Briteller: Shining a Light on AI Recommendations for Children

TL;DR

Briteller introduces a light-based tangible interface to demystify AI recommendation systems for children, using data vectors encoded as light beams and the dot product as a core computation. Through two iterative studies, initial tangible Briteller and a tablet AR-enhanced version, the authors demonstrate learning gains, reveal affordances and limitations of light-based learning, and derive design implications for embodied AI literacy. The work shows that light-based metaphors, combined with AR augmentation, can scaffold understanding of user/item vectors, the dot product, and AI outputs while engaging diverse learners, albeit with challenges in quantitative transfer and scalability. Overall, the study advances accessible AI education by linking image schemas, data physicalization, and XAI concepts in classroom-like settings, offering design guidance for future graspable AI tools.

Abstract

Understanding how AI recommendations work can help the younger generation become more informed and critical consumers of the vast amount of information they encounter daily. However, young learners with limited math and computing knowledge often find AI concepts too abstract. To address this, we developed Briteller, a light-based recommendation system that makes learning tangible. By exploring and manipulating light beams, Briteller enables children to understand an AI recommender system's core algorithmic building block, the dot product, through hands-on interactions. Initial evaluations with ten middle school students demonstrated the effectiveness of this approach, using embodied metaphors, such as "merging light" to represent addition. To overcome the limitations of the physical optical setup, we further explored how AR could embody multiplication, expand data vectors with more attributes, and enhance contextual understanding. Our findings provide valuable insights for designing embodied and tangible learning experiences that make AI concepts more accessible to young learners.

Figures (4)

• Figure 1: Briteller examples for demonstrating (1) the dot product between a user vector u = (1, 1, 1) and an item vector v = (1, 1, 1) for the content-based recommendation system; (2) the dot product between a user vector u = (1, 1, 0) and an item vector v = (0.1, 0, 1); (3) placing a red filter in front of the third attribute of the user vector, the final light dot turns red because the second attribute has a higher contribution to the final output; placing it to the second attribute, the color of the final light dot doesn't change much because the second attribute has a relatively lower impact. Please note that the blue color in the figure is for figure demonstration. We used white light in the study.
• Figure 2: Embodied representations and interactions in Briteller.
• Figure 3: Evaluation of Briteller with middle school students (N=10): (1) Students tagged data attribute labels to corresponding parts of the physical data bars (PART-WHOLE); (2) students rotated the knobs to block a specific amount of light for setting the value of a specific data attribute (BLOCKAGE); (3) students observed how individual light beams changed while passing through the light-based recommendation system (LINKAGE); (4) a student unmounted and mounted the convex lens to add three light beams together (MERGING); (5) students rotated the knobs to block a certain amount of light (BLOCKAGE) and compared the output of light intensity (DARK-BRIGHT) under different conditions; (6) a student got curious about the optical principles applied in Briteller.
• Figure 4: In Study 2, the tangible interface of Briteller is augmented by virtual objects in AR: (1) the item visibility (DARK-BRIGHT, SUPERIMPOSITION) represents how likely the item should be shown/recommended to the target user; the user sees the item more clearly when the recommendation system predicts a higher user preference; (2) the data vectors and dot product in this light-based recommendation system is augmented by numeric values of individual data attributes; (3) a learner added a new flashlight to use describe users and items with a new pair of data attributes; the learner edited the attribute's name "fiber" and set the value for it.