A Structured Unplugged Approach for Foundational AI Literacy in Primary Education

University-led primary education intervention

Italy (primary school context)

Researchers and educators designing and delivering AI-math integrated modules

• In-school (K-12)

Unplugged, structured multi-module classroom learning path

4 modules over 4 days (about 8 hours total)

31 fifth-grade students across two classes (23 completed final assessments)

Primarily unplugged activities, plus selected AI tool exposure (e.g., AI for Oceans)

• Primary students often have limited prior CS/programming experience.
• Conceptual abstraction and transfer across representations remain challenging for some learners.
• Small sample and single-school context limit broad generalizability.

Primary school, grade 5

Daily consumer exposure but low conceptual understanding

Minimal or no formal programming background

• Build foundational AI literacy beyond tool usage.
• Strengthen classification reasoning and data representation understanding.
• Develop awareness of AI limitations, errors, and decision-making boundaries.

• Reinforce mathematical thinking through AI-linked tasks.
• Improve argumentation and justification in classification decisions.
• Support critical, responsible interaction with AI technologies.

• Not a coding-first AI curriculum.
• Not a long-term longitudinal study of retention.
• Not a purely tool-driven exposure model.

Unplugged-first AI literacy pathway with selective digital demonstrations

Classroom-native language plus visual/symbolic mathematical representations

• Evaluator

• Students classify objects using progressively refined feature-based rules.
• Learners move across Euler-Venn diagrams, tables, and decision trees.
• Students test classification models and inspect failures (false positives/negatives).
• Selected digital activity demonstrates effects of training data and labeling quality.

• Teacher-guided framing of AI as human-engineered and fallible, not autonomous intelligence.
• Explicit discussion of misclassification risks in high-impact domains.
• Anonymized classroom data collection and structured scoring protocol.

• Module 1 introduces AI, data/information distinction, and machine limitations.
• Module 2 develops classification principles and evaluation of rule performance.
• Module 3 deepens representations via Venn, tables, and decision trees.
• Module 4 conducts post-test and satisfaction reflection for outcome analysis.

• Humans define labels, select features, and evaluate model adequacy.
• AI systems execute predefined or learned classification behavior.
• Students critically interpret outputs and identify where decisions fail.

• Constructivist/constructionist design with spiral progression of concepts.
• Learning-by-doing and learning-by-necessity tasks to provoke model refinement.
• Multiple semiotic representations to support conceptual transfer and reasoning.

• Students initially carried common misconceptions about AI agency and emotions.
• Some learners relied on procedural strategies rather than deductive reasoning.
• Abstraction in accuracy reasoning and representational transfer required additional support.

• Integrated AI concepts with existing math topics to lower entry barriers.
• Used collaborative floor-based and board-based tasks for embodied understanding.
• Reintroduced core concepts with increasing complexity via spiral pedagogy.
• Combined quantitative post-test evidence with qualitative learner reflections.

• Most students reported high enjoyment and comfort during activities.
• Interactive classification tasks were frequently cited as the most engaging.

• Students improved terminology use, feature description, and classification justification.
• Post-test patterns showed satisfactory gains in AI concept understanding and reasoning.
• Math-related skills (frequency interpretation and set reasoning) showed solid performance for most learners.

A structured unplugged path can make AI concepts accessible in primary school while simultaneously reinforcing mathematical reasoning and critical citizenship.

• Curriculum explicitly addresses AI limitations and consequences of misclassification in real systems.
• Learners are encouraged to question AI output reliability rather than accept it uncritically.
• Study procedures used anonymized student identifiers and privacy-aware evaluation protocols.

• Post assessment
• Practitioner observation
• Activity documentation

• Provides a replicable unplugged AI literacy sequence aligned with primary curricula.
• Demonstrates practical integration of AI literacy and foundational mathematics.
• Offers design evidence for balancing conceptual depth, accessibility, and engagement.

• Primary AI literacy curriculum design
• Unplugged AI pedagogy
• AI-mathematics interdisciplinary learning
• Early critical AI education

• Completed

Primary (grade 5)

Formal classroom (unplugged-first)

Classification reasoning and model evaluation fundamentals

Structured spiral learning with constructivist hands-on activities

Low to Medium

Low (classroom tasks and anonymized assessment artifacts)