Generative artificial intelligence (AI) and adaptive learning technologies are expanding in higher education, raising questions about their alignment with pedagogical goals. This qualitative-integrative review synthesizes 57 publications (21 empirical, 25 conceptual or framework, 9 systematic reviews or meta-analyses retained for triangulation, and 2 design/development papers) from 2024 to 2026, employing Whittemore and Knafl’s methodology to examine AI-supported tutoring, implementation strategies, and learner/instructor experiences. Three interpretive themes organize this heterogeneous evidence: (1) personalization as a recurring target, from micro-level adaptation to curriculum design; (2) technology-pedagogy co-design as a recurring implementation pattern; and (3) contextual and ethical conditions surrounding adoption, including culture, policy gaps, well-being, and academic integrity. These themes are interpretive, not a validated framework, and draw on evidence of uneven strength. The present review does not estimate effect sizes or pool outcomes across studies; g = 0.70, reported by X. Wang et al. in a separate meta-analysis of 45 studies (2010–2022) of pre-generative adaptive systems, which is cited only as background, not as a finding of this review. Implications are framed as evidence-tiered hypotheses, while the layered model is offered as an interpretive scaffold for future research, not a confirmatory model.
Effective educational feedback is vital for enhancing student learning, yet providing personalized formative feedback in large-class settings remains challenging. Generative Artificial Intelligence (GAI) offers a potential means of scaling formative feedback. Guided by a self-determination theory-based model, this study examines the features, implementation, and pre-service teachers’ perceptions of generative AI-assisted formative feedback (GAI-FF) at a university in China. Participants ( n = 114) received four types of GAI-FF, varying in complexity and focus, across three feedback intervals during a semester. Feedback content and written reflections were analyzed using content analysis, epistemic network analysis, and thematic analysis. Results indicate that (1) GAI can generate formative feedback with distinct cognitive and affective feature patterns; (2) overall implementation of GAI-FF was limited; (3) feedback containing problem identification, suggestions, and solutions was more likely to be enacted, whereas feedback emphasizing affective features was less frequently implemented; and (4) participants expressed mixed perceptions, recognizing the usefulness of GAI-FF for collaborative learning while also noting constraints related to rigidity, accuracy, and practical applicability. These findings suggest that the effectiveness of GAI-FF depends on how it supports learners’ cognitive, motivational, and affective appraisals, which shape subsequent feedback use. The study also highlights the importance of deliberate instructional design and cautious integration of GAI tools in formative assessment practices.
Open-ended computational thinking (CT) inquiry challenges one teacher to support learners progressing at different rates. This clustered quasi-experiment evaluated TALPer, a generative-AI tutor providing student-initiated, artifact-contingent Socratic prompts during a six-lesson 6E unit. Ten intact Grade 3 classes in Taiwan (N = 229) completed the same curriculum, tasks, teacher, platform, and instructional time. Conditions differed in a bundled scaffolding configuration involving source, delivery modality, initiation, privacy, and artifact requirements. A two-level model controlling for pretest CT estimated an adjusted posttest difference favoring TALPer ( B = 4.05, SE = 1.58, Wald 95% CI [0.94, 7.15], p = .011; sensitivity p = .034; Hedges’ g = 0.52). Class-clustered analyses found no Holm-adjusted difference on any Intrinsic Motivation Inventory subscale; Perceived Choice had the largest positive estimate ( B = 0.27, cluster-adjusted SE = 0.12, 95% CI [−0.02, 0.55], p = .061, pHolm = .306; Hedges’ g = 0.38). The Condition × Pretest interaction was nonsignificant ( B = −0.109, SE = 0.125, 95% CI [-0.354, 0.137], p = .385). Descriptive coding of 101 matched responses showed different distributions of expected and planned AI-use orientations; no inferential transition test was conducted. Findings provide preliminary evidence concerning the bundled intervention but do not isolate component contributions.
Traditional micro-teaching offers limited opportunities for pre-service teachers to rehearse classroom contingencies in repeatable, standardized ways. This study developed a multi-agent micro-teaching simulation (MAMS), informed by professional noticing, to expand practice-based learning opportunities for pre-service mathematics teachers (PSTs). MAMS integrates differentiated virtual students, standardized challenge events, and Dean Agent prompts to create comparable classroom contingencies. Sixty-four PSTs completed two simulated micro-teaching sessions and were analytically classified into higher and lower baseline mathematics pedagogical content knowledge (MPCK) profiles. Outcomes included teacher self-efficacy, MPCK performance, and Community of Inquiry presence; interaction logs were analyzed using event-anchored lag sequential analysis. Under the standardized MAMS condition, outcome analyses showed observed pre–post changes and modest baseline-adjusted profile differences. The higher baseline MPCK profile showed stronger MPCK response-strategy performance and higher Teaching Presence; Social and Cognitive Presence showed no reliable profile differences. Log analyses revealed divergent feedback-to-action pathways: higher baseline MPCK PSTs more often moved from Dean Agent prompts to diagnostic questioning and scaffolding, followed by substantive student responses, whereas lower baseline MPCK PSTs more often shifted to direct instruction or process regulation, followed by continued low participation. These findings suggest a need for prior-knowledge-responsive scaffolding in AI-supported teacher education.
The metaverse provides new opportunities for designing learning activities in Information Systems (IS) education. This study integrated a pedagogical approach with the metaverse and examined its effectiveness on learning achievement, critical thinking awareness, collaborative tendencies, and creative thinking skills. A total of 50 university students ( M age = 19.4 years) participated in this quasi-experimental study. The experimental group engaged in a peer-assessment-based metaverse AICDR (Asking, Investigating, Creating, Discussing, and Reflecting) inquiry task (PA-MIT), while the control group followed a conventional metaverse AICDR inquiry task (C-MIT) over an 11-week intervention. Data were analyzed using ANCOVA to compare group differences in learning achievement, critical thinking awareness, and collaboration tendency, while epistemic network analysis (ENA) was applied to explore creative thinking patterns. Results showed that the PA-MIT group achieved significantly higher learning achievement and greater critical thinking awareness than the C-MIT group. However, no significant difference was found in collaboration tendencies. ENA further revealed stronger links among originality, flexibility, and fluency in the PA-MIT group, while the C-MIT group demonstrated strengths in elaboration, flexibility, and fluency. Results highlighted the role of the PA-MIT approach in fostering originality-oriented creative thinking. Peer feedback and idea exchange may lead to the interconnectedness of higher frequency originality-related discourse.
