FOLL-E: Teaching First Order Logic to Children

Simon Vandevelde, Joost VennekensAAAI 2023 · DOI

Introduction

While block-based programming environments like Scratch have successfully introduced algorithmic thinking ("the how") to primary education, First-Order Logic (FO)—the foundation of "the what" (requirements and properties)—remains largely absent. Logic is often perceived as having a steep learning curve, characterized by dry pen-and-paper exercises and a lack of immediate feedback.

FOLL-E (First-Order Logic Learning Environment) addresses these challenges by turning abstract logical formulas into physical puzzles. By replacing finicky syntax with wooden blocks and utilizing a formal reasoning engine, FOLL-E makes complex logic intuitive and engaging for young learners.

The FOLL-E System

FOLL-E employs a multi-faceted approach to lower the barrier to entry for formal logic:

Syntax-Free Tactile Notation

Instead of traditional symbolic strings (e.g., "For all x: Human(x) implies Mortal(x)"), children use laser-cut wooden blocks. These blocks feature a "pegs-and-slots" design that physically prevents syntax errors, ensuring that only logically valid structures can be assembled.

The FOLL-E block library, including predicates, functions, logical connectives, and quantifiers.

Model-Theoretic Puzzle Solving

The learning environment centers on a "Robot Domain." Children are presented with hand-drawn cartoon robots with various components (arms, legs, head, body) that can be colored and may possess specific attributes, such as wearing a hard hat.

An example robot model from the Robot Domain, featuring various components and attributes like color and accessories.

The core task is to distinguish "good" robots from "bad" robots by constructing a formula that captures the underlying rule. For instance, if all "good" robots have green limbs, the child must assemble blocks representing that specific property.

Real-Time Reasoning Engine

The system uses a Raspberry Pi and camera to scan physical blocks tagged with ArUco markers. Behind the scenes, the IDP-Z3 reasoning engine evaluates the assembled formula against the robot models. It provides immediate visual feedback and uses "propagation" to visualize the consequences of the child's logic on a central robot model.

Learning Outcomes and Observations

The system was evaluated with approximately 50 students ranging from ages 8 to 18.

  • Tactile Engagement: Physical blocks encouraged collaboration and focus more effectively than screen-based tools. Children were more naturally inclined to work together when manipulating physical objects.
  • Declarative vs. Imperative Thinking: Students with prior coding experience initially struggled to shift from imperative "how-to" instructions to declarative "what-is" descriptions, highlighting a key conceptual hurdle in logic education.
  • The Power of Visualization: The "propagation" feature proved crucial, allowing children to see if their formulas were too restrictive or too vague by observing the resulting robot visualization.
  • Performance: High schoolers typically mastered the curriculum within 20 minutes, while younger students (ages 8-13) required 40–50 minutes. While a post-intervention test showed lower scores for the younger group, researchers attributed this to fatigue from late-afternoon testing sessions.

Future Directions

While successful, the study identified several areas for improvement. The transition to universal and existential quantifiers remains a significant hurdle for younger children, suggesting a need for more intermediate "bridge" levels. Future iterations will focus on improving software robustness and implementing automated error explanations to help students understand why a particular logical construction fails to match the intended model.

Summary updated

BibTeX
@inproceedings{vandevelde2023foll,
  author = {Simon Vandevelde and Joost Vennekens},
  title = {{FOLL-E: Teaching First Order Logic to Children}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2023},
  doi = {10.1609/aaai.v37i13.26884},
}
Figure