Parse: Teaching How AI Learns Through Tangible Unplugged Modules

Parse: Unplugged AI Learning Modules

We present Parse, a set of three unplugged tangible modules designed for middle school students that externalize core computational mechanisms underlying how learning systems operate including search, constraint-based refinement, and statistical inference from feedback into hands-on interaction.

Explore Modules Read Full Paper

Unplugged AI Learning

No screens, internet access, or technical infrastructure required. Optimized for low-resource environments.

Tangible & Embodied

Physical interaction with manipulable objects supports conceptual grounding of abstract STEM and ML processes.

Feedback-Derived

Surfaces pattern discovery through random search, constraint reduction, and model updating.

The Three Modules

Each module isolates a distinct problem in ML: where to search (Module 1), how to narrow (Module 2), and how to build a model (Module 3).

Download Workshop Program
01 / MODULE ONE

Random Search

Random Exploration
Core ML Concept

Exhaustive exploration of a solution space in the absence of proximity indicators (brute-force or grid search).

The module contains two wheels, each with eight orientations (7 walls, 1 passage). The learner rotates both wheels, locks them, and drops a ball. Only one combination out of 64 permits passage. Without a proximity signal, the learner must explore exhaustively.

  • 64 unique configurations
  • Exhaustive search strategy
  • No proximity feedback
  • Rotatable physical wheels
Module 1: Random Search
Download STL File
02 / MODULE TWO

Iterative Refinement

Constraint Propagation
Core ML Concept

Constraint satisfaction and version space reduction. Progressively narrowing hypotheses based on structured matches.

Features six unique cylindrical profiles and six matching keys. Three profiles are enclosed. A correct fit gives immediate auditory click feedback. Each successful match decreases the remaining options, modeling constraint propagation.

  • Cylindrical profile tubes
  • Iterative narrowing
  • Immediate click feedback
  • Hypothesis reduction
Module 2: Iterative Refinement
Download STL File
03 / MODULE THREE

Strategic Inference

Statistical Inference
Core ML Concept

Model building through observation aggregation. Estimating hidden distributions over time to optimize prediction.

Consists of seven sliding keys with hole configurations and a grid holding three hidden balls. Resolving coordinates across trials builds a probability map of the hidden layout, mimicking statistical model estimation.

  • Sliding keys grid
  • Probability mapping
  • Delayed distributed feedback
  • Hidden distribution mapping
Module 3: Strategic Inference
Download STL File
The Team

About Us

Meet the designers and researchers from Hisar School who developed Parse.

Research & FabLab Development

Parse was designed and tested in classrooms through five iterative Research through Design cohorts involving 120 middle school students. The project shows that unplugged physical modules can make complex machine learning concepts accessible through immediate, tactile interactions.

Ediz Umur

Co-Author

Researcher and programmer at ideaLab FabLab, Hisar School, Turkiye.

Email ORCID

Irmak Ureten

Co-Author

Researcher and designer at ideaLab FabLab, Hisar School, Turkiye.

Email ORCID

Kaan Koca

Co-Author

Researcher at ideaLab FabLab, Hisar School, Turkiye.

Email ORCID

Sedat Yalcin

Project Advisor

Advisor and FabLab Coordinator at ideaLab FabLab, Hisar School, Turkiye.

Email ORCID

ideaLab FabLab Hisar School

Istanbul, Turkiye