How to Teach Primary Science | Representation, Experiments, Models & AI

Teaching Science is not a competition to use the most worksheets, experiments, videos, animations or AI tools. Every representation should have a job.

Use the representation that makes the missing relationship visible. Then change the representation to test whether the learner actually owns the idea.

This is the Representation & Teaching Tools specialist room in the Primary Science building. It does not replace the Tutor Role pages. Its job is narrower: decide which form of the Science should be shown next, and why.

Science must move between the world and representations

World → observation → object/model → image/video → diagram → table/graph → scientific language → explanation.

A strong learner can move in both directions. They can look at the world and construct a useful representation, then look at a representation and reconstruct the scientific relationship behind it.

If understanding exists only in one form, it is fragile.

Representation is not decoration

Representation	Best used when	What it can reveal
Real object / direct observation	The learner needs contact with the phenomenon.	Features, changes, sequence, comparison.
Experiment	A causal relationship needs to be tested.	Variables, evidence, fair comparison, outcome.
Physical model	Structure or spatial relationship is difficult to imagine.	Parts, positions, movement, system relationships.
Photograph / video	The event is difficult to bring into the classroom.	Observation and sequence.
Animation / simulation	A process is too fast, slow, small, large or invisible to observe directly.	Proposed mechanism and sequence.
Diagram	Irrelevant surface detail needs to be removed.	Structure, pathway, relationship.
Table / graph	Patterns across measurements matter.	Trend, comparison, relationship between variables.
Scientific language	The learner must compress a model into a precise explanation.	Whether the relationship can be expressed clearly.
AI-generated variation	The learner needs new examples, counterexamples or changed representations.	Transfer and hidden dependency—if the output is checked.

Choose the tool from the failure

Start with the learner state, not the technology.

Failure	Useful next representation	Why
Cannot notice the relevant feature.	Real object / photograph with guided observation.	Returns attention to evidence.
Knows labels but cannot connect parts.	Diagram or physical model.	Makes structure visible.
Cannot understand a process over time.	Sequence cards, animation or repeated snapshots.	Exposes order and transition.
Cannot interpret an experiment.	Variable map + table of observations.	Separates changed condition, measurement and evidence.
Understands orally but writes vaguely.	Short causal sentence frame.	Converts model into language.
Only succeeds on familiar worksheet diagrams.	Change orientation/context/representation.	Tests transfer rather than recognition.
Gets answer only after tutor explanation.	Remove the explanation and give a changed question.	Tests whether support became capability.

Tool choice should follow diagnosis.

Experiments: evidence before entertainment

An experiment is useful when it changes what the learner can infer from evidence. It is not automatically better than a diagram simply because it is hands-on.

A strong experiment lesson exposes this chain:

Question → changed condition → measured/observed response → pattern → scientific explanation → conclusion.

If the child remembers the activity but cannot explain what was changed, observed or concluded, the experiment produced experience without enough scientific reconstruction.

Models: useful because they are incomplete

A model removes some details so an important relationship becomes easier to see. That is its strength—but also its danger.

• A digestive-system diagram is not a human body.
• A circuit diagram is not a photograph of the circuit.
• A food web is not the whole ecosystem.
• An animation is a constructed explanation, not the phenomenon itself.

Students should therefore learn both what the representation shows and what it leaves out.

Animation and video: slow down what the eye cannot hold

Animation can be useful when a process is invisible or difficult to observe directly. It can compress long time periods, enlarge tiny processes, remove distracting details and show a proposed sequence repeatedly.

But watching is not evidence of learning. After the animation:

1. Remove the animation.
2. Ask the learner to reconstruct the process.
3. Change the representation.
4. Ask for a prediction or explanation.
5. Compare the answer with new evidence.

If understanding disappears when the video stops, the representation was carrying the learner.

Do not teach to fixed “learning styles”

The goal is not to label one child “visual”, another “auditory” and another “kinesthetic”. Science itself arrives in multiple forms. A learner ultimately needs to handle diagrams, words, experiments, tables, objects and unfamiliar questions.

Vary representation because the Science requires transfer—not because the learner has been assigned a permanent sensory type.

AI in Primary Science: generator, challenger and translator—not authority

AI can be useful in Science learning when its job is bounded. It can rapidly generate alternate questions, simpler explanations, counterexamples, comparison tables, oral quizzes and changed contexts.

Useful AI job	Example	Required control
Change the surface form	Create three different contexts using the same scientific relationship.	Tutor checks the underlying model.
Generate counterexamples	Produce examples that test a classification rule.	Verify factual accuracy.
Translate complexity	Explain the same mechanism at P3 and P6 levels.	Keep within syllabus boundary.
Quiz retrieval	Ask short questions without chapter labels.	Do not reveal answers too early.
Challenge an answer	Ask “What evidence supports that conclusion?”	Require the learner to justify, not copy.
Create transfer variants	Change diagram, organism, apparatus or direction of question.	Preserve the same scientific dependency.

AI can also produce incorrect or over-level explanations. Therefore:

AI output → check against syllabus/evidence → use as representation → learner reconstructs → tutor retests.

The learner should not outsource the reasoning step to the AI. If the machine supplies the model, evidence and final wording before the child attempts the problem, the apparent performance may be borrowed.

The representation-switch test

A useful way to test understanding is to preserve the relationship while changing its form.

1. Teach or repair the idea in one representation.
2. Remove the original representation.
3. Present the same relationship as a diagram, table, verbal scenario or experiment.
4. Ask the learner to identify the same mechanism.
5. Require an explanation without the original prompt.

If the relationship survives the representation change, it is becoming transferable.

Three students make representation failure visible

Give three students the same scientific relationship in different forms. One may understand the real experiment but fail the diagram. Another may read the table but not explain the mechanism. A third may write well but misread the evidence.

That comparison tells the tutor whether to repair the concept, the representation, the evidence selection or the language conversion.

A teaching-tool control loop

Observe failure → choose representation → expose relationship → learner reconstructs → change representation → retest → fade the tool.

The final step matters. A successful tool should eventually become unnecessary for that task.

Where this specialist room connects

• How Primary Science Works | Observe → Model → Explain → Test → Transfer
• Primary Science Learning Architecture
• P4 Scientific Inquiry, Variables & Structured Reasoning
• PSLE Science Language Specialist
• Primary Science OS | Technical Master Map

Specialist-room rule

This room survives because representation choice is a distinct teaching function. It does not claim that more technology is better. It asks a harder question:

What must become visible next—and can the learner still reconstruct it after the tool is removed?