Abstract:

Introduction:
Artificial intelligence (AI) is rapidly becoming a foundational technology across science, industry, and everyday life. As AI systems increasingly influence how information is processed and decisions are made, basic AI literacy is becoming an important component of modern education. However, existing educational tools present machine learning as a “black box,” allowing students to train models but providing no insight into how those models organize and interpret data. AISoundLab was developed to address this gap by making the internal representations of machine learning systems visible and interactive. Using sound as an intuitive medium, the platform allows users to record, label, and explore sounds while observing how a neural network organizes them into clusters based on similarity, and how the network learns to predict the identity of new sounds. This paper describes the design of AISoundLab and its use as a tool to support understanding of core artificial intelligence concepts.

Methodology:
AISoundLab is a website developed using Claude Code for rapid prototyping and development. Users record sounds through the microphone and assign labels to them (e.g., by recording and typing words). The sounds appear as colored dots on the screen, with each label corresponding to a different color. The system then trains a neural network to group sounds with the same labels, forming clusters of similar sounds. When a new sound is recorded, the model predicts its identity by comparing how close it is to these learned clusters. Both the location of each sound and the training process are visualized in real time, allowing users to see how the system gradually organizes the sounds as it learns.

Results:
Several design features support learning. First, real-time movement of sound-dots allows users to observe training as clusters form and reorganize. Second, the system emphasizes representation learning rather than just prediction accuracy, helping students understand the similarity relationships within the data. Third, rapid recording and feedback encourage experimentation with everyday sounds. Through these interactions, users can develop an intuitive understanding of basic AI concepts such as training, inference, probabilistic decision-making, and how biased or unbalanced data can influence a model's behavior.

Conclusions:
AISoundLab illustrates how interactive visualization can support AI literacy. By showing how sounds are organized in a learned space and how new sounds are classified relative to existing clusters, the system turns machine learning into an observable process rather than a hidden one. Because users directly record, label, and retrain the model, they can see how their actions change the system’s behavior. This helps learners develop a clearer understanding of how AI systems learn from data and make decisions.

Keywords:

AI, STEM, Education, Technology.