Frontiers in Education: Digital Learning Innovations
1 day 12 hours ago
Generative AI is increasingly reshaping how learners engage in informal English learning beyond the classroom, making AI-mediated informal digital learning of English (AI-IDLE) an important emerging learning context. Yet research has predominantly focused on intra-personal psychological factors, the social dimensions of learning, particularly how learners construct their social identities in relation to AI, remain significantly underexplored. This qualitative study examines the social identity categories and their construction processes among 12 Chinese university EFL learners engaged in AI-IDLE. Drawing on Social Identity Theory (SIT), data from semi-structured interviews, reflective journals, and AI chat logs were analyzed using thematic analysis. The findings reveal two distinct identity pathways: learning-oriented academic identities, characterized by deep-learning commitment, humble mindset toward AI, and sustained immersion in the student role; and an instrumental identity, organized around a user category defined by transactional engagement, service satisfaction and surface learning approaches. Within this instrumental pathway, a more intensified fast-food consumer orientation emerged, which was explicitly self-identified by one participant and analytically inferred from behavioral patterns across several others. The study demonstrates that AI-IDLE does not simply erase or replace existing social identities but selectively reinforces and reconfigures them through group norms, prototypical markers, and evaluative comparison. These findings extend SIT by revealing how technological affordances and social configuration shape identity pathways that support or undermine meaningful learning engagement. Practical implications are proposed for AI developers, digital learning designers, educators, and curriculum designers to foster learning-oriented identity pathways in AI-mediated informal learning environments
Wu Fei
1 day 14 hours ago
Generative AI is now a routine presence in higher-education classrooms, shifting the operative question from whether to permit its use toward when and how access should be structured within a learning activity. This paper reports a design-based research programme examining phase-based AI constraints, deliberately restricting AI access to specific phases of a collaborative teaching method rather than permitting or prohibiting it wholesale, as a design principle for structuring human and AI contributions within collaborative learning activities. The wider DBR programme examined AI-augmented designs across six collaborative teaching methods, administered repeatedly to five student groups over two academic years; the phase-based AI-constraint principle reported here as DP1 was developed through, and its evidence anchored in, Think-Pair-Share, with the other five methods contributing broader DBR context rather than independent confirmatory evidence. Its strongest evidence comes from an AI-free Think phase within Think-Pair-Share, replicated across three independent student groups spanning three degree programs, where learners consistently endorsed the AI-free phase as appropriate; this endorsement pattern was typically the highest-scoring item on its questionnaire. This design itself resulted from an earlier design cycle within the same programme: an initial administration permitting AI access during this phase prompted redesign after the instructor's classroom observation indicated that independent reflection was being displaced by AI-mediated responses. We treat this endorsement as an empirical finding, not as a direct assessment of the quality or originality of what learners produced, and not as evidence that the phase in fact protects independent cognitive engagement; that stronger claim is a theoretically motivated proposition this study does not test, and we frame DP1 accordingly across three explicit levels: an empirical endorsement finding, a transferable design principle of differentiating AI access by activity phase rather than deciding it once for a whole activity, and an untested theoretical proposition about the cognitive mechanism this differentiation may support. A complementary, independent stakeholder strand surveyed higher-education professionals at a practitioner festival session, offering a modest, exploratory social-validity signal: the phase-based procedure itself was almost universally endorsed, while personal adoption intention was somewhat more reserved. An exploratory qualitative coding of practitioners' open responses showed only slight inter-coder reliability (κ=0.17), so we do not use these responses to derive additional design principles, retaining only one response on function-based AI restriction as an illustrative example. The study is exploratory and formative, without a no-AI control condition or random assignment, and we frame its claims as design-principle evidence within a design-based research programme rather than as causal effect estimates. Anchored in independently replicated student endorsement data from Think-Pair-Share and supplemented by a practitioner social-validity signal, phase-based AI constraints offer higher-education instructors an actionable, low-friction design principle, one we expect to be conceptually transferable to other collaborative teaching methods, though this transfer remains to be empirically demonstrated.
Felix Höpfl
2 days 2 hours ago
In an era of expanding digital finance, fintech innovation and AI-supported decision-making, financial literacy has become a critical educational priority for informed, inclusive and responsible participation in contemporary financial systems. This study examines financial literacy in digital and AI-supported educational contexts through bibliometric mapping and a PRISMA-based systematic review. A bibliometric analysis of 3,105 Web of Science and Scopus records identified the broader conceptual structure of the field, followed by thematic synthesis of 29 studies with educational relevance. Methodological quality was assessed using design-specific JBI Critical Appraisal Tools. The findings show that financial literacy and financial inclusion form the main conceptual hubs, while fintech and digital finance act as bridging themes. AI-related research is visible, but is mainly linked to prediction, decision-making, fintech adoption and digital services rather than direct pedagogy. The review further shows that digital tools are used across diverse educational and assessment contexts, with the more consistent evidence emphasizing structured instructional design and pedagogical integration rather than technology exposure alone. AI-related evidence remains limited and is predominantly analytical, diagnostic, advisory or supportive rather than directly pedagogical. The main research gap lies in the limited empirical evidence on pedagogically grounded AI-supported approaches to financial literacy education. Overall, the study provides an integrated framework for identifying conceptual structures, emerging digital and AI-supported trends, and critical pedagogical gaps relevant to evidence-informed and inclusive financial literacy education.
Nika Drnovšek
3 days 12 hours ago
Despite scientific argumentation being a fundamental competency of the PISA assessment framework, prospective biology teachers often demonstrate insufficient development in evidence-based reasoning and rebuttal construction skills. Considering the limitations of large-scale implementation of individualized approaches in higher education, this study evaluates the effectiveness of Generative Artificial Intelligence (ChatGPT-4) configured as a “Socratic tutor” in developing argumentation skills among future educators. An 8-week quasi-experimental study involving 124 students from a pedagogical university employed traditional peer discussion methods in the control group (n = 62), while the experimental group (n = 62) utilized an AI agent that induced cognitive dissonance regarding socio-scientific issues. Analysis of Covariance (ANCOVA) of data obtained through Toulmin's Argument Pattern (TAP) and PISA rubrics confirmed that the experimental group achieved statistically significantly higher overall argumentation quality compared to the control group (F (1,121) = 14.23, p < .001, η2 = .105). Particularly, participants who received AI-assisted instruction showed marked improvement in forming rebuttals and evaluating counterevidence, which are considered among the most cognitively demanding components within the PISA scientific literacy framework (OECD). These findings suggest that a Generative AI-mediated Socratic instructional package can be an effective and scalable approach for developing epistemic cognition, and they propose a pedagogical model for integrating structured AI-based tutoring into teacher preparation systems aligned with international assessment standards. These findings should be interpreted as evidence for the effectiveness of the structured, AI-mediated Socratic instructional package as a whole, rather than as evidence of AI's inherent superiority over other methods; the relative contribution of the AI medium versus the structured pedagogical protocol it delivers was not directly tested in this design.
Gulrukhsor Surkhonidinovna Ergasheva
6 days 13 hours ago
As generative artificial intelligence (GenAI) becomes increasingly integrated into education, understanding whether students possess the foundational conditions needed to benefit from AI-mediated learning has become an important educational challenge. This study conceptualizes dispositional foundations for AI-mediated learning as the configuration of ICT self-efficacy, attitudes toward technology and expectations of future ICT use that may support meaningful engagement with AI-mediated learning environments, and examines how these configurations relate to objectively assessed digital competence and to student characteristics. ICILS 2023 contains no measure of AI literacy or AI use, so these are candidate foundations rather than a direct measure of preparedness for AI. Using data from 118,764 Grade-8 students across 35 education systems participating in the International Computer and Information Literacy Study (ICILS 2023), we conducted sampling-weighted latent profile analyses on four internationally comparable indicators. Computer and information literacy (CIL) was examined as an external criterion using the BCH approach, and profile membership was predicted by socioeconomic background, gender, educational aspirations and ICT access. Four profiles emerged: average (55.9%), attitude-oriented (25.2%), low-disposition (11.0%) and efficacy-dominant (7.9%). Differences in CIL across profiles were modest, and students in the efficacy-dominant profile, who reported ICT self-efficacy more than two standard deviations above the mean, did not outperform the average profile in objectively assessed competence, a pattern we interpret as a confidence-competence mismatch rather than as high preparedness. On a common metric of average marginal effects, gender showed larger and more consistent contrasts than socioeconomic background. These findings suggest that the dispositional foundations for AI-mediated learning are distinct from computer and information literacy and should be understood as a multidimensional configuration of learner characteristics whose relevance to actual AI-mediated learning remains to be tested directly.
Vanessa Izquierdo-Álvarez
6 days 13 hours ago
IntroductionThis study examined the relationships between the application of software technologies and the delivery of Business Education content in federal universities in South South Nigeria. Specifically, it determined the individual and joint relationships of Cognitive Artificial Intelligence (AI), Voice on the PowerPoint (VOPPT) lectures, Skillshare and Google Forms with Business Education content delivery.MethodsA correlational research design was adopted. The population comprised 306 respondents, and the entire population was studied. Data were collected using a structured questionnaire and analysed using simple and multiple linear regression at the .05 level of significance. Five research questions were answered and five null hypotheses were tested.ResultsCognitive AI had a significant positive relationship with Business Education content delivery (R = .574, R² = .330, β = .574, t = 12.235, p < .001). VOPPT also had a significant positive relationship (R = .626, R² = .392, β = .626, t = 14.011, p < .001), while Skillshare demonstrated a significant positive relationship (R = .578, R² = .335, β = .578, t = 12.363, p < .001). Google Forms recorded the strongest individual relationship (R = .804, R² = .646, β = .804, t = 23.572, p < .001). Collectively, the four technologies explained 65.6% of the variance in Business Education content delivery (R = .810, R² = .656, adjusted R² = .652), with the overall regression model being significant, F(4, 301) = 143.634, p < .001. In the joint model, Skillshare (β = .185, t = 2.921, p = .004) and Google Forms (β = .808, t = 14.746, p < .001) made significant unique contributions, whereas Cognitive AI (β = −.107, t = −1.395, p = .164) and VOPPT (β = −.062, t = −.775, p = .439) did not.DiscussionThe results indicate that the four software technologies are individually associated with Business Education content delivery, but their unique contributions differ when considered simultaneously. The findings highlight the importance of strategic technology integration, lecturer capacity development, reliable digital infrastructure and institutional support for the effective delivery of Business Education content in federal universities.
Edet E. Okon
1 week ago
BackgroundEngagement and achievement gaps in higher education remain persistent, especially for students from lower socioeconomic and minority cultural backgrounds in digital learning environments.ObjectiveThis study examined whether participation in LMS-based courses implementing a Culturally Responsive Digital Differentiated Learning (CR-DDL) Framework was associated with greater improvements in student engagement and learning outcomes and with narrower observed engagement disparities compared with standard LMS-based instruction.MethodsAn explanatory sequential mixed-methods design was used with 252 undergraduates in public health and agrotechnology programmes at three Indonesian institutions, comparing CR-DDL courses (n = 129) with standard LMS-based courses (n = 123). Quantitative data included pre- and post-intervention Higher Education Student Engagement Scale scores, LMS-derived behavioural indicators, and final course scores. Outcomes were analysed using linear mixed-effects models with a random intercept for course to account for the clustering of students within the six participating course clusters. Eighteen students also participated in semi-structured interviews to provide contextual insight into their learning experiences.ResultsThe CR-DDL group demonstrated a larger increase in total engagement than the comparison group (mean change, +0.52 vs. + 0.13; group × time estimate = 0.31, SE = 0.07, approximate Wald 95% CI 0.17–0.45, p < 0.001). Greater increases were also observed in LMS-based participation. After adjustment for prior-semester GPA and course-level clustering, CR-DDL participation was associated with a 3.41-point higher final course score on the 0–100 reporting scale (SE = 1.12, approximate Wald 95% CI 1.21–5.61, p = 0.003). Lower-SES students in CR-DDL courses showed larger descriptive engagement gains than their counterparts in comparison courses. Because the formal group × time × SES interaction could not be verified from the currently reported summary results, the SES findings are interpreted as exploratory subgroup patterns rather than definitive evidence of effect modification. Documented interview excerpts provided limited contextual insight into students’ experiences but were not treated as independently reproducible qualitative evidence.ConclusionThe quantitative findings provide evidence that CR-DDL was associated with improved engagement, LMS participation, and course performance and with a narrowing of observed engagement disparities. The available qualitative material provides limited contextual insight into students’ experiences but should be interpreted cautiously because complete original qualitative records are currently unavailable.
Reno Renaldi
1 week ago
IntroductionGenerative artificial intelligence has diffused rapidly among learners, but a question remains whether that reach becomes useful for learning or stalls at access. This distinction matters most where learners already have connectivity, so the barrier is no longer reaching the technology but converting access into learning use.MethodsThis study draws on a survey of 14,653 computer science learners across 166 countries, a population for whom first-level access is broadly shared. A two-part model separates whether an everyday user converts to any learning use from how widely converters then use the tool, estimated across World Bank income tiers.ResultsEveryday adoption is high across income levels and is not lower in poorer settings. The quantity that varies is conversion. Among everyday users, those in low-income countries are markedly less likely to convert to any learning use, at an odds ratio of 0.49 (95% CI 0.27, 0.90), while breadth among converters does not differ by income tier. A latent typology shows that disengaged learners are found across every income tier rather than in one region, and the association between AI use and self-rated proficiency is strongest where other support is weakest.DiscussionThe gap sits at the threshold of first learning use, not in the breadth beyond it. This article names the pattern the access-conversion gap and develops it as a measurable construct at the second and third levels of the digital divide. The findings describe a connected population of engaged learners, so the divide to address is in use, not access.
Krishnashree Achuthan
1 week 1 day ago
This mixed-method exploratory study examined the development of AI literacy among (N = 19) master's-level special education graduate students as they explored the integration of generative AI tools to support students with high-incidence disabilities. Participant responses were examined to identify perceptions of AI, intended uses of generative AI, and considerations related to responsible implementation in special education settings. Findings revealed a conceptual shift in participants' perspectives, from viewing generative AI primarily as an administrative tool for increasing efficiency to recognizing its potential as an adaptive instructional scaffold. Participants increasingly emphasized the need for critical human oversight, FERPA compliance, data privacy, and individualized alignment between AI-generated resources and students' educational needs. Intentional AI literacy development can support future special educators in moving beyond basic technical proficiency toward the critical judgment needed to evaluate and implement generative AI responsibly. Preparing special educators to use AI effectively requires attention not only to its potential benefits for instruction and efficiency but also to legal, ethical, and pedagogical considerations, including student confidentiality and individualized educational planning.
Randa Keeley
1 week 3 days ago
IntroductionThe modern higher education classroom is incredibly diverse, with students displaying various strengths, challenges, prior knowledge, and learning preferences. Traditional “one-size-fits-all” teaching methods often fail to meet different learners' needs and hinder student engagement and academic achievement. To address this, Universal Design for Learning (UDL), a research-based, learner-focused approach grounded in cognitive neuroscience, provides a proactive path to design more inclusive and accessible teaching practices.MethodsEmploying a quasi-experimental design, this study examined written lesson-planning performance among 45 Moroccan pre-service English as a Foreign Language (EFL) teachers across three conditions: a control group (CG), a UDL-only training group (TG1), and a UDL training supplemented with AI tools group (TG2).ResultsAnalysis of the 38 participants with matched pretest and posttest scores indicated an adjusted between-group difference in posttest lesson-plan scores. After controlling for pretest scores, TG2 had higher adjusted posttest scores than both CG and TG1, whereas TG1 did not differ significantly from CG. The TG2-versus-TG1 comparison was sensitive to the estimation method and is interpreted cautiously.DiscussionThe findings indicate an adjusted difference associated with the UDL + AI training package as delivered. Because the UDL + AI condition also included additional training time, demonstrations, guided prompts, practice, and feedback, the design cannot isolate the contribution of AI tools alone. The evidence is limited to rubric-scored lesson plans in a single Moroccan institution, and a larger, time-matched study is needed to examine which components of the training account for the observed difference.
Abdelkabir Mokhlesse
1 week 6 days ago
The adequate application of artificial intelligence (AI) technology in legal education in the United Arab Emirates (UAE) has been barely covered by scholarly literature, especially, from the perspectives of legal and educational sciences. The current paper attempts to fill in an existing gap in literature about the appropriate usage of AI in legal education. The research is intended to help curriculum designers in UAE universities to improve undergraduate legal education programs using innovative AI systems and, thus, to provide UAE policy makers with the academic and ethical standards for the use of AI in legal education. The current paper critically discusses the possible contribution of AI systems in the improvement of the quality of legal education. Moreover, the research analyzes the probable implications of using AI in law curricula for the learning process and for the development of critical legal skills in UAE law students, stressing on the importance of ensuring academic integrity. Furthermore, the current research uses a qualitative approach to analyze the legal, educational, pedagogical, and ethical aspects of AI integration into legal education in the UAE and suggests a legal-pedagogical model of its responsible integration. The goal of the analysis is to determine what opportunities may be offered by AI systems in improving the quality of legal education in the UAE and to discuss their ability to educate students to perform independent legal practice. The discussion focuses on four main areas: (i) the quality of legal education in the UAE’s legal context; (ii) understanding of the AI-enabled legal education systems; (iii) identifying the educational and ethical risks of using AI in legal education; and (iv) setting the standards for the responsible use of AI in legal education. To conclude, the article argues that the adequate use of AI can improve significantly the quality of legal education, if appropriate ethical and academic standards will be developed to ensure the integrity of the process and to develop legal competencies of the students.
Mohamed Morsi Abdou
2 weeks ago
IntroductionGenerative AI creates a curriculum-transformation problem by redistributing activities through which disciplinary expertise is learned, practised, demonstrated, and assessed. This study examines that problem within graphic design education as a bounded empirical case.MethodsA survey of 270 participants was analysed using exploratory factor analysis (N = 265 complete cases), and semi-structured interviews with 12 design academics and practitioners were analysed using a frozen rule-based codebook across 650 meaning units. The quantitative and qualitative strands were analysed independently and integrated through a joint display.ResultsThe quantitative analysis yielded a broad, sample-specific Transformative–Normative response orientation and a narrower Curricular Diagnostic Critique cluster; neither is interpreted as a validated latent construct. The final qualitative dataset contained 504 normative/transformative units (77.5%), 35 diagnostic units (5.4%), and 111 units retained as X0 (17.1%). Both the N-family and D-family occurred in all 12 interviews.DiscussionThe integrated interpretation informed the Bot-Haus Curriculum Architecture, which organises human creative agency, experiential human–AI practice, adaptive curriculum, critical and ethical governance, foundational design principles, and educational outcomes. The architecture is presented as an empirically informed, theory-guided design proposition developed from the graphic design case, not as a validated causal model. Transfer to other practice-based disciplines remains a proposition for future empirical testing.
Khaled Mostafa Ahmed Mohamed
2 weeks 1 day ago
IntroductionPromoting scientific argumentation in the context of socio-scientific issues remains a major challenge in science education. Artificial intelligence offers new opportunities to provide adaptive feedback on students’ written arguments; however, little is known about learners’ perceptions of hybrid learning environments that combine AI-supported feedback with peer feedback under authentic classroom conditions. This study presents the development and pilot evaluation of SMOKEChem, a web-based learning environment designed to foster scientific argumentation through AI-supported and collaborative feedback processes.MethodsAn exploratory mixed-methods design was employed to investigate learners’ perceptions of SMOKEChem during regular chemistry lessons. A total of 130 secondary school students (mean age = 15 years) participated in the study. Quantitative data were collected using questionnaires assessing technology acceptance, cognitive load, and chatbot usability, while qualitative responses explored students’ perceptions of peer feedback.ResultsThe findings indicate that SMOKEChem was perceived as useful, user-friendly, and cognitively balanced, with low levels of extraneous cognitive load and high levels of germane cognitive load. Furthermore, students perceived AI-supported and peer feedback as fulfilling complementary functions: acceptance and usability data indicate that AI provided immediate, structured guidance for revising arguments, while qualitative data on peer feedback point to dialogic reflection, perspective-taking, and collaborative knowledge construction.DiscussionThe findings provide initial empirical evidence that hybrid feedback designs combining AI-supported and peer feedback can effectively support scientific argumentation in socio-scientific contexts. The complementary strengths of both feedback sources highlight the potential of integrating AI into collaborative learning environments rather than replacing peer interaction. These results provide practical implications for the design of AI-supported argumentation environments and inform future research on hybrid feedback approaches in science education.
Laura Celine Leppla
2 weeks 1 day ago
Learning analytics promises early identification of struggling students, but evidence comes mainly from learning-management-system counts rather than authentic laboratory telemetry. We tested the predictive value of fine-grained process traces in a first-year digital electronics laboratory. The open Educational Process Mining dataset contains 115 engineering undergraduates observed across six simulator sessions (230,318 records) and 93 students with examination outcomes. Under a pre-specified, leakage-aware protocol—nested repeated cross-validation, pipeline-internal preprocessing, a 5×3 feature-group-by-model ablation matrix, and bootstrap confidence intervals—we predicted below-median final-examination performance. The pre-specified behavioural logistic model, using eleven features spanning volume, engagement, and interaction intensity, achieved an AUC of 0.617 (95% CI 0.491–0.737), not reliably above chance; the best non-pre-registered behavioural model reached 0.648. Four in-course-grade features achieved an AUC of 0.764 (0.644–0.854), the best matrix cell. Behaviour underperformed grades (ΔAUC −0.147, 95% CI −0.263 to −0.028), and adding telemetry to grades produced no detectable improvement (ΔAUC −0.040, −0.083 to 0.006). With one session, behavioural prediction was below chance and plateaued near 0.62 from session four, while the single grade available at session two already outperformed six sessions of telemetry; its signal concentrated in exercise coverage and workflow and partly reflected a small lower-participation group that often missed the examination. Full telemetry predicted non-participation well (AUC 0.872), but exercise coverage alone matched it (0.879). Continuous-score regression also favoured grades as standalone predictors (best MAE 21.29 vs. 22.38), although nonlinear unions modestly improved error. In this setting, telemetry is primarily a coarse disengagement flag; the course’s assessment loop remains the more trustworthy signal for participating students.
Mingyu Wu
2 weeks 1 day ago
Artificial Intelligence (AI)-generated feedback is increasingly transforming mathematics education by providing timely, personalized, and adaptive learning support. However, limited evidence exists on the behavioural mechanisms through which students’ perceptions of AI-generated feedback influence mathematics learning outcomes, particularly in Sub-Saharan Africa. This study examined the direct and indirect relationships among Perceived AI Feedback Quality (PAQ), AI Dependence (AID), and Mathematics Learning Outcomes (MLO) among 360 undergraduate students in Ghana. A quantitative cross-sectional survey design was adopted, and data were analysed using Structural Equation Modelling (SEM) with IBM SPSS Statistics 23 and AMOS 23. The measurement model demonstrated satisfactory reliability and validity. The structural model showed that PAQ significantly predicted AID (β = 0.673, p < .001) and MLO (β = 0.449, p < .001), while AID also significantly predicted MLO (β = 0.441, p < .001). Bootstrapping analysis further revealed that AID partially mediated the relationship between PAQ and MLO (β = 0.314, p < .001). The findings indicate that students who perceive AI-generated feedback as accurate, relevant, and useful are more likely to depend on AI, and this dependence is in turn associated with stronger perceived mathematics learning outcomes. The study extends Feedback Theory and Social Cognitive Theory by identifying AI Dependence as a key behavioural mechanism linking AI feedback quality to mathematics learning. The findings provide practical implications for educators, instructional designers, and policymakers seeking to promote the effective and responsible integration of AI technologies into mathematics education.
Emmanuel Kusi
2 weeks 2 days ago
IntroductionArtificial intelligence (AI) is rapidly transforming research practices in higher education, yet empirical evidence on how AI supports research productivity among students remains limited, particularly in developing contexts. This study examined the influence of AI tools used during the research preparation, analysis, and dissemination stages on the research productivity of postgraduate students in public universities in Nigeria. The study was guided by the Unified Theory of Acceptance and Use of Technology 2 (UTAUT2) and Experiential Learning Theory.MethodsThe study adopted a descriptive survey design involving 312 respondents. Data were collected using the AI Tools Utilisation and Research Productivity Questionnaire (AITURPQ) and analysed using simple linear regression. The analysis examined the predictive effects of AI-assisted research preparation, analytical, and dissemination tools on research productivity.ResultsThe findings reveal that AI-assisted research preparation tools significantly predicted research productivity (R² = .241, p < .05). AI-supported analytical tools also exerted a significant positive influence on research productivity (R² = .137, p < .05), while AI-driven dissemination tools demonstrated the strongest predictive effect (R² = .329, p < .05).DiscussionThe findings underscore the role of AI across the research lifecycle in supporting knowledge creation, analytical competence, and scholarly dissemination. They highlight the need to strengthen AI literacy and institutional support and promote responsible AI integration to enhance postgraduate research productivity.
Edet E. Okon
2 weeks 6 days ago
Generative artificial intelligence (GenAI) and artificial intelligence (AI) are transforming STEAM education by reshaping instructional design, learning facilitation, assessment, creativity, and interdisciplinary learning. Their rapid evolution presents both opportunities and challenges for educators and educational institutions. This article synthesized contemporary scholarship on GenAI, AI, pedagogy, ethics, and digital innovation. Structured searches of Scopus, Web of Science, ERIC, and Google Scholar, supplemented by the UNESCO Digital Library and the OECD iLibrary and by citation chaining, identified 107 records, of which 15 met the eligibility criteria and were included in the reviewed corpus; additional literature cited in the findings and discussion attaches the pedagogical argument and contextualizes the synthesis. The primary search window covered 2016–2025, with foundational theoretical works purposively retained. Records were screened using explicit inclusion and exclusion criteria in two stages. The synthesis examined the conceptual foundations of Gen AI and AI in STEAM education, pedagogical theories supporting responsible integration of Gen AI and AI, and Gen AI and AI tools applicable to STEAM classrooms. Findings highlight applications in lesson planning, personalized learning, inquiry- and project-based learning, formative assessment, feedback, academic integrity, coding, media generation, and interdisciplinary learning. GenAI and AI can enhance teacher productivity, creativity, collaboration, adaptive instruction, and learner engagement when supported by sound pedagogy, AI literacy, and ethical governance. The article proposes an integrated framework for responsible human–AI collaboration linking pedagogical principles, ethical considerations, and discipline-specific applications. It offers practical guidance for fostering innovative, equitable, and contextually responsive STEAM education in an increasingly AI-enabled educational environment.
Niroj Dahal
3 weeks ago
BackgroundLarge language models (LLMs) are increasingly being considered for assessment support in health professions education; however, evidence of their performance in essay-style examinations remains limited. In particular, little is known about the reproducibility and operational stability of LLM-based grading under different conditions.ObjectiveThis study compared the grading performance of several LLMs with that of human examiners in a master's-level public health course and assessed the consistency of LLM-based grading across repeated sessions and different file upload volumes.MethodsWe conducted a method comparison and validation study using anonymized student submissions from a 5-hour essay-style examination in a master's-level course in public health, empowerment, and health promotion. Four LLMs—ChatGPT, Gemini, LeChat, and Kimi—were prompted to assign final grades on an A-F scale using the same grading guidance as human examiners. The agreement between the LLM- and human-assigned grades in a single-file upload setting was assessed using weighted Cohen's kappa. Internal consistency across file upload volumes was assessed using Krippendorff alpha. Repeated grading across multiple sessions was performed for the best-aligned model.ResultsAgreement with human examiners was limited in fast mode and improved in reasoning or thinking modes; however, reproducibility across sessions and implementation conditions remained limited. In the single-file upload setting, ChatGPT showed the strongest agreement with human examiners (weighted kappa 0.718, 95% CI 0.578–0.859), followed by Kimi (weighted kappa 0.571, 95% CI 0.373–0.768). ChatGPT achieved 50.0% exact agreement and 90.6% agreement within ±1 grade, with corresponding values of 28.1% and 78.1% for Kimi. Gemini and Gemini Pro showed the highest internal consistency across the file upload conditions. Repeated grading by ChatGPT across 5 days showed moderate variation. LLMs used fewer extreme grades than did human examiners.ConclusionsSome LLMs, particularly ChatGPT and Kimi, showed moderate alignment with human examiners, but the alignemnet was not consistently stable across repeated sessions or operational settings. LLMs may currently be better suited as supervised grading assistants than as autonomous graders.
Asgeir Brevik
3 weeks ago
Teaching Virtual Reality (VR) design software to students with disabilities remains challenging in Ecuadorian higher education, where tool complexity, accessibility constraints, and diverse learning needs can limit participation. This study examined Hercules VR, a pedagogical framework integrating gamification, co-design, and biomimicry to support Unreal Engine learning among university students with cognitive, sensory, and motor disabilities. Using a single-group, mixed-methods pre-post pilot design (quasi-experimental, no control group), 26 students participated in a 16-week semester including 12 structured instructional sessions. The framework combined a co-design stage adapting activities to participants’ needs, a gamified progression system, and a biomimetic approach inspired by the life cycle of the Dynastes hercules beetle to make complex software concepts more accessible. Pre- and post-test instruments assessed VR design comprehension, collaborative participation, and intrinsic motivation. Results showed statistically significant within-group improvements across all self-reported dimensions (Wilcoxon signed-rank tests, all exact p < .001), with very large effect sizes (Rosenthal’s r = 0.86–0.87), computed from a dataset independently verified during revision. Post-test scores across the three outcomes were not significantly correlated, suggesting they changed relatively independently rather than as a single cohesive process. Given the absence of a control group and the combined delivery of the three intervention components, findings are interpreted as descriptive evidence of within-group change rather than confirmation of a causal effect or of any individual component's specific contribution. Qualitative findings indicated that participants perceived the biomimetic model as helpful for understanding abstract concepts and experienced the learning process as more accessible, while the co-design component was associated, in participants’ accounts, with greater autonomy and active participation. As an exploratory pilot study, these results suggest Hercules VR may offer a promising pedagogical approach for inclusive digital design education in similar higher-education contexts, though controlled, multi-institutional research is needed before broader replicability claims can be made. The framework aligns with Sustainable Development Goal 4 (Quality Education).
Richard P. Sánchez Sánchez
3 weeks 1 day ago
BackgroundHandwriting is central to children's written-language development, but conventional assessment usually emphasizes the completed product and provides limited information about the processes that produced it. Digital technologies can capture handwriting as an unfolding motor and orthographic activity.ObjectiveThis scoping review mapped technologies for capturing children's handwriting process data, the raw signals and derived features obtained, the assessment constructs supported by process measures, and their educational, developmental, and clinical applications.MethodsFollowing scoping-review guidance and PRISMA-ScR, we searched Scopus, PubMed, ERIC, Web of Science Core Collection, and IEEE Xplore and conducted backward citation searching of 11 relevant review or framework articles. Of 198 reports sought for full-text assessment, one could not be retrieved, 197 were assessed, 39 were excluded, and 158 were included. All included reports were charted with the same 16-field framework. Independent verification by a second researcher covered randomly selected samples of 40 of 197 full-text eligibility decisions (20.3%; 97.5% agreement; Cohen's κ = 0.918) and 32 of 158 extractions (20.3%).ResultsDigitizing or graphics tablets were used in 136 reports (86.1%), and 117 (74.1%) used Wacom hardware. Position or trajectory (145, 91.8%) and time or timestamps (124, 78.5%) were the most common raw signals, followed by pen pressure or normal force (82, 51.9%) and pen-state or in-air information (71, 44.9%). Common derived feature families included speed or velocity (97, 61.4%), temporal or duration measures (84, 53.2%), spatial or geometric measures (67, 42.4%), and pressure- or force-derived measures (59, 37.3%). Developmental or normative characterization (75, 47.5%) and clinical characterization or differential assessment (74, 46.8%) were the most frequent applications. Latin-alphabetic handwriting appeared in 114 reports (72.2%), followed by Hebrew (23, 14.6%), Chinese (19, 12.0%), and Arabic (5, 3.2%).ConclusionsDigital handwriting assessment is well established in process capture but less consistently translated into applied educational or clinical decisions. Classification performance should not be equated with diagnostic validity. A theory-informed and review-refined four-level framework organizes raw signals, derived features, assessment constructs, and empirical interpretation or action without implying a validated developmental or psychometric hierarchy.
Siyuan Liu