.
Less than what? Build a chain from the claim to independent support.
First, state exactly what is being claimed.
AI is not only a chatbot.
Some systems answer. Others classify, select or assess without asking the child anything at all.
Creates content
Produces new text, images or sound.
A chatbot or generated imageYou see the output directly in front of you.
When a system acts without a direct question from the user, its error may only become visible later.
Who signs off on the result?
For ages 10-12, this is a reflective map over a concrete task, not a separate physical material. The child considers data, tool suitability, checking and responsibility.
PROTECT
Which data is really needed to suggest a route?
DECIDE
Who or what should hold each step?
Who is responsible for the final plan?
Same result. Different conclusion.
A closed task isolates one piece of reasoning. In real inquiry, even the two hypotheses we started with may be incomplete.
We know that exactly A or B is true.
If both possibilities expect the same result, the question cannot distinguish them.
One contradictory result can eliminate a rule.
Which test do A and B predict differently?
Ages 6-12. Judgement grows with the child.
A physical scaffold first makes thinking visible. It later recedes as the same questions transfer to real systems.
Preparation
This layer belongs to the target architecture. Its detailed design remains open and will need a separate developmental review.
Which symbols and habits can prepare later inquiry without pretending this layer is complete?
Current core
Two hypotheses, a closed world, a distinguishing test, commitment to both predictions before reveal and revision of the conclusion in light of evidence.
Which test do A and B predict differently?
Name and compare
The child names AI and compares it with the earlier closed task. The boundary of the analogy is made explicit. The physical material is not a model of AI.
How is the simple task like a real system, and how is it different?
Investigate and decide
A teacher mediates work with real systems. PROTECT, DECIDE, repetition and human responsibility are added.
What changes when we use an output for a real decision?
Learning sequence
- EXPERIENCE
- COMMITMENT
- EVIDENCE
- BRIEF NAMING
- CONTRAST
- TRANSFER
Claims follow evidence.
Sense & Think separates what has been designed, what is being tested and what data already support. We will not present a mechanism as a confirmed effect if it does not survive testing or transfer without its original scaffolds.
If the reasoning disappears with the token, layout and symbols, the principle has not transferred yet.View the methodology
Help change how children learn to trust.
We are looking for people and organisations that want to develop children's judgement before systems begin making decisions about them.
Schools
Create room for short physical activities and practical discussion about evidence.
Involve a schoolPartners
Connect pedagogy, child development, safety, privacy or AI subject expertise.
Connect expertiseLet us begin with the right question.
Write if you want to open a pilot partnership, review the methodology, support development or bring the topic to a conference.
By submitting the form, you voluntarily provide the information needed to reply and consider a potential collaboration. Privacy