Augmented reality mask for teaching artificial intelligence and machine learning in teacher training and K-12

Introducing core concepts of Artificial Intelligence and machine learning in educational contexts requires instructional approaches that are engaging, time-efficient, and accessible to learners without prior expertise. This paper presents an augmented-reality mask activity (AR mask) that guides participants through a machine learning implementation cycle within a single lesson. In a browser-based environment, participants capture and label a small dataset of four facial expression categories and train an image-classification model using transfer learning with a pre-trained convolutional neural network. Learners then deploy the resulting neural network in an interactive, Scratch-based AR application that reacts to live webcam input by overlaying expression-dependent masks. The activity is designed to connect abstract ML concepts to embodied, immediate feedback, make the role of data quality and variability immediately tangible through iteration, and surface typical limitations of ML systems (e.g., uncertainty, fragility, and dataset bias in small samples). In this exploratory study we evaluated the approach using a mixed-methods design with 32 in-service teachers (online survey) and 40 lower-secondary students (in-class questionnaire and guided reflection). Teachers perceived the AR mask as suitable for addressing neural network training processes, computer vision, reliability and error mechanisms, and dataset-related bias, and judged the cost-benefit ratio as favorable. Student responses indicated high enjoyment and showed that learners linked model performance to the quantity and diversity of training data. Overall, the AR mask appears to be a low-threshold entry point for foundational AI literacy.