BACKGROUND:Self-explaining is an engagement activity that supports learners' active use of instructional scaffolds such as worked examples. Self-explanation quality is assumed to mediate the worked example effect on learning outcomes. However, prior investigations have relied on analytical approaches limited in scope and quality or have yielded inconclusive results AIMS: We replicate a previous investigation that examines whether self-explanation quality mediates the worked example effect in a worked examples - problem-solving paradigm, while simultaneously considering key cognitive aptitudes (prior knowledge, working memory capacity (WMC), fluid intelligence, shifting ability) as moderating variables METHOD: We analysed self-explanations from 115 university students solving six ill-defined statistics problems with or without worked examples. Self-explanation quality was defined as using specific case information from problem descriptions to justify statistical claims and coded by trained raters RESULTS: Results showed a moderated mediation of the worked example effect via self-explanation quality for prior knowledge and WMC, with stronger indirect effects at lower aptitude levels. The indirect effect was not moderated by fluid intelligence. No self-explanation mediation emerged for shifting ability; shifting ability was confirmed as a moderator of the worked example effect, with stronger scaffold benefits at lower ability levels CONCLUSIONS: Whereas the original study found no evidence for self-explanation quality as a mediator, the replication, drawing on stronger methodology, provides evidence for such mediation. However, this mechanism is not universal. Rather, worked examples exert their benefits either directly or via self-explanation, contingent on learners' cognitive resources.

Students’ difficulties with mathematical argumentation, conjecturing, and proving, as well as with collaborating effectively, have long been documented in the literature. While there is some research on what constitutes successful proving processes, systematic investigations of proving process quality are rare. We investigate how individual-mathematical and social-discursive process characteristics during a collaborative conjecturing-and-proving activity relate to students’ performance in conjecturing and proving, as well as their proof-related resources, such as prior proof skills. Based on a multidimensional high-inferent coding of collaborative conjecturing-and-proving processes of N=98 prospective mathematics students who worked in dyads, we test a range of hypotheses on the overall role of process characteristics for students’ performance. Moreover, we investigate which single process characteristics best predict conjecturing-and-proving performance. We find that individual-mathematical process quality mediates the effect of students’ resources on their performance, whereas the investigated social-discursive characteristics were largely unrelated to students’ proof performance. Mathematical argument structure was particularly predictive for performance. Results further indicate that process quality is predicted by the partners’ resources, but not by the partners’ contributions earlier in the process. However, students within a dyad converge in terms of process quality. We derive implications for the conceptualization, measurement, and investigation of process characteristics.

With digital learning environments becoming more prevalent, the ease with which generative AI enables the scalable production of real-time, automated feedback holds the potential to reshape learning and teaching experiences. This meeting report synthesizes the interdisciplinary perspectives of 50 scholars from educational psychology, computer science, science education, and the learning sciences on the use of generative AI for feedback and its promises and risks in educational practice. We highlight points of convergence in the scholarship, identify areas of debate and unresolved challenges, and outline open questions and future directions for research and educational practice that emerged from structured small-group activities designed to bridge disciplinary barriers.

Given the increasing integration of Artificial Intelligence (AI) into everyday life and professional contexts, it is essential to investigate learners' existing capabilities regarding AI tools to inform possible interventions to equip them with necessary AI skills, but also advance the theoretical frameworks on digital skills measurement and development. In this vein, this study aims to validate a Scenario-Based AI Literacy Scale (SAILS) tailored to vocational learners. In this study, we differentiate between instrumental (e.g., using AI to prepare a presentation) and critical-reflective AI skills (e.g., recognizing AI-generated deepfake content). The scale is operationalized through a scenario-based self-assessment approach, ensuring a context-driven evaluation of AI skills. We validated the AI skills scale consisting of 12 scenarios (5 instrumental and 7 critical-reflective) with a sample of police officers in training (N = 420), investigating reliability and validity of the scale. Additionally, we compared SAILS with a traditional digital skills scale to investigate convergent and discriminant validity. The analysis resulted in excellent reliability of the subscales (Cronbach's alpha = 0.88 for critical-reflective skills and 0.89 for instrumental skills) and the entire scale as a whole (Cronbach's alpha = 0.92). A 3-factor-model (including critical-reflective and instrumental AI skills subscales as well as digital skills) shows an acceptable model fit (RMSEA = 0.06, TLI = 0.93) with the standardized factor loadings ranging from 0.56 to 0.81 indicating an acceptable construct validity. AI skills were not only found to be related to more general digital skills, but also exhibiting some unique features, emphasizing that AI use requires skills which were not yet covered by digital literacy. These results provide support for an easy-to-use template, to be used for additional research in different contexts where objective performance measurements cannot be used and with a broader array of learners.Practitioner notes What is already known about this topic Basic AI skills are important in an ever-growing AI-driven world. Currently available measurements for AI literacy include self-assessments and objective performance-based tests. AI literacy is closely linked with digital skills. What this paper adds AI literacy can be seen as a related but separate construct from traditional digital skills. AI literacy can be measured through a scenario-based approach combining strengths from self-assessment and objective measures. AI literacy skills can be empirically differentiated into instrumental and critical-reflective skills, for example, recognizing AI-generated content. Implications for practice and/or policy The SAILS instrument provides educators with a reliable tool to assess learners' AI literacy when objective measurement is problematic. AI literacy should be regarded-and measured-in more than one dimension, that is, differentiated between instrumental and critical-reflective skills in order to identify learners' strengths and needs. SAILS can be adapted to a vast variety of thematic contexts and is thus applicable in many different venues of education.

This study, conducted as a registered report, investigates how expectancy beliefs and task values predict diagnostic performance in simulation-based learning environments for prospective teachers and physicians. Using a meta-analytic approach, we analyzed individual participant data from 16 studies (N = 1,492) conducted within a single research unit. Despite the homogeneity in measures and methodologies, measurement invariance analyses revealed two clusters of studies (Cluster 1: nine studies, Cluster 2: five studies) with differing patterns. For expectancies for success and diagnostic accuracy, the aggregated Fisher's z correlation was .07 (p = .03, 95% CI [.01, .13]) in Cluster 1 and -.04 (p = .63, 95% CI [-.18, .11]) in Cluster 2. For utility value and diagnostic accuracy, the correlations were .14 (p < .05; 95% CI [.02, .26]) and .08 (p = .10; 95% CI [-.01, .18]), respectively. The interaction between expectancy beliefs and utility value showed weak, inconsistent, and nonsignificant relations with diagnostic accuracy in both clusters. Findings also varied depending on whether motivational variables were assessed at a task-specific or more general level. No differences emerged between teacher and medical education contexts. These findings highlight the context-dependency of motivational processes and suggest that variations may stem from the situated nature of learning environments and how constructs are operationalized. The results also emphasize the need to tailor interventions to specific learning contexts and caution against generalizing from single studies. Overall, our study underscores the situated nature of expectancy-value frameworks and the importance of multistudy syntheses in understanding their role in professional education.

Clinical reasoning in acute care unfolds under time pressure as teams must continuously interpret evolving patient information while coordinating treatment. While research has predominantly focused on diagnostic reasoning, it remains insufficiently understood how care teams generate, coordinate, and enact interventions, and how these processes are organized. To address this gap, we introduce the Collaborative Diagnostic–Intervention Reasoning (CDI-R) model, a theory-driven process model that conceptualizes clinical reasoning as the interplay of diagnostic activities (DAs), intervention activities (IAs), and collaborative activities (CAs). We provide an initial empirical examination of the model using a virtual reality cardiac arrest simulation with 29 teams ( N = 116 participants). Team interactions were coded at the utterance level. Lag Sequential Analysis (LSA) was used to examine transitions between DAs and IAs, and Epistemic Network Analysis (ENA) was used to examine the structural organization of CAs. Findings reveal three key patterns. First, DAs and IAs constitute structurally distinct reasoning modes, characterized by different configurations of CAs. Second, reasoning unfolds in non-linear and recurrent sequences, with sustained engagement within and transitions between reasoning modes. Third, expertise shapes the organization of reasoning: expert-led teams engaged in a higher proportion of IA-oriented activity, transitioned more frequently from diagnosis to intervention, and showed more differentiated collaborative structures, whereas trainee-led teams exhibited more loops within reasoning modes. By explicitly integrating intervention reasoning into models of clinical reasoning, this study advances a process-oriented account of team-based reasoning. The CDI-R model provides a framework for examining how teams coordinate diagnosis and intervention in dynamic clinical settings.

As digitalization progresses and technologies advance rapidly, digital simulations offer great potential for learning professional practices in contexts such as medical or teacher higher education. The technological advancements increasingly facilitate the personalization of learning support to meet the individual needs of learners, whose diverse prerequisites influence their learning processes, activities, and outcomes. However, systematic approaches to combining technologies with educational theories and evidence are scarce. In this article, we propose to use data on relevant learning prerequisites and learning processes as a basis for personalizing feedback and scaffolding to facilitate learning with simulated practice representations. We connect theoretical concepts with methodological and technical approaches (e.g., using artificial intelligence) for modeling important learner variables as a basis for personalized learning support. The interplay between the learner and the simulation environment is outlined in a conceptual framework which may guide systematic research on personalized learning support in digital simulations. Educational relevance statement This paper introduces a conceptual framework, which aims to advance personalized simulation-based learning in higher education. Digital simulations can provide tailored learning experiences that adapt to students' individual differences and needs, using artificial intelligence and other technological advances. This approach might have the potential to transform learning in higher education by increasing student engagement and the effectiveness of learning professional knowledge and skills. The framework is discussed along five central questions of personalized learning, which may guide systematic research on how simulations can accommodate learners' diverse prerequisites and processes. In doing so, the framework provides a starting point for interdisciplinary research collaborations aimed at developing design principles for personalized simulation-based learning in higher

Technology has shown to be beneficial for initiating cognitive engagement. In the present study, cognitive engagement was conceptualized by the ICAP framework, proposing four levels of cognitive engagement (interactive, constructive, active, passive), which can be determined from observable student activities. To initiate cognitive engagement, teachers require diagnostic skills. With this study, we aimed to foster those skills. We designed and validated a simulation with N = 213 pre-service teachers to investigate the validity of the simulation. Moreover, we evaluated the difficulty of diagnosing the levels of cognitive engagement within planning and implementing lessons. We used linear regressions for the validation and confusion matrices for insights into the diagnostic process. The study results show a varying difficulty of diagnosing levels of cognitive engagement due to (a) challenges in inferring the involved cognitive processes and (b) different phases of teaching. Levels of cognitive engagement that require inferential processes to identify them are more difficult to diagnose. This highlights the importance of adding scaffolds to our simulation to help pre-service teachers understand the processes of generating knowledge and co-generating knowledge. More importantly, the study reveals shortcomings of the ICAP framework and presents first suggestions for its further development.

ABSTRACTBackgroundArtificial intelligence, particularly natural language processing (NLP), enables automating the formative assessment of written task solutions to provide adaptive feedback automatically. A laboratory study found that, compared with static feedback (an expert solution), adaptive feedback automated through artificial neural networks enhanced preservice teachers' diagnostic reasoning in a digital case‐based simulation. However, the effectiveness of the simulation with the different feedback types and the generalizability to field settings remained unclear.ObjectivesWe tested the generalizability of the previous findings and the effectiveness of a single simulation session with either feedback type in an experimental field study.MethodsIn regular online courses, 332 preservice teachers at five German universities participated in one of three randomly assigned groups: (1) a simulation group with NLP‐based adaptive feedback, (2) a simulation group with static feedback and (3) a no‐simulation control group. We analysed the effect of the simulation with the two feedback types on participants' judgement accuracy and justification quality.Results and ConclusionsCompared with static feedback, adaptive feedback significantly enhanced justification quality but not judgement accuracy. Only the simulation with adaptive feedback significantly benefited learners' justification quality over the no‐simulation control group, while no significant differences in judgement accuracy were found.Our field experiment replicated the findings of the laboratory study. Only a simulation session with adaptive feedback, unlike static feedback, seems to enhance learners' justification quality but not judgement accuracy. Under field conditions, learners require adaptive support in simulations and can benefit from NLP‐based adaptive feedback using artificial neural networks.

Recently, the option to use large language models as a middleware connecting various AI tools and other large language models led to the development of so-called large multimodal foundation models, which have the power to process spoken text, music, images and videos. In this overview, we explain a new set of opportunities and challenges that arise from the integration of large multimodal foundation models in education.