Understanding how Computers Learn: AI Literacy for Elementary School Learners

University-school collaboration (HTW Berlin with educational foundation)

Berlin, Germany

Researchers and educators co-designing and facilitating elementary AI workshops

• In-school (K-12)

Hands-on workshop in school computer lab

Approx. 2-hour workshop session

Two 5th-grade classes (20 and 23 learners)

School computers using Scratch and Machine Learning for Kids

• Limited AI background among elementary teachers and few ready-to-use curricular materials.
• Learners showed large variance in prior exposure, understanding, and confidence.
• Workshop requires computer-lab access and facilitator support for differentiated pacing.

10-11 years (5th grade)

Mostly consumer exposure (e.g., Alexa, Siri, ChatGPT mentions)

No required prior programming or computer science knowledge

• Understand basic concepts and terminology around AI and machine learning.
• Describe how labeled data is used to train and test simple ML models.
• Recognize why data quality and bias affect model outcomes.

• Differentiate teaching computers via explicit programming versus ML training cycles.
• Develop age-appropriate ethical reflection on AI errors and consequences.
• Increase confidence to explore AI concepts in school contexts.

• Not a full-semester graded AI curriculum.
• Not advanced algorithm/math instruction.
• Not an assessment of long-term retention.

Beginner AI literacy tools for elementary ML and programming activities

German instructional context

• Evaluator

• Scratch activity to illustrate algorithmic instruction and sequencing.
• Machine Learning for Kids activity to train and test image classifiers.
• Lab-log documentation of predictions and misclassifications.
• Wrap-up discussion connecting model errors to data quality and bias.

• Age-appropriate framing and no high-risk personal data collection.
• Parental consent and child-sensitive research procedures (no audio/video recording).
• Guided discussion of ethical implications such as bias and safety in AI systems.

• Warm-up on where computers/AI appear in daily life.
• Algorithmic teaching analogy (e.g., sandwich instructions) and Scratch practice.
• ML classification task with butterfly/caterpillar image dataset.
• Reflection on misclassification causes, data quality, and ethical implications.

• Learners and instructors select classes, labels, and training examples.
• Model performs predictions that learners evaluate and debug.
• Humans interpret errors and decide how to improve data quality.

• Constructive alignment between learning goals, activities, and reflection.
• Short instructor inputs plus group/plenary discussion and hands-on experimentation.
• Differentiated materials for fast learners and learners needing extra support.

• Technical terms (model, data, training) needed strong age-appropriate simplification.
• Homogeneous-age groups still showed high variance in AI understanding.
• Some learners required substantial prompting in reflective discussion phases.

• Iterative workshop refinement after field test and stakeholder feedback.
• Standing discussion segments to reduce screen distraction and improve attention.
• Prepared template programs and support paths for differentiated classroom pacing.
• Misclassification-by-design examples to make bias/data issues concrete.

• Teachers reported strong participation and positive engagement during the workshop.
• Learners responded positively to hands-on experimentation and immediate feedback.

• Learners articulated core ML cycle elements: labeling, training, testing, and checking correctness.
• Students identified plausible misclassification causes such as small/biased data and ambiguous images.
• Children demonstrated age-appropriate critical awareness of data quality and fairness implications.

The concept is implementable by teachers without formal CS background when materials and facilitation supports are provided.

• Ethics content included data bias, data quality, and consequences of classification errors.
• Research with minors used consent-based, low-intrusion methods and avoided audio/video recording.
• Discussion activities connected technical outcomes to responsibility and safe real-world AI use.

• Activity documentation
• Practitioner observation
• Post assessment

• Provides a practical elementary-level AI literacy design pattern with replicable workshop phases.
• Contributes evidence that younger learners can engage with basic ML and ethics concepts.
• Supports future work on equity-oriented and teacher-deliverable early AI education.

• Elementary AI literacy pedagogy
• Hands-on machine learning education
• Child-centered ethics and bias education
• Teacher-support models for early AI curriculum

• Completed

Elementary (10-11 years)

In-school computer lab

Introductory machine learning classification and reflection

Hands-on workshop with discussion and guided experimentation

Low to Medium

Low (classroom activity datasets and non-sensitive learner artifacts)