Behavioral-pattern exploration and development of an instructional tool for young children to learn AI

University department (technology application and HR development) with MIT collaborators

Taiwan and USA (MIT)

Researchers designing materials and analyzing behavioral traces

• In-school (K–12)

9-week case study with staged individual then cooperative activities

9 weeks total (5 individual + 4 cooperative)

Elementary upper-grade cohort (pairs in robotics stage)

MIT App Inventor, Personal Image Classifier, robot cars, board game integration

• Interdisciplinary design complexity for teachers
• Hardware logistics for robot construction
• Analytics interpretation requires researcher expertise
• Generalization beyond single context not established

Elementary upper grades (paper: high-grade elementary)

Typical school ICT exposure; not expert programmers

Introductory App Inventor scaffolding in phase 1

• Teach applied AI via image classification students author themselves
• Integrate CT, STEM, and cooperative making
• Identify meaningful sequential learning behaviors via analytics

• Respond to national AI-in-education strategy with feasible classroom materials
• Support portfolio-based evidence of learning

• Not a large-scale RCT
• Not focused on generative LLMs
• Not a cross-country replication suite

MIT App Inventor + Personal Image Classifier + physical robot + board game

• Co-creator
• Automation tool

Instructional context in Taiwan compulsory education setting

• Students label/collect training images
• Deploy classifier to interact with tangible robot/game

• Student-generated images need privacy and consent norms
• Discuss misclassification and dataset bias with young learners
• Safe robotics construction and classroom management

• Phase 1: individual App Inventor + classifier skills
• Phase 2: pairs build robot car
• Integrate classifier app with robot and CT board game
• Collect and mine sequential interaction data

• Students own training data and model behavior for their app
• Teachers facilitate pacing and safety

• Progressive shift from individual mastery to collaborative integration
• Concrete game board coupling symbolic CT with ML outputs

• Interdisciplinary AI curriculum design is demanding for schools
• Sequential analytics surface non-obvious learning pathways
• Hardware+software integration increases failure modes needing support

• Custom modular teaching aids alongside digital toolchain
• Two-stage design isolates programming/classifier skills before robotics

• Learning analytics reveal meaningful behavioral patterns during AI application learning

• Demonstrates feasibility of integrated AI+robot+game instructional tool in case context

Positions analytics as evidence for refining materials while noting case-study limits.

• Protect images of people and classmates in personal classifiers
• Secure storage of student-created apps and datasets
• Equitable access to robotics kits
• Transparent parental information for multi-week data collection

• Post assessment
• Activity documentation
• Learning analytics

• Scale with larger samples and standardized knowledge assessments
• Compare analytics-derived patterns to teacher rubric scores

• K-12 AI instructional design
• Learning analytics
• Educational robotics

• Completed

Upper elementary

In-school

Applied image classification + robotics

Individual then cooperative PBL

Medium

Medium–High (images, logs)