
This study investigates recursive cycles of knowledge building-modeling (KBM) supported by knowledge building analytics to foster reflection, model improvement, and systems thinking. A total of 43 grade 5 students in China constructed models of Earth Science systems that served as objects of inquiry in Knowledge Forum. Models were refined as students engaged in knowledge building discourse, using knowledge building analytics to visualize the interconnectedness of ideas and progressively enhance models that link Earth Science phenomena to sustainability. Successive model iterations showed increased modeling sophistication and progression of systems thinking from individual variables to cause–effect relationships, secondary effects, temporal accounts, and predictions of behaviors. Students also demonstrated significant gains in their domain knowledge of Earth Science. Qualitative analyses identified mediating processes and system-level dynamics: (1) idea generation and improvement through collaborative knowledge building discourse, (2) analytics used for meta-reflection and conceptual reorganization, and (3) model construction for system expansion and refinement. These findings illustrate how students appropriated KBM and analytics as epistemic frameworks to engage in recursive design and refinement for systems thinking and knowledge advances. The research demonstrates how computer-supported collaborative learning (CSCL) enriched with model-based inquiry and knowledge building analytics can foster students’ engagement with complex systems.
Teachers’ ability to assess pupils’ knowledge and skills is a crucial component of their daily responsibilities, as accurate diagnoses influence both the support provided to pupils and their learning outcomes. Traditionally, diagnosing pupils’ skills is an individual task, with little consideration given to the potential of collaborative diagnosis. However, collaboration may offer significant benefits, particularly when diagnosing cross-domain skills, such as scientific reasoning, where teachers from different subjects can contribute observations from different classroom situations and integrate their perspectives into a joint diagnosis. Drawing on parallels with the medical field, where collaborative diagnoses have already been shown to improve diagnostic accuracy, potential advantages may also occur in an educational context. Nonetheless, collaboration also comes with difficulties: Distributed information is seldom shared, and collaboration can be ineffective. Thus, using video-based simulations, this study investigates how collaboration, distribution of information, and collaboration support, by means of collaboration scripts, influence diagnostic accuracy – potentially via the application of collaborative diagnostic activities – consequently contributing to the development of diagnostic competences in pre-service teachers. The results indicate that pre-service teachers’ collaborative diagnostic accuracy does not per se exceed individual diagnostic accuracy, but that collaboration support can significantly enhance the added value of collaborative diagnosis. The reason for this seems to be increased information sharing. Thus, this study suggests that successful collaborative diagnoses require collaboration support and that collaboration scripts can help overcome the challenges of collaboration. Future research should explore collaborative diagnosis in different contexts and for other (cross-domain) skills.
In collaborative learning, regulation is crucial for meaningful and effective collective learning. As generative artificial intelligence (AI) becomes readily available to learners as an external, on-demand information resource, regulatory behavior dynamics may shift from information seeking toward evaluating, selecting, integrating, and coordinating externally generated content within group discourse. This study examined the learners’ behavioral dynamics in their cognitive, task, and emotional regulation and investigated the interplay of individual performance and behavior transition patterns against the backdrop of group achievement in an AI-supported collaborative learning environment. A total of 126 undergraduate and graduate students participated in a face-to-face learning task, with access to online AI tools at the learners’ discretion. Audio recordings were transcribed into 16,816 semantic units for content analysis. Using k-means clustering, the research identified four distinct learner types, namely evaluative, curious, expressive, and passive. Lag sequential analysis (LSA) results showed divergent behavioral patterns emerged between high- and low-performance groups and individuals. Group performance level did not warrant similar individual performance levels. The study highlights the complex interplay between individual and group behaviors and their impact on performance, in particular the behaviors of high-performance individuals in low-performing groups and vice versa. Notably, these patterns underscore how group outcomes can hinge on how contributions are regulated and coordinated, rather than on information availability alone. Collectively, the findings suggest that, in AI-supported collaboration, performance is associated with the learner’s capacity to critically appraise and purposely appropriate information among the group members and to align its use with task goals and social emotional coordination. This research offers insights for the facilitators on grouping strategies, differentiated scaffold design for learner types, and intervention considerations for low-performance groups and individuals. Limitations and future research directions are also discussed.