ABSTRACT. Scientific sensemaking, the process of constructing mechanistic explanations to resolve perceived gaps or conflicts, is a central goal of science education reform. Yet students frequently remain at the descriptive level, and teachers lack tools to scaffold individual sensemaking trajectories. This doctoral research investigates whether an intelligent tutoring agent grounded in sensemaking theory can diagnose students' sensemaking phases and deliver adaptive scaffolding that redirects LLM behavior from answer-giving to sensemaking support. I present SenseHelper, a four-agent system operationalizing Odden and Russ's (2019) framework and Chen et al.'s (2023) four-phase trajectory into a diagnosis-scaffolding loop. Preliminary evaluation shows SenseHelper achieved zero direct concept revelations across 10 dialogue rounds, whereas a baseline agent revealed key concepts six or more times. Students interacting with SenseHelper independently proposed five core scientific concepts versus two for the baseline. I describe the research questions, contribution, methodology, and planned future work for this doctoral study.

ABSTRACT. Recent advances in generative artificial intelligence (GenAI) have enabled the creation of realistic images directly from textual descriptions, opening new possibilities for multimodal language learning. While current AI-assisted writing systems primarily provide textual feedback on grammar and vocabulary, they rarely exploit generated image feedback to support learners' understanding of meaning, content development, and self-expression. This doctoral research proposes an adaptive multimodal English writing support system that generates image feedback from learner compositions and investigates how such feedback influences writing development among English as a Foreign Language (EFL) learners. Building upon preliminary work that demonstrated positive effects of generated image feedback on learner engagement and revision behavior, the proposed research aims to (1) develop an AI-assisted writing environment integrating textual and image feedback, (2) investigate methods for visualizing language errors without directly correcting learner output, and (3) personalize image-based feedback according to learners' writing styles and proficiency. Through a series of classroom-based experiments involving Japanese secondary school students, the project seeks to establish theoretical and empirical foundations for image feedback in AI-supported language learning.

ABSTRACT. Generative Artificial Intelligence (GenAI) has become an important tool for counselor training by providing interactive, scalable, and psychologically safe environments for simulated counseling practice. However, current large language model (LLM)-based virtual clients frequently exhibit overly agreeable conversational behaviors, showing limited resistance, ambivalence, or emotional variability. Such interactions may encourage routine rather than adaptive empathy strategies, raising concerns about whether counseling skills acquired through GenAI simulations can achieve cross-context behavioral transfer. Moreover, previous studies have primarily relied on self-report measures, providing limited objective evidence of behavioral transfer during counselor–client interactions.This study investigates whether an Anti-Over-Accommodation Mechanism embedded in GenAI virtual clients facilitates the cross-context behavioral transfer of empathy strategies among novice counselors. An exploratory sequential mixed-methods design is employed. In Phase 1, interaction transcripts from 15 counseling students are qualitatively analyzed to develop a bidirectional behavioral coding framework. In Phase 2, a randomized controlled trial compares 60 novice counseling trainees interacting with GenAI virtual clients with and without the proposed mechanism. Counseling transcripts are coded using the developed framework, and Lag Sequential Analysis (LSA) is employed to examine sequential relationships between counselors' empathy strategies and client responses across counseling scenarios.The proposed mechanism is expected to promote more adaptive counselor–client interactions and strengthen the cross-context behavioral transfer of empathy strategies. By combining an AI interaction design mechanism with objective behavioral sequence analysis, this study contributes empirical evidence for AI-supported counselor education and offers design implications for developing adaptive and pedagogically effective GenAI-based counseling systems.

ABSTRACT. Artificial intelligence (AI) is becoming common in learning, but it may also take over some of the decision-making and regulatory activities that young learners need to develop. This doctoral research proposes a learner agency–system autonomy alignment framework in which AI control and support are calibrated to learners’ developing capacity to regulate learning. Two preliminary classroom studies suggest that self-regulated learning (SRL) support and interaction designs that require learners to explain, evaluate, and correct can help preserve regulatory engagement. The proposed research will develop process indicators of learner agency, use them to design an adaptive AI learning companion, and evaluate learning during and after AI support. The project aims to explain how system autonomy can be aligned with learner agency and SRL and to provide design principles for AI support that fades or withdraws as learners become better able to regulate their learning independently.

ABSTRACT. Generative artificial intelligence (GenAI) is increasingly used in recent times for university academic writing, yet most current systems can either scaffold learning or complete substantial parts of a writing task for the learner. This doctoral research investigates how an AI tutor can be designed to support international students' academic writing while preserving learner agency and authorship. Using Design-Based Research (DBR), the study develops and iteratively evaluates an AI writing tutor informed by process-oriented writing pedagogy, the Zone of Proximal Development (ZPD), socio-cultural theory. Rather than generating complete academic responses, the tutor guides learners through audience and genre identification, planning, section-by-section drafting, revision, editing, and reflection. An expert-evaluation phase has informed refinements concerning process-oriented scaffolding, academic literacy, feedback, usability, and responsible AI use. The current student phase examines international students' writing experiences, existing AI practices, perceptions of AI support, and interactions with the developed tutor through a background survey, observation, and semi-structured interviews. Reflexive Thematic Analysis will examine where pedagogically structured tutoring aligns with, or conflicts with, students' expectations and established GenAI practices. The anticipated contribution is a set of empirically informed design principles for AI tutoring that balance scaffolding, learner agency, efficiency, and meaningful academic writing development.

ABSTRACT. Abstract: Computer-Supported Collaborative Learning (CSCL) relies on complex socio-metacognitive interactions, yet its pedagogical efficacy is frequently hindered by the high cognitive burden of teacher orchestration. While generative AI and "Simulated Students" offer theoretical solutions to overcome this bottleneck, current Large Language Model (LLM) agents suffer from an "alignment tax", autocorrecting cognitive flaws and failing to preserve authentic learning difficulties. To address this critical gap, this doctoral research proposes a Data-Grounded Cognitive (DGC) simulation paradigm. The research is structured across three interconnected phases. First, we validate the DGC architecture's capacity to suppress the alignment tax, generating high-fidelity multimodal logs for the formative evaluation of instructional designs. Second, we deploy the simulation as a "Human-on-the-loop" diagnostic sandbox, demonstrating that a "double-loop recursive mechanism" enhances teachers' design self-efficacy. Finally, we propose extending this out-of-loop architecture into an in-the-loop digital facilitator to regulate dynamic group interactions. This work provides a universal methodology to fulfill the historical applications of simulated students, bridging the theoretical pedagogy of CSCL with generative AI implementations.

ABSTRACT. Artificial intelligence (AI) is becoming increasingly important in higher education (HE), with universities using it in teaching, research, administration, decision-making, and other institutional activities. However, adoption of AI tools does not translate to conditions that an institution is prepared to achieve efficient and effective AI transformation. It also requires the alignment of people, technology and data, organizational processes, and governance. This study therefore aims to examine the motivations of public higher education institutions (PHEIs) to pursue AI into the capabilities needed to integrate, expand, and sustain its use across institutional functions. Guided by Sociotechnical Systems Theory (STS), the study focuses on the role of socio-technical alignment in linking institutional motivation with AI transformation capability. It also considers how internal and external conditions, such as leadership, resources, policies, regulations, and other institutional pressures, may influence this process. A multiphase mixed-methods design will be employed. Interviews and focus group discussions will first be conducted to identify and refine the key dimensions of the study constructs. These findings will guide the development of a survey instrument for knowledgeable informants from multiple PHEIs. Quantitative data will then be analyzed to test the proposed relationships, followed by additional interviews to explain significant, weak, or unexpected findings. The study is expected to contribute to theory by extending discussions of AI in HE from adoption and individual use toward an institution-level explanation of AI transformation. Methodologically, it seeks to develop and validate measures of institutional motivation, socio-technical alignment, and AI transformation capability. In practice, the resulting framework may help PHEIs identify gaps in people, technology, processes, and governance and guide policies, investments, and capacity-building initiatives for coordinated, responsible, ethical and sustainable AI transformation.

ABSTRACT. In special needs education context, understanding students’ sustained attention is important for adapting instruction to individual needs. However, existing approaches for assessing sustained attention often rely on laboratory-based tasks or specialized equipment, which may limit their use in regular classroom settings. Learning analytics provides opportunities to utilize data generated during everyday learning activities to understand students’ learning processes. Among these data, digital handwriting logs can capture students’ writing behaviors and may provide insights into their attention-related patterns without requiring additional sensors or specialized equipment. The research explores handwriting behaviors of students with special needs and their association with attention states and develops an automated approach for detecting sustained attention. It further examines how attention-related information provided through a dashboard can support teachers to adjust their instructional responses. Together, this research investigates how digital handwriting logs collected from everyday learning activities can provide insights into students’ sustained attention to support teachers in special needs education.

ABSTRACT. Generative artificial intelligence (GenAI) is increasingly transforming educational practices, yet existing research has focused primarily on student learning, with limited attention to how GenAI reshapes teachers' own professional development. This gap is particularly significant for graduate teachers in regional and geographically isolated contexts, where limited access to mentoring and professional networks creates additional challenges for early-career development. This study investigates how graduate teachers in Australian regional secondary schools integrate GenAI chatbots into their professional learning ecologies. Drawing on a constructivist grounded theory approach, this study examines longitudinal qualitative data from eight graduate secondary teachers in regional Australian schools. Through iterative coding and constant comparative analysis, the study developed an AI-mediated Teacher Professional Learning Ecology Framework. Findings reveal that GenAI chatbots functions not merely as a technological resource but as an emerging ecological actor that reshapes teacher learning through three interconnected mechanisms: digital scaffolding, hybrid human–AI support ecologies, and cognitive filtering.

ABSTRACT. The integration of Generative Artificial Intelligence (GenAI) into mathematics courses has elevated the pedagogical challenge, moving beyond routine calculation to the collaborative management of human-AI partnerships. In general education mathematics courses, such as Mathematics in the Modern World (MMW), traditional assessment models frequently collapse as students utilize Large Language Models (LLMs) to completely bypass internal cognitive struggle. Grounded in Lev Vygotsky’s Sociocultural Theory, this work-in-progress paper introduces a Taxonomy of Student-AI Collaboration designed to engineer "AI-resilient" math assessments that preserve the student's Zone of Proximal Development (ZPD). This study investigates how the structural openness of an assessment task—contrasting closed procedural problems with open-ended mathematical modeling workflows—functions as a mechanism for digital scaffolding. This paper presents conceptual framework used to operationalize the design across core MMW modules and outline a mixed-methods action research methodology to evaluate how task design prevents the premature offloading of computation and promotes independent mathematical internalization.