Self-Regulated Learning (SRL), defined as learners' ability to systematically plan, monitor, and regulate their learning activities, is crucial for sustained academic achievement and lifelong learning competencies. Emerging AI developments profoundly influence SRL interactions by potentially either diminishing or strengthening learners' opportunities to exercise their own regulatory skills. Recent literature emphasises a balanced approach termed Hybrid Human-AI Regulated Learning (HHAIRL), in which AI provides targeted, timely scaffolding while preserving the learners' role as active decision-makers and reflective monitors of their learning process. Central to HHAIRL is the integration of adaptive and personalised learning systems; by modelling each learner's knowledge and self-regulation patterns, AI can deliver contextually relevant scaffolds that support learners during all phases of the SRL process. Nevertheless, existing digital tools frequently fall short, lacking adaptability and personalisation, focusing narrowly on isolated SRL phases, and insufficiently supporting meaningful human-AI interactions. In response, this paper introduces the enhanced FLoRA Engine, which incorporates advanced generative AI features and state-of-the-art learning analytics, and grounds in solid educational theories. The FLoRA Engine offers tools such as collaborative writing, multi-agent chatbots, and detailed learning trace logging to support dynamic, adaptive scaffolding of self-regulation tailored to individual needs in real time. We further present a summary of several research studies that provide the validations for and illustrate how these tools can be utilised in real-world educational and experimental contexts. These studies demonstrate the effectiveness of FLoRA Engine in fostering SRL, providing both theoretical insights and practical solutions for the future of AI-enhanced learning contexts.

In primary education, adaptive learning technologies (ALTs) personalise tasks and provide teacher dashboards that visualise real-time student performance and progress. These dashboards can help teachers to provide targeted, real-time feedback based on the dashboard data, a practice known as dashboard-prompted feedback (DP-feedback). This study examined how teachers give different types of DP-feedback (task, process, metacognitive, social, and personal) across different lesson phases (e.g., instruction, guided practice, independent practice). We observed 25 teachers giving dashboard-supported mathematics lessons in primary education. Using visualised behavioural clustering, we analysed the distribution of teachers' DP-feedback across lesson phases. Our results showed that DP-feedback mainly consisted of task and process feedback and was mostly provided in the later phases of the lessons. Three clusters of DP-feedback were identified: feedback provided in one, two, or three lesson phases. No significant relationship was found between the type of feedback and the number of lesson phases in which DP-feedback was given. In addition, we explored the relationship between the clusters and teacher characteristics. Overall, teachers’ dashboard use differed primarily in when DP-feedback was enacted across lesson phases, rather than in which feedback types were used, highlighting the relevance of phase-sensitive support and technology-specific experience.

In het onderwijs wordt Artificial Intelligence (AI) vaak benaderd vanuit twee perspectieven: leren met AI (AI als hulpmiddel) en leren over AI (AI als onderwerp). Dit artikel benadrukt de derde invalshoek: AI om te leren over leren. Hoe kan AI ons helpen om leerprocessen zichtbaar te maken en studenten te ondersteunen bij het ontwikkelen van zelfregulatievaardigheden?

Debates about Artificial Intelligence (AI) in education are often framed in terms of teacher replacement, yet the majority of teaching tasks and skills will involve some form of teacher-AI complementarity. However, the field still lacks a shared analytical framework for systematically describing how pedagogical tasks and the professional skills and knowledge change when AI is introduced. This paper proposes an analytical approach combining Hierarchical Task Analysis (HTA) and Skill and Knowledge Analysis (SKA) to capture task-, skill- and knowledge level changes in AI-supported teaching. We apply this approach to two contrasting cases: one in primary education where teachers use a curriculum-adaptive AI tool for differentiated math lesson planning, and one in secondary education where a generative AI tool supports the planning of classroom feedback interventions. The analysis reveals two distinct complementarity configurations: one in which AI informs teacher decisions through data-driven suggestions, generating new skill and knowledge demands around evaluating algorithmic output; and one in which teacher expertise becomes the primary input shaping AI output through iterative dialogue, requiring new skills in knowledge externalisation. This novel analytical approach provides a principled foundation for moving beyond replacement narratives toward a fine-grained understanding of teacher-AI complementarity.

Research on emotions during learning has mainly compared students' emotional responses with and without feedback. It is therefore unknown how sequential instances of the same feedback influence emotions. To answer this question, 109 fifth graders solved mathematics problems and received immediate feedback on their answers. Emotional responses were captured multimodally to address physiological (arousal), experiential (valence), and behavioural responses (emotion types). The latter two responses became more negative over time, whereas one component of arousal increased. Students had more positive valence and emotion types after feedback that indicated success instead of failure. Finally, sequential feedback had a cumulative effect on students' arousal and self-reported valence, but only failure feedback had a cumulative effect on expressed emotion types. Individual variability emerged in valence and emotion types, but not in physiological responses. These findings underscore the multidimensional nature of emotional responses to feedback and the value of investigating emotions at a micro-level. Educational relevance and implications Feedback has a cumulative effect on students' emotions during learning, particularly on physiological arousal and self-reported valence. Individual variability is found in self-reported valence and expressed emotion types after feedback. Teachers should attend to students' emotional states when adjusting their instruction, recognising that not all emotions are outwardly expressed but may still manifest internally through physiological arousal.

Teachers increasingly make use of student data to inform instruction, yet evidence on the factors shaping primary teachers' in-lesson data use remains fragmented, limiting translation to classroom practice. This systematic review synthesised findings from 22 empirical studies examining teachers' use of static and real-time student data, with explicit attention to comparing factors that shape data use across these two contexts. Findings were coded across three dimensions: dispositional teacher characteristics, skills and knowledge, and contextual conditions. Data literacy, self-efficacy, and school-level collaboration were consistently found to support teachers’ data use. Differences were also evident: studies on static data emphasised attitudes, critical reflection, and external pressure, whereas studies on real-time data highlighted cognitive load, technological trust, and the need for pedagogical expertise and targeted training with adaptive learning technologies. Results suggest supporting data competencies and context-sensitive professional and organisational support, with real-time student data placing additional demands on teachers and schools.

This special issue examines advances in the measurement and support of self-regulated learning (SRL), emphasizing the integration of multimodal data and artificial intelligence (AI) in educational contexts. SRL is a goal-directed process in which learners plan, monitor, and control their learning, influenced by the interplay of cognition, affect, metacognition, and motivation (CAMM). Traditional methods in educational psychology, such as self-reports and interviews, often fall short of capturing the dynamic and recursive nature of SRL. Recent research employs multimodal data and process-oriented approaches to better understand the complex interactions among CAMM processes. The self-regulated learning, multimodal data, and analysis grid (SMA) is used as a framework for mapping and analyzing these processes across diverse data streams. The special issue includes three review papers and two empirical studies that illustrate the benefits and challenges of integrating multiple data sources and analytical techniques, while emphasizing the need for reliable and valid measures to enable personalized support for SRL. Collectively, the studies provide a multidisciplinary perspective on the current state and future directions of SRL research, advocating for innovative, theory-driven approaches that leverage existing technological capabilities to empower agentic learners in digital environments.

Generative Artificial Intelligence (GenAI) holds a potential to advance existing educational technologies with capabilities to automatically generate personalised scaffolds that support students' self-regulated learning (SRL). While advancements in large language models (LLMs) promise improvements in the adaptability and quality of educational technologies for SRL, there remain concerns about the hallucinations in content generated by LLMs, which can compromise both the learning experience and ethical standards. To address these challenges, we proposed GenAI-enabled approaches for evaluating personalised SRL scaffolds before they are presented to students, aiming for reducing hallucinations and improving overall quality of LLM-generated personalised scaffolds. Specifically, two approaches are investigated. The first approach involved developing a multi-agent system approach for reliability evaluation to assess the extent to which LLM-generated scaffolds accurately target relevant SRL processes. The second approach utilised the "LLM-as-a-Judge" technique for quality evaluation that evaluates LLM-generated scaffolds for their helpfulness in supporting students. We constructed evaluation datasets, and compared our results with single-agent LLM systems and machine learning approach baselines. Our findings indicate that the reliability evaluation approach is highly effective and outperforms the baselines, showing almost perfect alignment with human experts' evaluations. Moreover, both proposed evaluation approaches can be harnessed to effectively reduce hallucinations. Additionally, we identified and discussed bias limitations of the "LLM-as-a-Judge" technique in evaluating LLM-generated scaffolds. We suggest incorporating these approaches into GenAIpowered personalised SRL scaffolding systems to mitigate hallucination issues and improve the overall scaffolding quality.

This study explored how learners engage in shared control in math in Adaptive Learning Technologies (ALT). In shared control, learners adjust task difficulty (easy, medium, or hard), while the ALT selects tasks based on performance. These adjustments to task difficulty influence the probability of solving the next task correctly. This study aimed to understand (1) differences in how learners use shared control and (2) how this relates to general math ability, regulation of practice behaviour (number of finished problems and accuracy), and learning outcomes. In this exploratory study, 98 grade 5 learners practiced three math topics using an ALT combined with an app, including personalised visualisations of learners’ real-time progress on the math topics and shared control selection options. Results showed four clusters reflecting differences in learners’ use of shared control in quantity and direction of task difficulty changes: learners making no changes (cluster 1), learners making some changes preferring hard difficulty (cluster 2) or easy difficulty (cluster 3), and learners who frequently changed across all task difficulties (cluster 4). Shared control was related to general math ability and influenced learners’ regulation of practice behaviour. Although a comparison with the ALT control was absent, learners seem to choose task difficulties in line with their needs and benefit from these choices, resulting in learning gain. From a self-regulated learning perspective, this indicated how learners engaged in regulation and were aware of their needs.

Learning Analytics Dashboards (LADs) have the potential to serve as a Self-Regulated Learning (SRL) support tool in secondary education. While LADs have demonstrated their potential in higher education contexts, secondary education remains largely undiscovered. Using a Design-Based Research approach, an existing higher education student-facing LAD was adapted in co-design with teachers throughout three design cycles. Across the three design cycles, five design propositions for SRL dashboards in secondary education were derived from empirical evidence and informed by theory. Together, the design propositions suggest that student-facing SRL LADs in secondary education should use accessible language, clearly communicate relevance, tightly integrate assessment and SRL features to support meaningful student reflection, make assessment criteria clear, and make relevant information easy to access. Although the dashboard did not fully meet all propositions, qualitative findings indicate that most students were able to use the LAD to reflect on their learning, thereby demonstrating that the co-designed dashboard achieved its goal of SRL support.

Generative artificial intelligence (GenAI) is rapidly becoming a pervasive technology across education, work, research, and everyday life. Yet growing evidence suggests that fluent, coherent, and seemingly transparent AI assistance can improve task performance while weakening the metacognitive engagement and deeper processing required for durable learning. This tension raises a central question: how does offloading to AI reshape human learning, and under what conditions do performance and learning decouple? In this Perspective, we advance a metacognitive account of human-AI interaction by conceptualising metacognitive offloading and metacognitive onloading and by introducing the Synergetic Notion of Human-AI Skills and Knowledge Construction (SYNC) model. The SYNC model explains how human-AI interaction unfolds across learning and performance dimensions, clarifying how learners may drift into overreliant AI-supported performance or move toward human-AI synergy. We further identify key pitfalls in ineffective human-AI interaction and outline strategies for scaffolding metacognitive engagement. This Perspective provides a conceptual foundation for understanding when AI-supported performance contributes to durable human capability and when it instead risks undermining durable learning.

Developing self-regulated learning (SRL) is essential for students to ensure lifelong learning in a changing world. As students often do not develop SRL skills independently, teachers play a crucial role in nurturing these skills through metacognitive feedback. To provide this type of feedback, teachers need an accurate assessment of students’ metacognition. Teachers gain valuable insights from dashboards, but these dashboards often lack students’ metacognitive indicators. Therefore, this study engaged in a participatory design approach with primary school teachers to design an SRL teacher dashboard. The goals were to (1) identify teachers’ needs for information regarding primary school students’ metacognitive indicators, and (2) design an SRL teacher dashboard that analyzes, organizes, and presents this information. Two design rounds were conducted: focus groups gathered preferences from 19 educational professionals regarding metacognitive indicators and interface features, followed by a questionnaire administered to 45 primary school teachers to further refine these indicators and features, guiding the design of the SRL teacher dashboard. In our findings, teachers expressed a need for metacognitive indicators in the goal setting, enactment, and adaptations phases of the COPES model, while giving less priority to the task definition phase. Regarding dashboard interface, teachers valued personalization options, along with additional information (e.g. students’ growth by grade), and clarity for quick data overview. The present study reveals primary school teachers’ need for metacognitive indicators and exemplifies how participatory design can integrate COPES phases and teacher input to design an SRL teacher dashboard, effectively merging theoretical frameworks into practical classroom tools.

This study addresses the growing need for ethical guidelines for artificial intelligence (AI) in education research by testing and refining procedures for responsible experimentation with AI in educational settings. The researchers engaged 21 Dutch Institutional Review Board (IRB) members across two feedback rounds to evaluate available ethical procedures from the literature. These procedures focus on four key areas: gradual scaling, side-effects monitoring, proportional stopping rules, and stakeholder consultation, with different requirements based on levels of AI automation and human oversight. Results showed strong consensus among IRB members that AI experiments require specialized ethical procedures beyond standard human subjects protections. After incorporating feedback from the first round, including clarifying terminology, distinguishing between general-purpose AI and educational AI, and providing clearer process descriptions, the second round achieved unanimous agreement on the procedures' clarity and utility. Notably, while IRB members agreed on the need for specialized procedures, they believed regular IRBs could handle most cases rather than requiring automatic escalation to specialized committees. The study contributes practical, stakeholder-validated procedures that balance ethical rigor with bureaucratic efficiency, enabling IRBs to better assess AI education research while distinguishing between higher and lower risk experiments based on AI type and automation level.

This study investigates secondary education students' self-regulated learning (SRL) processes with digital trace data, particularly whether SRL processes found in secondary education are comparable to those observed in higher education. We therefore adapted a digital learning environment and rule-based AI algorithm originally designed to measure SRL in higher education and collected multi-trace data from 13-year-old students (N = 179) across three European countries during an essay-writing task. Hidden Markov modeling was employed to capture latent SRL processes. Four latent SRL processes emerged: orientation, first-reading, writing, and re-reading combined with monitoring. By clustering sequences of these latent SRL processes, we identified four sequential patterns of SRL processes at the task level: writing with metacognitive monitoring, writing intensively, reading first, writing next, and reading and writing simultaneously. Our findings highlight how AI and multitrace data can be used to measure SRL during learning, providing a basis for enhancing personalized support. Educational relevance and implications statement: Self-regulated learning (SRL) is vital in the digital world. In this study, we investigated secondary education students' SRL processes with digital trace data. We also demonstrated that the instrumentation of a digital learning environment and rule-based AI algorithm originally designed to measure SRL processes in higher education can be leveraged to measure SRL processes in secondary education students. The real-time measurement of covert SRL processes is important as this information can 1) raise students' awareness on their learning, 2) help teachers to support students' learning, and 3) form the basis for providing personalized support for these processes with the help of AI-enhanced learning technologies.

The introduction of artificial intelligence (AI) into education holds promise for supporting and augmenting teaching and learning‐related activities. Yet, despite its potential, there is limited empirical research on the use of AI in K‐12 settings exploring the pedagogical grounding, impact and implications of the technological solutions. The current paper reports a systematic literature review of 28 papers that addresses the topic of AI technologies in authentic settings. We applied etic and emic content analysis to the collected data. Our results indicated that most AI research is conducted in China and the United States, and investigates AI solutions for primary education. As main AI technologies we found social robots, augmented reality and intelligent systems. The review also revealed that half of the papers lack a solid pedagogical foundation and a contextualisation under concrete course conditions. Finally, most AI solutions focus mainly on learning outcomes without exploring the teacher needs and impressions. Our findings suggest the need for research to focus on: (a) AIEd localisation to fit the AI context of different regions; (b) contextualisation of the tool design under several course conditions and learning theories; (c) equal attention on teacher agency and autonomy within the AIEd era; and (d) systematisation of the evaluation of sustained pedagogical effects (e.g., cognitive, relational, affective effects) apart from student learning outcomes. We envision that our research can support the design, development and deployment of AI tools in K‐12 under such pedagogical considerations, hence contributing to strengthen the ties between pedagogy and AI practices in authentic settings.

Adaptive Learning Technologies generate data traces as children interact with them, offering a unique opportunity to estimate self-regulated learning (SRL) support needs. Providing timely, data-driven support for these needs may enhance children's ability to self-regulate their learning and improve learning outcomes. While previous work has identified different levels of SRL support needs using Bayesian non-parametric clustering, these classifications were determined after completing a learning session, delaying potential support. In this study, we present a novel method for identifying children's SRL support needs during a learning session, utilizing a Dirichlet-process Gaussian-process mixture model (DPGP). The model clustered based on children's response count and response time, resulting in a Matthews' correlation coefficient of 0.75 after children solve 33 problems out of an average of 71 - less than half the session. Our findings demonstrate that real-time identification of SRL support needs is feasible and effective. This work opens new possibilities for enhancing personalized, online learning experiences by enabling timely, data-driven support tailored to each child's needs.