Recommendation systems shape much of what people, including youth, encounter online, influencing their exposure to information and ideas. Understanding their workings and potential downsides, such as filter bubbles, is increasingly important. At the same time, Minecraft remains one of the most popular and accessible game platforms among students worldwide, making it a promising medium for AI literacy outreach. Building on a previous in-person augmented reality application called BeeTrap, in which players, acting as bees, pollinate flowers and see how similar choices narrow their environment, we created BeeTrap-MC, a Minecraft-based version aimed at broader reach and accessibility. Unlike the original facilitator-led group workshop, BeeTrap-MC is designed for students individual playthrough. We conducted a study with nine middle school participants, using pre-/post-assessments and qualitative interviews to evaluate its effectiveness. Results showed significant learning gains in key AI concepts, such as understanding filter bubbles and their consequences. We also discuss key differences in design, usability and outcomes between BeeTrap-MC and the original, reflecting on trade-offs in adapting a group-based embodied experience into a shorter, self-guided digital format.

Abstract The complex processes of collaborative knowledge construction require a multimodal approach to capture the interplay between learners, tools, and the environment. While existing studies have recognized the importance of considering multiple modalities, there remains a need for a comprehensive framework that explicitly models the dynamics of knowledge representation and construction. Drawing on theoretical perspectives from collaborative knowledge-building, distributed cognition, and multiple representations in science education, we propose a multimodal representation framework that captures the diverse ways in which learners externalize, negotiate, and advance their understanding. We employ Transmodal Ordered Network Analysis to examine the interplay between knowledge representations across three distinct yet interconnected spaces: the virtual space of the digital environment, the conceptual space of internal knowledge, and the physical space of gestures. This approach enables a more granular and accurate modeling of the temporal dynamics and influences associated with different modalities. Investigating 16 groups of college students (n = 77) who utilized an immersive astronomy simulation in their introductory astronomy course, results reveal distinct patterns between high- and low-learning groups. Notably, high-learning groups demonstrated more frequent and stronger cross-modal connections, linking verbal explanations with digital representations within the simulation and with embodied representations through gestures. It extends the theory of multiple representations by demonstrating its importance not only for individual learning but also for collaborative processes. The findings highlight the need for designing learning environments and analytic approaches that can support and capture the rich multimodal interactions through which students co-construct scientific understanding.

This study aims to better understand norms that are potentially relevant for student engagement in peer critique within the modeling context.By analyzing multiple data from a 5 th /6 th grade classroom, we identified social and socioscientific norms within small groups.Students demonstrated productive within-group engagement in peer critique governed by norms, while encountering some difficulty with adopting certain socioscientific norms.

Fostering AI literacy among youth from diverse communities with unique values and challenges is crucial.This requires critical pedagogy and education inclusion that empower students-especially those from marginalized backgrounds-to engage with technology meaningfully and address societal inequities.We developed a summer camp with the theme "Protect the Freedom to READ".By inviting students to collaboratively design AI for their community, the camp deepened their AI understanding and empowered them to create applications that honor their community's values.Collaborative AI design developed higherorder thinking skills.Additionally, critical pedagogy played a key role in enabling children to apply AI skills toward innovative, community-focused applications, positioning them as advocates for ethical and inclusive technology.

Teaching complex AI concepts to children presents significant challenges, as educators often struggle to both grasp nuanced mechanisms and translate them into engaging learning experiences. Drawing on embodied cognition and analogical reasoning theories, this work aims to create embodied metaphors and analogies that simplify the mechanisms and ethical issues of AI recommendation systems for middle-school students. We employed a multi-phase design pipeline grounded in the Design-Based Research (DBR) methodology to integrate feedback from key stakeholders, including domain experts, students, and teachers, for linking abstract AI concepts to concrete and familiar experiences. This process began with collaborative brainstorming with seven AI experts to generate diverse embodied representations. The resulting designs underwent mixed-method evaluations with both middle-school students and teachers, which revealed positive shifts in student attitudes and provided insights for further iteration. The design cycles ultimately uncovered key challenges, such as balancing feature fidelity with cognitive overload, and yielded design implications for developing effective embodied and analogical representations for children's AI literacy.

Understanding the inner workings of Artificial Intelligence (AI) recommendation systems may benefit children in becoming more sensible consumers of the ever-growing information in their daily lives. It may further enable deeper reflections on related ethical issues such as the filter bubble. With limited prior knowledge in math and computing, children often find AI concepts overly abstract. Inspired by optical computation, we propose a novel tangible interface, OptiDot. Through exploratory manipulation with light beams, OptiDot supports children in learning the dot product—a building block for numerous AI algorithms—and AI recommendations through embodied learning experiences. Findings of a preliminary user study with ten middle school students indicate the effectiveness of the key embodied metaphors. We also discuss the design implications and challenges of developing optical-inspired learning tools for children.

AI recommendations influence our daily decisions. The convenience of navigating personalized content goes hand-in-hand with the notorious filter bubble effect, which may decrease people's exposure to diverse options and opinions. Children are especially vulnerable to this due to their limited AI literacy and critical thinking skills. In this study, we propose a novel Augmented Reality (AR) application BeeTrap. It aims to not only raise children's awareness of filter bubbles but also empower them to mitigate this ethical issue through sense-making of AI recommendation systems' inner workings. By having children experience and break filter bubbles in a flower recommendation system, BeeTrap utilizes embodied metaphors (e.g., NEAR-FAR, ITERATION) and analogies (bee pollination) to bridge abstract AI concepts with sensory-motor experiences in familiar STEM contexts. To evaluate our design's effectiveness and accessibility for a broad range of children, we introduced BeeTrap in a four-day summer camp for middle-school students from underrepresented backgrounds in STEM. Results from pre- and post-tests and interviews show that BeeTrap developed students' technical understanding of AI recommendations, empowered them to break filter bubbles, and helped them foster new personal and societal perspectives around AI technologies.

Young learners today are constantly influenced by AI recommendations, from media choices to social connections. The resulting "filter bubble" can limit their exposure to diverse perspectives, which is especially problematic when they are not aware this manipulation is happening or why. To address the need to support youth AI literacy, we developed "BeeTrap", a mobile Augmented Reality (AR) learning game designed to enlighten young learners about the mechanisms and the ethical issue of recommendation systems. Transformative Experience model was integrated into learning activities design, focusing on making AI concepts relevant to students’ daily experiences, facilitating a new understanding of their digital world, and modeling real-life applications. Our pilot study with middle schoolers in a community-based program primarily investigated how transformative structured AI learning activities affected students’ understanding of recommendation systems and their overall conceptual, emotional, and behavioral changes toward AI.

This study aims to better understand norms that are potentially relevant for student engagement in peer critique within the modeling context.By analyzing multiple data from a 5 th /6 th grade classroom, we identified social and socioscientific norms within small groups.Students demonstrated productive within-group engagement in peer critique governed by norms, while encountering some difficulty with adopting certain socioscientific norms.

In collaborative problem solving (CPS), people's actions are interactive, interdependent, and temporal. However, it is unclear how actions temporally relate to each other and what are the temporal similarities and differences between successful vs. unsuccessful CPS processes. As such, we apply a temporal analysis approach, Multilevel Vector Autoregression (mlVAR) to investigate CPS processes. Our data were collected from college students who collaborated in triads via a video-conferencing tool (Zoom) to collaborately engage a physics learning game. Video recordings of their verbal interactions were transcribed, coded using a validated CPS framework, and organized into sequences of 10-second windows. Then, mlVAR was applied to the successful vs. unsuccessful CPS sequences to build temporal models for each. A comparison of the models together with a qualitative analysis of the transcripts revealed six temporal relationships common to both, six unique to successful level attempts, and another eight unique to unsuccessful level attempts only. Generally, for successful outcomes, people were likely to answer clarification questions with reasons and to ask for suggestions according to the current game situation, while for unsuccessful CPS level attempts, people were more likely to struggle with unclear instructions and to respond to inappropriate ideas. Overall, our results suggest that mlVAR is an effective approach for temporal analyses of CPS processes by identifying relationships that go beyond a coding and counting approach.