Abstract
Inspired by explanations of machine learning concepts in children’s books, we developed an approach to introduce supervised, unsupervised, and reinforcement learning using a block-based programming language in combination with the benefits of educational robotics. Instead of using blocks as high-end APIs to access AI cloud services or to reproduce the machine learning algorithms, we use them as a means to put the student “in the algorithm’s shoes.” We adapt the training of neural networks, Q-learning, and k-means algorithms to a design and format suitable for children and equip the students with hands-on tools for playful experimentation. The children learn about direct supervision by modifying the weights in the neural networks and immediately observing the effects on the simulated robot. Following the ideas of constructionism, they experience how the algorithms and underlying machine learning concepts work in practice. We conducted and evaluated this approach with students in primary, middle, and high school. All the age groups perceived the topics to be very easy to moderately hard to grasp. Younger students experienced direct supervision as challenging, whereas they found Q-learning and k-means algorithms much more accessible. Most high-school students could cope with all the topics without particular difficulties.
Introduction
Most AI education tools for children focus on high-level applications—like training a model to recognize a face or a voice—where the actual "learning" happens behind a curtain of complex code. This creates a "black box" effect: students see the input and the output but fail to form a mental model of the underlying process. This paper, presented at AAAI-21, introduces a novel pedagogical approach to demystifying machine learning (ML) for K-12 students by moving away from "black-box" APIs and toward interactive, block-based "playgrounds."
Methods: Playful Learning Environments
The researchers developed extensions for the Open Roberta Lab, a popular block-based programming platform for educational robotics. These tools allow students to "step into the algorithm's shoes" by manually tuning neural network weights and configuring reinforcement learning environments.
Supervised Learning: Neural Network Playground
Instead of an automated backpropagation algorithm, students perform "direct supervision." They manually adjust the weights (numerical values) on links between sensor input neurons (e.g., light or ultrasonic sensors) and motor output neurons. This provides immediate visual feedback: as the student tweaks a weight, the simulated robot’s behavior changes in real-time.
Reinforcement Learning: Q-learning Playground
Students create a labyrinth for a robot and configure the parameters of a Q-learning algorithm. To make it accessible, technical terms were mapped to intuitive concepts. For example, the learning rate (Alpha) became "learn speed," and the exploration rate (Rho) became "teleportation frequency."
Unsupervised Learning: Unplugged k-means
To teach clustering, the authors designed a physical activity where a facilitator groups objects (like different drinking vessels) based on hidden criteria, and students must guess the sorting principle, effectively acting out the k-means algorithm.
User Study and Key Results
The framework was evaluated through a user study involving 24 participants across primary (Grades 3-4), middle (Grades 5-6), and high school (Grades 7-9).
- High Engagement: 75% of all participants expressed a desire to continue working with machine learning after the workshop.
- Intuition vs. Complexity: Surprisingly, younger students (Grades 3-4) found the concepts of Reinforcement Learning and Unsupervised Learning easier to grasp than the manual weight-tuning required for Neural Networks. Supervised learning was rated as the most difficult by primary schoolers (avg. score 3.3/5 for ease).
- Conceptual Shift: Qualitative feedback indicated that students moved from viewing AI as "magic" to seeing it as a system that "improves by checking which way is the better one."
Conclusion
By forcing students to manually adjust weights and parameters, the "magic" of AI is replaced by the realization that learning is essentially a process of numerical optimization. While the study was small and limited to 2D simulations, it demonstrates that technical ML concepts can be made tangible for young learners through constructionist robotics. Future work will focus on migrating these tools to physical hardware like LEGO EV3 and conducting larger-scale pedagogical evaluations.
Summary updated