AI-Supported Feedback as Assessment for Learning: Learner Agency, Trust, and Ethical Sensemaking in a Postgraduate Music Education Context

AI-Supported Feedback as Assessment for Learning: Learner Agency, Trust, and Ethical Sensemaking in a Postgraduate Music Education Context | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article AI-Supported Feedback as Assessment for Learning: Learner Agency, Trust, and Ethical Sensemaking in a Postgraduate Music Education Context Chamil Arkhasa Nikko Mazlan, Muhammad Atiullah Othman This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8554027/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The integration of artificial intelligence (AI) into educational assessment has intensified ethical and pedagogical debates concerning learner agency, trust, and responsibility. While AI-enabled systems are frequently promoted for efficiency and personalization, their use in assessment contexts raises concerns that extend beyond technical performance into questions of judgment, power, and ethical governance. Framed within an Assessment for Learning (AfL) perspective, this qualitative interpretive study examines how postgraduate music education students make sense of an AI-based Assessment and Feedback Assistant embedded within a Master’s-level course at a Malaysian public university. Drawing on reflective forum posts produced as part of routine coursework, the study explores how learners articulate perceptions of usefulness, negotiate trust, assert agency, and establish ethical boundaries around AI-supported feedback. Data was analysed using reflexive thematic analysis, informed by concepts of AfL, feedback literacy, trust, and human-centred AI ethics. The findings indicate that students do not experience AI as an authoritative assessor, but as a provisional and dialogic resource that supports reflective sense-making when embedded within an AfL-oriented pedagogical design. Trust in AI emerged as conditional and negotiated, grounded in alignment with human judgment and contextual relevance rather than technological authority. The study argues that ethical and trustworthy AI use in assessment is enacted through pedagogical governance rather than guaranteed by system design alone. By foregrounding reflection, transparency, and learner agency, AfL-oriented environments can enable AI to function as support for assessment literacy without displacing human responsibility. The study contributes empirical insight into human-centred approaches to AI ethics in assessment, particularly within qualitative, judgment-intensive disciplines such as music education. Artificial intelligence in education Assessment for Learning ethical AI feedback literacy learner agency music education reflective practice trustworthy AI Introduction The rapid integration of artificial intelligence (AI) into educational contexts has intensified longstanding debates surrounding assessment ethics, learner agency, and institutional responsibility. Within higher education, AI-enabled systems are increasingly promoted as solutions to challenges associated with efficiency, personalisation, and scalability, particularly through adaptive learning platforms, automated feedback tools, and data-driven instructional support (Mohamed, 2025; Kaswan et al., 2024; Monika Singh et al., 2024; Portilla et al., 2025). These developments are commonly positioned as pragmatic responses to learner diversity, staff workload pressures, and demands for flexible delivery. Yet as AI systems move from instructional support into assessment-related practices, their implications extend well beyond questions of technical performance or instructional effectiveness. Instead, they raise ethical, relational, and pedagogical concerns that cannot be resolved through technological design alone (Marques-Cobeta, 2024; Miranda, 2025). Assessment has never been a neutral pedagogical mechanism. Rather, it is a socially embedded practice through which judgment, power, trust, and moral responsibility are exercised, often with lasting consequences for learner identity, participation, and equity (Pollard & Filer, 2009; Elwood, 2013). Decisions concerning what constitutes valid evidence, whose judgments are legitimised, and how performance is interpreted are inherently normative, shaped by institutional priorities, disciplinary traditions, and broader social values. Ethical tensions frequently emerge when assessment practices are positioned at the intersection of accountability demands and commitments to learner development, inclusion, and fairness (Pope et al., 2009; Thompson & Erickson, 2025). Within such conditions, trust has been identified as a foundational requirement for principled assessment, influencing whether learners are willing to engage honestly, disclose uncertainty, and act productively on feedback (Carless, 2009; Dolan et al., 2019). Assessment for Learning (AfL) provides a particularly useful framework for examining these tensions. AfL is commonly conceptualised as a formative approach in which assessment is used primarily to promote learning rather than to certify achievement, with emphasis placed on feedback, learner participation, and ongoing sense-making (Weeden & Simmons, 2017; Davison, 2013; Akib et al., 2016). Rather than positioning assessment as a terminal judgment, AfL foregrounds processes through which learners develop understanding of quality, monitor their progress, and make informed judgments about their own work. Research has consistently associated AfL practices with enhanced learner agency, motivation, and confidence, particularly when students are actively involved in self- and peer-assessment processes (Akib et al., 2016; Arnold & Holden, 2024). Beyond its pedagogical function, AfL can also be understood as an ethical and relational practice. Learning-oriented assessment depends on environments in which learners feel sufficiently safe to expose uncertainty, receive critique, and experiment with improvement without fear of punitive consequences (Carless, 2009; Dolan et al., 2019). Studies of assessment reform suggest that when trust is undermined, through excessive surveillance, high-stakes grading, or opaque evaluative criteria, innovative assessment practices become constrained regardless of their theoretical strength (Carless, 2009; Sutherland-Harris et al., 2025). In this sense, AfL operates not merely as a collection of techniques, but as a form of pedagogical governance that shapes how responsibility, authority, and agency are distributed within learning environments. Central to this governance is feedback and reflection. Feedback does not function as information transmitted unproblematically from expert to novice; rather, it becomes educationally meaningful only through learners’ active interpretation, judgment, and use (Carless, 2017; Dolan et al., 2019). The concept of feedback literacy captures this interpretive dimension by emphasising learners’ capacities to understand evaluative criteria, judge feedback quality, and act on feedback over time (Carless, 2017). Reflection, in turn, extends beyond retrospective description to function as a judgment-based practice through which learners interrogate experience, align action with professional values, and refine disciplinary reasoning (de Sousa Nunes Vieira et al., 2024). In postgraduate education, reflective practices are particularly significant because they mediate the formation of professional identity, ethical awareness, and evaluative judgment. Within this broader assessment landscape, ethical discussions of AI in education have increasingly focused on issues of fairness, transparency, accountability, and trustworthiness. Much of this literature has approached ethics from a system-centred perspective, prioritising algorithmic bias, explainability, and governance frameworks for responsible deployment (Li, 2025; Thalji & Alkhasawneh, 2025). While these contributions are essential, they often privilege technological and regulatory considerations, offering limited insight into how ethical AI use is enacted within everyday pedagogical practice. Human-centred approaches to AI ethics argue that ethical engagement cannot be reduced to system compliance alone but must attend to how AI reshapes human judgment, agency, and relational dynamics within educational settings (Mohamed, 2025; Marques-Cobeta, 2024). From this perspective, ethical engagement with AI in assessment contexts emerges not solely from technological safeguards but through pedagogical design. When AI is positioned as a support for learner sense-making rather than as an authority that issues judgments, it becomes possible to integrate AI tools into Assessment for Learning practices without displacing human responsibility. How learners experience, interpret, and negotiate this position remains under-explored empirically, particularly in disciplines such as music education where assessment relies heavily on qualitative judgment, contextual interpretation, and reflective reasoning. This study addresses this gap by examining how postgraduate music education students articulate their experiences of using an AI-based Assessment and Feedback Assistant within an explicitly framed AfL design. Rather than evaluating the accuracy or efficiency of AI-generated feedback, the study focuses on learners’ reflective sense-making: how students describe usefulness, negotiate trust, assert agency, and articulate ethical boundaries around AI use. Through analysis of reflective forum posts produced within a Master’s-level course on assessment in music education, the study seeks to illuminate how ethical and trustworthy AI use is shaped through pedagogical governance, rather than technological prescription alone. Conceptual Framework This study is situated at the intersection of Assessment for Learning, feedback literacy, and human-centred approaches to ethical AI in education. Rather than treating artificial intelligence as an autonomous instructional agent, the conceptual framework positions AI as a pedagogically mediated tool whose educational and ethical implications are shaped by assessment design, learner engagement, and institutional norms. The framework therefore foregrounds relational and interpretive processes through which AI is taken up within learning-oriented assessment practices. Assessment for Learning as Pedagogical Governance Assessment for Learning (AfL) provides the foundational lens for this study. AfL is not understood merely as a collection of formative techniques but as a coherent orientation toward assessment that prioritises learning, sense-making, and learner participation (Weeden & Simmons, 2017; Davison, 2013). Central to AfL is the idea that assessment functions productively when learners develop understanding of quality, engage with feedback, and exercise judgment in relation to their own work (Carless, 2017). In this respect, AfL redistributes authority within assessment processes, shifting emphasis from teacher-controlled evaluation toward shared responsibility and learner agency. Importantly, AfL also operates as a form of pedagogical governance. Decisions about task design, feedback practices, and opportunities for reflection shape how power, responsibility, and accountability are negotiated within learning environments (Carless, 2009). When assessment systems emphasise transparency, dialogue, and formative use of evidence, learners are more likely to engage actively and ethically with assessment processes. Conversely, when assessment is experienced as opaque or punitive, trust is eroded and learner agency is constrained, regardless of the tools employed. Within this framework, AI does not determine assessment practice; rather, its role is conditioned by the AfL-oriented design within which it is embedded. AI tools may amplify, support, or undermine AfL principles depending on how they are positioned in relation to human judgment and learner participation. Feedback Literacy and Reflective Sense-Making Feedback literacy forms a second core component of the conceptual framework. Feedback literacy refers to learners’ capacities to interpret, evaluate, and use feedback productively over time, rather than merely receiving feedback as information (Carless, 2017). This perspective reframes feedback as an active, judgment-based process that requires learners to make sense of criteria, assess the credibility and relevance of feedback, and decide how to act upon it. Reflection plays a central role in this process. Reflective practice enables learners to connect feedback with experience, values, and professional goals, supporting deeper learning and the development of evaluative judgment (de Sousa Nunes Vieira et al., 2024). In postgraduate contexts, reflection is particularly significant because it mediates the formation of professional identity and ethical reasoning. Reflection thus becomes not only a learning strategy but also an ethical practice through which learners articulate boundaries, responsibilities, and standards of quality. Within this study, students’ reflective accounts are treated as sites of sense-making where feedback literacy is enacted. Rather than assessing whether AI-generated feedback is “correct,” the framework attends to how learners interpret usefulness, negotiate trust, and assert agency in relation to AI-supported feedback. Trust, Agency, and Ethical Engagement in Assessment Trust is a critical relational condition underpinning both AfL and feedback literacy. Trust influences whether learners feel able to disclose uncertainty, engage critically with feedback, and take responsibility for improvement (Carless, 2009; Dolan et al., 2019). In assessment contexts, trust is shaped by transparency of criteria, clarity of purpose, and perceptions of fairness and care. Ethical engagement with assessment therefore emerges through relationships rather than rules alone. Research on ethical dilemmas in assessment highlights that tensions often arise when institutional accountability pressures conflict with commitments to learner development and inclusion (Pope et al., 2009; Elwood, 2013). AfL-oriented environments seek to mitigate these tensions by foregrounding learning, dialogue, and shared responsibility. Within this framework, learner agency is not defined as independence from guidance, but as the capacity to exercise informed judgment within supportive structures. AI tools, when introduced into assessment, may either support or constrain this agency depending on whether they are perceived as authoritative evaluators or as resources for reflection and sense-making. Human-Centred AI in Assessment Contexts Ethical discussions of AI in education have increasingly emphasised system-centred principles such as fairness, transparency, explainability, and accountability (Li, 2025). While these principles are essential, a human-centred perspective highlights that ethical AI use is ultimately realised through pedagogical practice. AI systems do not operate in isolation; their ethical implications are mediated by how learners and educators interpret, trust, and act upon AI-generated outputs (Mohamed, 2025). From a human-centred perspective, ethical AI in assessment is not achieved solely through technical safeguards but through pedagogical design that preserves human judgment and responsibility. When AI is positioned as a dialogic support rather than an evaluative authority, it can be integrated into AfL practices without displacing ethical accountability. Conversely, when AI outputs are treated as definitive judgments, they risk undermining trust, agency, and reflective engagement. Music Education as Contextual Anchor Music education provides a particularly salient context for examining these dynamics. Assessment in music education frequently relies on qualitative judgment, contextual interpretation, and reflective reasoning, rather than solely on standardised metrics. As such, issues of trust, transparency, and ethical judgment are especially pronounced. The integration of AI into assessment within this discipline therefore raises critical questions about how technological tools intersect with professional judgment and pedagogical values. Within this study, music education is not treated as an exceptional case but as a context that renders visible broader tensions inherent in AI-supported assessment. By examining postgraduate music education students’ reflective accounts, the framework illuminates how ethical and trustworthy AI use is shaped through pedagogical governance, feedback literacy, and relational trust. Methodology This study is situated within a qualitative, interpretive research tradition grounded in reflective inquiry. Its purpose is not to evaluate the technical performance of an artificial intelligence (AI) system, nor to measure learning outcomes in a causal or experimental sense. Instead, the study examines how learners make sense of AI-supported assessment within an Assessment for Learning (AfL) framework, with particular attention to issues of usefulness, trust, agency, and ethical responsibility. Learners’ reflective accounts are therefore treated as primary sites through which pedagogical, relational, and ethical meanings are articulated. Qualitative approaches are especially appropriate for investigating assessment practices understood as socially embedded and ethically situated phenomena. Assessment involves judgment, interpretation, and negotiation of values, processes that cannot be meaningfully reduced to quantitative indicators alone (Elwood, 2013; Pollard & Filer, 2009). By focusing on reflective writing produced within authentic assessment contexts, this study seeks to capture how learners interpret feedback, negotiate authority, and articulate ethical boundaries when engaging with AI-supported tools. The study is informed by principles of Assessment for Learning, which emphasise the formative use of evidence, learner participation, and reflective sense-making rather than summative measurement or certification (Weeden & Simmons, 2017; Davison, 2013). Within this orientation, reflection is not treated as supplementary data, but as a central pedagogical and methodological resource through which evaluative judgment and ethical reasoning become visible. Importantly, AI is not conceptualised as an independent variable exerting causal influence on learners. Rather, it is understood as a mediating artefact embedded within a specific pedagogical design. This positioning aligns with human-centred approaches to AI ethics, which argue that ethical implications emerge through use, interpretation, and governance rather than through technical properties alone (Mohamed, 2025; Li, 2025). Context and pedagogical design The study was conducted within a Master’s-level course on assessment in music education offered at a Malaysian public university. The course is designed to develop students’ assessment literacy, with a particular emphasis on formative assessment principles, feedback practices, and reflective evaluation of pedagogical decision-making. Participants were primarily in-service and pre-service music educators teaching in primary and secondary schools, as well as in specialised contexts such as military music education. This professional positioning is analytically significant, as it situates participants not only as postgraduate learners but also as practitioners engaged in the development of assessment practices for real-world teaching contexts. The course explicitly adopts an Assessment for Learning orientation. Assessment activities are designed to foreground process, judgment, and improvement rather than grading or product evaluation alone. Students engage in ongoing reflective tasks intended to support conceptual understanding of assessment, ethical considerations, and application to authentic teaching situations. Within this pedagogical framework, an AI-based Assessment and Feedback Assistant was introduced as a support tool rather than an evaluative authority. Students were permitted to engage with the AI tool only after submitting their reflective writing to the learning forum. This sequencing was deliberately designed to preserve authorship, authenticity, and learner responsibility, while enabling students to use AI-generated feedback as a secondary input for sense-making. Explicit guidance was provided to ensure ethical use, including clear statements that AI should not be used to generate reflective content, but only to support understanding of assessment criteria, feedback processes, and reflective quality. AI was therefore positioned as dialogic and provisional rather than authoritative, embedded within AfL governance structures that prioritise trust, learner agency, and human responsibility (Carless, 2009; Dolan et al., 2019). Music education provides a particularly relevant disciplinary context for this inquiry. Assessment in music frequently relies on qualitative judgment, contextual interpretation, and reflective evaluation, rendering issues of trust, transparency, and ethical responsibility especially salient. The integration of AI into this context thus offers a productive lens for examining broader tensions surrounding AI-supported assessment without reducing assessment to technical scoring or optimisation. Participants The participants comprised 18 postgraduate students enrolled in the course during the semester in which the study was conducted. All students engaged in the learning activities from which the data were drawn. Participation in the research was non-intrusive and pedagogically embedded; students were not recruited into a separate intervention, and no additional research tasks were introduced beyond normal course requirements. This approach aligns with ethical guidance for educational research that emphasises minimal disruption to learning and avoidance of coercion in assessed contexts (Elwood, 2013). The relatively small cohort size is appropriate for qualitative inquiry focused on depth of interpretation rather than statistical generalisation. The study does not seek representativeness across populations but rather examines how learners articulate experiences of AI-supported assessment within a specific pedagogical design. Given participants’ professional engagement with assessment in their own teaching contexts, their reflections offer conceptually rich insights into issues of judgment, agency, and ethical responsibility. Data sources The primary data source consisted of written reflective forum posts produced by students as part of the course’s Assessment for Learning design. These reflections were submitted in response to structured prompts inviting students to reflect on assessment practices and, subsequently, on their experiences of using an AI-based Assessment and Feedback Assistant. Students were required to submit their initial reflections independently, without AI assistance. Only after submission where they permitted to engage with the AI tool to receive formative feedback related to clarity, alignment with assessment criteria, and depth of reflection. Following interaction with the AI tool, students were invited, on a voluntary basis, to post additional reflective comments in the forum addressing the perceived usefulness, limitations, trustworthiness, and ethical implications of AI-supported feedback. These reflective commentaries constitute the core dataset analysed in this study. In addition to forum reflections, a secondary source of qualitative data was obtained through a voluntary Google Form completed by 17 participants. The form invited students to reflect explicitly on similar dimensions, including usefulness of AI feedback, understanding of assessment rubrics, impact on reflective quality, and perceived challenges or ethical concerns. These responses were treated not as survey data but as short-form reflective narratives. Analysed alongside forum posts, they served to corroborate, extend, and refine emerging interpretive themes. Reflective writing is an established data source in qualitative research on assessment, feedback literacy, and professional learning. Reflection enables access to learners’ interpretive processes, revealing how individuals make sense of feedback, articulate values, and negotiate responsibility (Carless, 2017; de Sousa Nunes Vieira et al., 2024). In this study, both forum reflections and Google Form responses are treated as naturally occurring pedagogical data, enhancing ecological validity by capturing sense-making as it unfolds within authentic assessment conditions. Analytic approach Data was analysed using reflexive thematic analysis informed by an interpretive qualitative orientation. The analysis did not seek to quantify frequencies of opinion or evaluate system effectiveness, but to examine how students articulated meaning around AI-supported assessment within an AfL context. Analytic attention focused on how participants described usefulness, negotiated trust, asserted agency, and articulated ethical boundaries in relation to AI-generated feedback. The analysis followed an iterative, multi-stage process. All reflective texts were read repeatedly to develop familiarity with the dataset, with attention to evaluative language, expressions of judgment, and ethical positioning. Initial codes were generated inductively, focusing on meaning units rather than isolated keywords. Codes were then examined for patterns of convergence and tension and grouped into candidate themes that captured recurring ways in which students made sense of AI within the pedagogical design. Themes were refined through iterative comparison with the dataset to ensure conceptual coherence and analytic depth. Reflexive memos were maintained throughout to document analytic decisions and surface researcher assumptions. Although primarily inductive, the analysis was theoretically sensitised by concepts drawn from Assessment for Learning, feedback literacy, trust, and human-centred AI ethics. Theory functioned as an interpretive lens rather than a coding template, providing a vocabulary for articulating patterns emerging from the data without constraining analytic openness. Ethical considerations Ethical considerations were addressed with particular attention to the dual role of the researcher as course instructor and investigator, the use of assessed learning activities as data, and the integration of AI within assessment-related practices. The data analysed were drawn from naturally occurring pedagogical activities, and no additional research interventions were introduced. Students were informed that anonymised reflections might be used for research purposes, and participation in reflective discussions about AI use was voluntary. Students were assured that non-participation would have no impact on grades or academic standing. Anonymity and confidentiality were maintained through removal of identifying information and the use of pseudonyms. Data was stored securely and accessed only by the researcher. Ethical use of AI was an explicit component of both the pedagogical design and the research ethics. AI was framed as a supportive, non-authoritative tool, and students were not monitored or evaluated on their AI use. Ethical engagement with AI was examined through students’ own reflective articulations rather than through surveillance or compliance checking, aligning with AfL principles that prioritise trust and learner agency (Carless, 2009). Finally, the study recognises assessment itself as an ethical practice. By foregrounding reflection, transparency, and dialogue around AI use, the research seeks not only to investigate ethical issues but also to enact ethically responsive assessment practices within the course. Findings The analysis of postgraduate students’ reflective accounts reveals a coherent yet nuanced picture of how AI-supported feedback was experienced within an Assessment for Learning (AfL) design. Across both forum reflections and Google Form responses, participants did not frame the AI-based Assessment and Feedback Assistant as an authoritative assessor. Instead, they positioned it as a provisional resource whose value depended on learners’ judgment, contextual knowledge, and ethical restraint. Five interrelated themes capture how students made sense of AI-supported assessment in practice. AI as a Clarifier of Criteria Rather Than an Evaluator A dominant theme across participants’ reflections was the use of AI as a tool for clarifying assessment expectations rather than evaluating performance. Students consistently described the AI assistant as helping them understand rubrics, criteria, and levels of achievement in more concrete and accessible terms. Rather than accepting feedback as verdicts, learners used AI output to interpret what constituted quality work within the course’s assessment framework. Many participants noted that AI feedback helped them “see” assessment criteria more clearly, particularly in relation to abstract dimensions such as depth of reflection, alignment between theory and practice, and clarity of argument. For several students, this clarification supported self-assessment by making implicit expectations more explicit. Importantly, this function did not replace students’ own evaluative judgment. Instead, AI feedback operated as a reference point against which students compared their own interpretations of the rubric. This pattern was evident across both longer forum reflections and shorter Google Form responses. Even when students expressed strong appreciation for AI feedback, they framed its usefulness in interpretive terms, emphasising understanding rather than decision-making. AI was valued for illuminating criteria, not for determining grades or judgments. Reflective Deepening Through Dialogic Feedback Participants frequently described AI-supported feedback as prompting deeper reflection rather than final answers. Rather than treating feedback as corrective instruction, students engaged with it dialogically, using suggestions and questions generated by the AI to re-examine their own assumptions, arguments, and professional experiences. Several students reported that AI feedback encouraged them to move beyond descriptive reflection towards more analytical and evaluative writing. This shift was often framed as a process of refinement rather than transformation. Students did not describe AI as producing better reflections for them, but as helping them recognise gaps, tensions, or underdeveloped reasoning within their own work. This dialogic engagement was particularly evident among participants who were practising teachers. These students frequently described adapting AI feedback to align with their specific teaching contexts, such as rural schools, ensemble-based instruction, or military music settings. In doing so, they asserted the primacy of contextual judgment over generic feedback, treating AI output as a stimulus for reflection rather than an endpoint. Agency Maintained Through Selective Acceptance and Adaptation A striking feature of the dataset is the extent to which participants articulated active control over how AI feedback was used. Rather than expressing dependency, students consistently emphasised the need to filter, adapt, and sometimes reject AI-generated suggestions. This selective engagement functioned as a form of agency, reinforcing learners’ responsibility for judgment rather than undermining it. Participants described several strategies for maintaining agency. These included cross-checking AI feedback against course requirements, aligning suggestions with personal teaching experience, and revising outputs to ensure authenticity of voice. In some cases, students explicitly cautioned against over-reliance on AI, noting that feedback could be overly general or insufficiently sensitive to disciplinary and contextual nuances. This theme was reinforced in Google Form responses, where students frequently acknowledged both benefits and limitations in the same reflection. The coexistence of appreciation and caution suggests that agency was not experienced as threatened by AI use, but as something that required conscious enactment. AI-supported assessment was therefore framed as a space of negotiation rather than submission. Trust as Pedagogically Framed Rather Than Technically Assumed Trust emerged as a relational and pedagogical issue rather than a technical one. Participants did not describe trust in AI as inherent or automatic. Instead, trust was contingent upon how the AI tool was introduced, constrained, and positioned within the assessment design. Students repeatedly referenced the importance of clear guidance regarding ethical use. Knowing that AI was intended as a support tool rather than an assessor contributed to learners’ willingness to engage with it critically. The sequencing of reflection before AI use was particularly salient, as it reassured students that authorship and responsibility remained with them. At the same time, participants expressed conditional trust in AI output. Trust was extended to AI feedback when it aligned with students’ own judgments or clarified existing understanding. Conversely, when feedback appeared disconnected from context or overly generic, trust was withheld. This conditionality suggests that trust was exercised as judgment rather than granted by default. Ethical Boundaries and Professional Responsibility Across the dataset, participants demonstrated explicit ethical awareness in relation to AI-supported assessment. Students did not treat ethical considerations as abstract policy concerns but as practical questions embedded in everyday academic and professional practice. Many reflections emphasised the importance of preserving originality, avoiding dependency, and ensuring that final decisions remained human. Several participants extended this ethical reasoning to their future roles as educators. They expressed interest in adapting AI tools for their own assessment practices while simultaneously asserting that AI should not replace teacher judgment. This forward-looking orientation indicates that ethical engagement with AI was not limited to the course context but integrated into participants’ broader professional identities. Ethical reflection was particularly pronounced among participants working in high-responsibility contexts, such as military education or high-stakes school environments. These students framed AI as a potential support for efficiency and consistency, but only under conditions of human oversight, contextual sensitivity, and ethical restraint. Summary of Findings Taken together, the findings indicate that AI-supported feedback, when embedded within an Assessment for Learning design, was experienced as a reflective resource rather than an evaluative authority. Students used AI to clarify criteria, deepen reflection, and interrogate their own judgments, while maintaining agency and ethical responsibility. Trust in AI was not assumed but constructed through pedagogical framing, sequencing, and explicit ethical guidance. Rather than displacing human judgment, AI engagement in this context appeared to foreground it. Ethical and trustworthy use of AI emerged not from technical safeguards alone, but from pedagogical governance that positioned learners as responsible agents in assessment practices. Discussion This study set out to examine how postgraduate music education students made sense of AI-supported assessment within an explicitly framed Assessment for Learning (AfL) design. Rather than evaluating AI accuracy or efficiency, the analysis foregrounded learners’ reflective articulations of usefulness, trust, agency, and ethical responsibility. The findings suggest that ethical and trustworthy engagement with AI in assessment contexts is not primarily a function of technological sophistication, but of pedagogical governance, on how AI is positioned, constrained, and interpreted within learning-oriented assessment practices. Across participants’ reflections, AI was consistently described as supportive but provisional. Students valued AI-generated feedback for clarifying assessment criteria, improving structural coherence, and prompting deeper reflection. At the same time, they resisted positioning AI as an authoritative judge of quality. This selective engagement aligns closely with AfL principles, which emphasise learner participation, evaluative judgment, and feedback as a dialogic rather than transmissive process (Weeden & Simmons, 2017; Davison, 2013; Carless, 2017). In this sense, AI functioned less as an assessor and more as a mediating artefact that enabled learners to engage more actively with criteria, standards, and reflective judgment. Importantly, students’ accounts demonstrate that usefulness was not equated with correctness. AI feedback was valued precisely because it could be questioned, adapted, or rejected. This finding complicates dominant narratives of AI efficiency and automation in education, which often frame usefulness in terms of speed, consistency, or objectivity (Mohamed, 2025; Kaswan et al., 2024; Monika Singh et al., 2024). Instead, the present study suggests that usefulness in assessment contexts is relational and interpretive, emerging through learners’ capacity to evaluate feedback against their own professional knowledge, contextual realities, and ethical commitments. This reinforces critiques of system-centred approaches to AI ethics that prioritise algorithmic properties over lived pedagogical practice (Marques-Cobeta, 2024; Li, 2025). Trust emerged as a central condition shaping students’ engagement with AI-supported feedback. However, trust was neither unconditional nor naïve. Students expressed confidence in AI when it aligned with their own judgments and course expectations, while simultaneously articulating caution regarding over-reliance, contextual mismatch, and potential erosion of professional responsibility. This calibrated trust reflects a sophisticated ethical stance: learners acknowledged AI’s affordances while maintaining human accountability for assessment decisions. Such positioning resonates with research on trust in assessment reform, which emphasises that trust is sustained not through surveillance or control, but through transparency, dialogue, and shared responsibility (Carless, 2009; Dolan et al., 2019; Sutherland-Harris et al., 2025). The AfL-oriented design of the course appears to have been instrumental in enabling this ethical positioning. By sequencing AI use after initial submission and framing AI explicitly as non-authoritative, the pedagogical design preserved learner authorship and reduced the risk of dependency. Students’ reflections indicate that this design supported agency rather than undermining it. Rather than deferring to AI judgments, learners used AI feedback as a resource for reflection, comparison, and refinement. These findings challenge deterministic accounts that frame AI as inherently displacing human judgment, suggesting instead that displacement is a consequence of design choices rather than technological inevitability (Mohamed, 2025; Li, 2025). Feedback literacy provides a useful lens for interpreting these findings. Students demonstrated emerging capacities to interpret feedback, judge its relevance, and decide how or whether to act on it. AI-supported feedback did not replace these processes but appeared to scaffold them, particularly in relation to abstract criteria such as depth of reflection and theoretical alignment. However, the need for careful adaptation and contextualisation was repeatedly emphasised, underscoring that feedback literacy remains a human capability that cannot be automated (Carless, 2017; Dolan et al., 2019). AI may amplify feedback availability, but it does not resolve the interpretive work required to make feedback meaningful. Reflection itself functioned as both a pedagogical and ethical practice within the study. Students’ reflections moved beyond surface description to consider responsibility, authorship, and professional integrity. Several participants explicitly articulated boundaries around ethical AI use, rejecting the idea that AI should generate reflective content or replace teacher judgment. These reflections suggest that ethical engagement with AI is not merely a matter of compliance with institutional policy, but of developing ethical discernment through practice. In this respect, reflection operates as a site where ethical reasoning is enacted rather than merely declared (Elwood, 2013; de Sousa Nunes Vieira et al., 2024). The disciplinary context of music education further sharpens the significance of these findings. Assessment in music education relies heavily on qualitative judgment, contextual interpretation, and professional discretion. Students’ insistence that final assessment decisions should remain with teachers reflects an acute awareness of the limits of standardisation and automation in such contexts. Rather than resisting AI outright, participants articulated a nuanced vision in which AI supports assessment literacy while human educators retain interpretive authority. This disciplinary sensitivity highlights the risks of importing generic AI assessment models into contexts where judgment cannot be meaningfully reduced to metrics (Pollard & Filer, 2009; Thompson & Erickson, 2025). Taken together, the findings suggest that ethical AI use in assessment is less about regulating systems and more about cultivating pedagogical conditions that support agency, trust, and reflective judgment. Human-centred AI ethics, when operationalised through AfL design, shifts the focus from what AI can do to how learners and educators ought to engage with it. This reframing has implications for both research and practice. For researchers, it underscores the need to study AI-in-use rather than AI-in-the-abstract. For educators, it highlights the importance of designing assessment environments that foreground dialogue, transparency, and ethical responsibility. At the same time, this study does not suggest that AI use in assessment is unproblematic. Students’ reflections surfaced tensions related to over-generality, contextual misalignment, and the risk of dependency. These tensions should not be viewed as failures, but as productive sites for ethical learning. Indeed, the presence of ambivalence may be a marker of ethical engagement rather than uncertainty or resistance. By making these tensions visible through reflection, AfL-oriented designs can support learners in developing critical relationships with AI rather than uncritical acceptance or rejection (Carless, 2009; Dolan et al., 2019). In sum, this study contributes to emerging conversations on trustworthy AI in education by demonstrating that trustworthiness is enacted through pedagogy rather than guaranteed by technology. When AI is embedded within learning-oriented assessment designs that preserve authorship, encourage reflection, and distribute responsibility transparently, it can support rather than undermine ethical assessment practice. The challenge for future work lies not in perfecting AI systems, but in refining the pedagogical and ethical frameworks through which they are engaged. Conclusion and Implications This study sets out to explore how postgraduate music education students experienced and interpreted the use of an AI-based Assessment and Feedback Assistant within an explicitly framed Assessment for Learning (AfL) design. Rather than approaching AI as a technical intervention whose value could be judged in terms of efficiency or accuracy, the study examined how ethical, pedagogical, and relational meanings were constructed through learners’ reflective engagement. The findings demonstrate that ethical and trustworthy AI use in assessment is not inherent in the technology itself, but is enacted through pedagogical design, learner agency, and reflective judgment. The analysis shows that students did not position AI as an authoritative assessor. Instead, they treated AI feedback as provisional, dialogic, and open to contestation. This positioning is significant because it counters dominant efficiency-driven narratives that frame AI as a solution to assessment workload or consistency. In contrast, participants’ reflections indicate that the value of AI lay in its capacity to prompt sense-making: clarifying criteria, supporting structural coherence, and stimulating reflective questioning. Usefulness, in this context, was defined not by correctness or automation, but by how well AI feedback could be interpreted, adapted, or rejected in relation to professional judgment and contextual realities. Trust emerged as a negotiated and conditional practice rather than a fixed disposition. Students expressed confidence in AI when it aligned with course expectations and their own evaluative reasoning, while simultaneously articulating caution about over-reliance and decontextualised feedback. This calibrated trust reflects a mature ethical stance in which responsibility for assessment decisions remained firmly human. Rather than eroding agency, the AfL-oriented design appeared to strengthen it by making the limits of AI visible and by preserving authorship through deliberate sequencing and explicit ethical guidance. The findings also highlight the central role of feedback literacy and reflection in mediating AI use. Students demonstrated increasing capacity to interpret feedback, judge its relevance, and decide how to act on it, capabilities that cannot be automated. Reflection functioned not merely as a reporting mechanism but as a site of ethical reasoning, where boundaries around acceptable AI use were articulated and negotiated. In this sense, ethical engagement with AI was enacted through practice rather than compliance, reinforcing the view that ethics in assessment is relational and situated rather than procedural. The disciplinary context of music education further underscores the importance of these insights. Assessment in music education relies heavily on qualitative judgment, contextual interpretation, and professional discretion. Participants’ insistence that final assessment decisions should remain with teachers reflects an acute awareness of the limits of standardisation and algorithmic judgment in such contexts. This suggests that AI-supported assessment models must be sensitive to disciplinary epistemologies rather than applied generically across fields. Implications for Practice For educators, the study highlights the importance of pedagogical governance when integrating AI into assessment practices. AI tools should be positioned as supportive resources rather than evaluative authorities, with clear guidance that preserves learner authorship, agency, and responsibility. Sequencing matters: requiring students to produce work independently before engaging with AI feedback helps prevent dependency and reinforces ethical use. Embedding AI within AfL designs that emphasise dialogue, reflection, and transparency can transform AI from a risk to assessment integrity into a resource for developing assessment literacy. Implications for Research For researchers, the findings suggest a need to shift attention away from AI performance metrics and towards AI-in-use within authentic pedagogical contexts. Qualitative, interpretive approaches are particularly well suited to examining how ethical and trustworthy AI use is enacted through learners’ sense-making and reflective judgment. Future research might explore how such designs function across different disciplines, levels of study, or institutional contexts, and how tensions around trust and agency evolve over time. Limitations and Future Directions This study is situated within a single Master’s-level course with a relatively small cohort, and the findings are not intended to be statistically generalisable. Instead, they offer theoretically informed insight into how ethical AI engagement can be pedagogically cultivated. Future studies could extend this work through longitudinal designs, comparative disciplinary analyses, or by examining educators’ perspectives alongside students’ reflections. Concluding Remarks Overall, this study contributes to emerging discussions on trustworthy and ethical AI in education by demonstrating that trustworthiness is not guaranteed by technology but is constructed through pedagogical design and reflective practice. When AI is embedded within Assessment for Learning frameworks that foreground agency, trust, and ethical responsibility, it can support rather than undermine principled assessment practices. The challenge ahead lies not in perfecting AI systems, but in designing learning environments that enable educators and learners to engage with AI critically, responsibly, and humanely. Declarations Ethics Approval and Consent to Participate This study used anonymised data derived from routine pedagogical activities in a postgraduate course. Students were informed that anonymised excerpts might be used for research and publication. Participation was voluntary and had no impact on assessment outcomes. Informed consent was obtained from all participants. Competing Interests The authors declare that they have no known financial or non-financial competing interests that could have influenced the work reported in this paper. Funding This research received no direct research funding. Open access publication was supported through the Springer Nature transformative agreement facilitated by the Konsortium Sumber Elektronik Pendidikan Tinggi (KONSEPt), with institutional support from Sultan Idris Education University, Malaysia. Author Contribution C.A.N.M. conceptualised the study, designed the methodology, conducted the analysis, and drafted the manuscript. M.A.O. contributed to conceptual refinement, critical review of the manuscript, and provided scholarly feedback throughout the writing process. Both authors approved the final version of the manuscript. Acknowledgement The author would like to acknowledge the postgraduate students enrolled in the Master’s-level course on assessment in music education (cohort M251PM2) for their thoughtful engagement and reflective contributions, without which this study would not have been possible. Their willingness to critically articulate their experiences, professional judgments, and ethical considerations provided valuable insight into the pedagogical enactment of Assessment for Learning and the responsible use of artificial intelligence in educational assessment. The author also acknowledges the institutional support of Universiti Pendidikan Sultan Idris, within which the course was designed and conducted. The study benefited from an academic environment that encourages reflective practice, ethical inquiry, and pedagogical innovation in music education. The author further acknowledges the support provided by the Konsortium Sumber Elektronik Pendidikan Tinggi (KONSEPt) and the participating institutions for facilitating access to scholarly resources and enabling open access publication. Gratitude is also extended to Springer Nature for supporting open access initiatives through the transformative agreement with KONSEPt. Finally, the author acknowledges the broader scholarly community whose work on Assessment for Learning, feedback literacy, trust in assessment, and human-centred approaches to artificial intelligence informed the conceptual framing of this study. Data Availability The datasets generated and analysed during the current study are not publicly available due to ethical considerations and the use of assessed student work but are available from the corresponding author on reasonable request. References Akib, I., Akib, E., Ghafar, M. N. A., & Latif, A. A. (2016). Measuring assessment for learning. Man in India, 96 (1–2), 267–277. https://www.researchgate.net/publication/298639911_Measuring_assessment_for_learning Arnold, J., & Holden, M. (2024). Designing inclusive Assessment for Learning in English Language Arts. In Designing Inclusive Assessment in Schools (pp. 27–38). Routledge. https://doi.org/10.4324/9781003463184-5 Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3 (2), 77–101. https://doi.org/10.1191/1478088706qp063oa Braun, V., & Clarke, V. (2021). One size fits all? What counts as quality practice in reflexive thematic analysis? Qualitative Research in Psychology, 18 (3), 328–352. https://doi.org/10.1080/14780887.2020.1769238 Carless, D. (2009). Trust, distrust and their impact on assessment reform. Assessment & Evaluation in Higher Education, 34 (1), 79–89. https://doi.org/10.1080/02602930801895786 Carless, D. (2017). Scaling up assessment for learning: Progress and prospects. In The Enabling Power of Assessment (pp. 3–17). Springer. https://doi.org/10.1007/978-981-10-3045-1_1 Davison, C. (2013). Innovation in assessment: Common misconceptions and problems. In Innovation and Change in English Language Education (pp. 263–276). Routledge. https://doi.org/10.4324/9780203362716-29 de Sousa Nunes Vieira, M., da Silva Campos Costa, N. M., & Pereira, E. R. S. (2024). Assessment of learning in health higher education. Medicina (Ribeirão Preto), 57 (2), e-214928. https://doi.org/10.11606/issn.2176-7262.rmrp.2024.214928 Dolan, B. M., Arnold, J., & Green, M. M. (2019). Establishing trust when assessing learners: Barriers and opportunities. Academic Medicine, 94 (12), 1851–1853. https://doi.org/10.1097/ACM.0000000000002982 Elwood, J. (2013). Educational assessment policy and practice: A matter of ethics. Assessment in Education: Principles, Policy & Practice, 20 (2), 205–220. https://doi.org/10.1080/0969594X.2013.765384 Kaswan, K. S., Dhatterwal, J. S., & Ojha, R. P. (2024). AI in personalized learning. In Advances in Technological Innovations in Higher Education (pp. 103–117). CRC Press. https://doi.org/10.1201/9781003376699-9 Li, S. (2025). Trustworthy AI meets educational assessment: Challenges and opportunities. Proceedings of the AAAI Conference on Artificial Intelligence, 39 (27), 28637–28642. https://doi.org/10.1609/aaai.v39i27.35089 Marques-Cobeta, N. (2024). Artificial intelligence in education: Unveiling opportunities and challenges. Lecture Notes in Educational Technology , 33–42. https://doi.org/10.1007/978-981-97-2468-0_4 Miranda, S. (2025). Artificial intelligence in education: An exploratory survey. Education Sciences, 15 (8), 975. https://doi.org/10.3390/educsci15080975 Mohamed, L. K. (2025). Implementation of AI in education: Promises and challenges. In Education, Future Jobs and Smart Systems in the Age of Artificial Intelligence (pp. 51–66). Emerald. https://doi.org/10.1108/978-1-83608-904-920251004 Monika Singh, T., Reddy, C. K. K., Murthy, B. V. R., Nag, A., & Doss, S. (2024). AI and education: Bridging the gap to personalized, efficient, and accessible learning. In Internet of Behavior-Based Computational Intelligence for Smart Education Systems (pp. 131–160). IGI Global. https://doi.org/10.4018/979-8-3693-8151-9.ch005 Pollard, A., & Filer, A. (2009). Social practices in school assessment and their impact on learner identities. In International Encyclopedia of Education (3rd ed., pp. 512–517). Elsevier. https://doi.org/10.1016/B978-0-08-044894-7.00300-6 Pope, N., Green, S. K., Johnson, R. L., & Mitchell, M. (2009). Examining teacher ethical dilemmas in classroom assessment. Teaching and Teacher Education, 25 (5), 778–782. https://doi.org/10.1016/j.tate.2008.11.013 Portilla, J. E. N., Cedeño, J. K. Z., Jácome, G. O. L., & Gallegos, L. A. M. (2025). Artificial intelligence in Education 4.0: A systematic review. Journal of Educators Online, 22 (3). https://doi.org/10.9743/JEO.2025.22.3.13 Sutherland-Harris, R., Ali, A., & Elkhoury, E. (2025). Practicing trust between academic developers and faculty for equitable assessments. International Journal for Academic Development, 30 (1), 132–136. https://doi.org/10.1080/1360144X.2024.2445617 Thalji, N. J., & Alkhasawneh, S. (2025). How can artificial intelligence shape the future of sustainable education? Journal of Theoretical and Applied Information Technology, 103 (9), 3836–3850. Thompson, W. C., & Erickson, J. D. (2025). Voice, recognition, and exit: A rights-based response to cultural bias in educational assessments. In Culturally Responsive Assessment (pp. 52–69). Routledge. https://doi.org/10.4324/9781003392217-5 Weeden, P., & Simmons, M. (2017). Formative assessment. In Debates in Geography Education (pp. 140–155). Routledge. https://doi.org/10.4324/9781315562452 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8554027","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":580901182,"identity":"33883397-a246-4481-a549-9e7d2362c6a2","order_by":0,"name":"Chamil Arkhasa Nikko Mazlan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA50lEQVRIiWNgGAWjYBACxgYwZQHEzAeAhIQMsVokgJgtAcTgIdYykBYeAxCLsBbm9uOPP3zcIyEnH5Hz+dWNGgseBvbDRzfgdVhPQoLhjGcSxoY3crdZ5xwDOownLe0GXi0NCQeSeQ5IJG6ckbvNOIcNqEWCxwy/lv6HDYf/gLXkPDPO+UeMlhnJjM0MQC3zJXKYH+e2EaXlGTNjzwEJYwOeZ2bMuX0SPGyE/GLYn/74w48DNnLy7cmPP+d8q5PjZz98DL+WBijD4EICGyh2GNjwKQcBeTij/wDzB0KqR8EoGAWjYGQCAG0xSRdgFaCCAAAAAElFTkSuQmCC","orcid":"","institution":"Sultan Idris Education University","correspondingAuthor":true,"prefix":"","firstName":"Chamil","middleName":"Arkhasa Nikko","lastName":"Mazlan","suffix":""},{"id":580901185,"identity":"96b6fa99-4c11-45a1-b359-99e430259ce4","order_by":1,"name":"Muhammad Atiullah Othman","email":"","orcid":"","institution":"Sultan Idris Education University","correspondingAuthor":false,"prefix":"","firstName":"Muhammad","middleName":"Atiullah","lastName":"Othman","suffix":""}],"badges":[],"createdAt":"2026-01-08 17:23:28","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8554027/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8554027/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104300314,"identity":"fe9c3717-4a40-4055-9155-5e451e407a08","added_by":"auto","created_at":"2026-03-10 08:43:19","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":482018,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8554027/v1/5ff57dd9-710f-45b6-be97-f3f66e459fcd.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"AI-Supported Feedback as Assessment for Learning: Learner Agency, Trust, and Ethical Sensemaking in a Postgraduate Music Education Context","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe rapid integration of artificial intelligence (AI) into educational contexts has intensified longstanding debates surrounding assessment ethics, learner agency, and institutional responsibility. Within higher education, AI-enabled systems are increasingly promoted as solutions to challenges associated with efficiency, personalisation, and scalability, particularly through adaptive learning platforms, automated feedback tools, and data-driven instructional support (Mohamed, 2025; Kaswan et al., 2024; Monika Singh et al., 2024; Portilla et al., 2025). These developments are commonly positioned as pragmatic responses to learner diversity, staff workload pressures, and demands for flexible delivery. Yet as AI systems move from instructional support into assessment-related practices, their implications extend well beyond questions of technical performance or instructional effectiveness. Instead, they raise ethical, relational, and pedagogical concerns that cannot be resolved through technological design alone (Marques-Cobeta, 2024; Miranda, 2025).\u003c/p\u003e\n\u003cp\u003eAssessment has never been a neutral pedagogical mechanism. Rather, it is a socially embedded practice through which judgment, power, trust, and moral responsibility are exercised, often with lasting consequences for learner identity, participation, and equity (Pollard \u0026amp; Filer, 2009; Elwood, 2013). Decisions concerning what constitutes valid evidence, whose judgments are legitimised, and how performance is interpreted are inherently normative, shaped by institutional priorities, disciplinary traditions, and broader social values. Ethical tensions frequently emerge when assessment practices are positioned at the intersection of accountability demands and commitments to learner development, inclusion, and fairness (Pope et al., 2009; Thompson \u0026amp; Erickson, 2025). Within such conditions, trust has been identified as a foundational requirement for principled assessment, influencing whether learners are willing to engage honestly, disclose uncertainty, and act productively on feedback (Carless, 2009; Dolan et al., 2019).\u003c/p\u003e\n\u003cp\u003eAssessment for Learning (AfL) provides a particularly useful framework for examining these tensions. AfL is commonly conceptualised as a formative approach in which assessment is used primarily to promote learning rather than to certify achievement, with emphasis placed on feedback, learner participation, and ongoing sense-making (Weeden \u0026amp; Simmons, 2017; Davison, 2013; Akib et al., 2016). Rather than positioning assessment as a terminal judgment, AfL foregrounds processes through which learners develop understanding of quality, monitor their progress, and make informed judgments about their own work. Research has consistently associated AfL practices with enhanced learner agency, motivation, and confidence, particularly when students are actively involved in self- and peer-assessment processes (Akib et al., 2016; Arnold \u0026amp; Holden, 2024).\u003c/p\u003e\n\u003cp\u003eBeyond its pedagogical function, AfL can also be understood as an ethical and relational practice. Learning-oriented assessment depends on environments in which learners feel sufficiently safe to expose uncertainty, receive critique, and experiment with improvement without fear of punitive consequences (Carless, 2009; Dolan et al., 2019). Studies of assessment reform suggest that when trust is undermined, through excessive surveillance, high-stakes grading, or opaque evaluative criteria, innovative assessment practices become constrained regardless of their theoretical strength (Carless, 2009; Sutherland-Harris et al., 2025). In this sense, AfL operates not merely as a collection of techniques, but as a form of pedagogical governance that shapes how responsibility, authority, and agency are distributed within learning environments.\u003c/p\u003e\n\u003cp\u003eCentral to this governance is feedback and reflection. Feedback does not function as information transmitted unproblematically from expert to novice; rather, it becomes educationally meaningful only through learners\u0026rsquo; active interpretation, judgment, and use (Carless, 2017; Dolan et al., 2019). The concept of feedback literacy captures this interpretive dimension by emphasising learners\u0026rsquo; capacities to understand evaluative criteria, judge feedback quality, and act on feedback over time (Carless, 2017). Reflection, in turn, extends beyond retrospective description to function as a judgment-based practice through which learners interrogate experience, align action with professional values, and refine disciplinary reasoning (de Sousa Nunes Vieira et al., 2024). In postgraduate education, reflective practices are particularly significant because they mediate the formation of professional identity, ethical awareness, and evaluative judgment.\u003c/p\u003e\n\u003cp\u003eWithin this broader assessment landscape, ethical discussions of AI in education have increasingly focused on issues of fairness, transparency, accountability, and trustworthiness. Much of this literature has approached ethics from a system-centred perspective, prioritising algorithmic bias, explainability, and governance frameworks for responsible deployment (Li, 2025; Thalji \u0026amp; Alkhasawneh, 2025). While these contributions are essential, they often privilege technological and regulatory considerations, offering limited insight into how ethical AI use is enacted within everyday pedagogical practice. Human-centred approaches to AI ethics argue that ethical engagement cannot be reduced to system compliance alone but must attend to how AI reshapes human judgment, agency, and relational dynamics within educational settings (Mohamed, 2025; Marques-Cobeta, 2024).\u003c/p\u003e\n\u003cp\u003eFrom this perspective, ethical engagement with AI in assessment contexts emerges not solely from technological safeguards but through pedagogical design. When AI is positioned as a support for learner sense-making rather than as an authority that issues judgments, it becomes possible to integrate AI tools into Assessment for Learning practices without displacing human responsibility. How learners experience, interpret, and negotiate this position remains under-explored empirically, particularly in disciplines such as music education where assessment relies heavily on qualitative judgment, contextual interpretation, and reflective reasoning.\u003c/p\u003e\n\u003cp\u003eThis study addresses this gap by examining how postgraduate music education students articulate their experiences of using an AI-based Assessment and Feedback Assistant within an explicitly framed AfL design. Rather than evaluating the accuracy or efficiency of AI-generated feedback, the study focuses on learners\u0026rsquo; reflective sense-making: how students describe usefulness, negotiate trust, assert agency, and articulate ethical boundaries around AI use. Through analysis of reflective forum posts produced within a Master\u0026rsquo;s-level course on assessment in music education, the study seeks to illuminate how ethical and trustworthy AI use is shaped through pedagogical governance, rather than technological prescription alone.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eConceptual Framework\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThis study is situated at the intersection of Assessment for Learning, feedback literacy, and human-centred approaches to ethical AI in education. Rather than treating artificial intelligence as an autonomous instructional agent, the conceptual framework positions AI as a pedagogically mediated tool whose educational and ethical implications are shaped by assessment design, learner engagement, and institutional norms. The framework therefore foregrounds relational and interpretive processes through which AI is taken up within learning-oriented assessment practices.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAssessment for Learning as Pedagogical Governance\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAssessment for Learning (AfL) provides the foundational lens for this study. AfL is not understood merely as a collection of formative techniques but as a coherent orientation toward assessment that prioritises learning, sense-making, and learner participation (Weeden \u0026amp; Simmons, 2017; Davison, 2013). Central to AfL is the idea that assessment functions productively when learners develop understanding of quality, engage with feedback, and exercise judgment in relation to their own work (Carless, 2017). In this respect, AfL redistributes authority within assessment processes, shifting emphasis from teacher-controlled evaluation toward shared responsibility and learner agency.\u003c/p\u003e\n\u003cp\u003eImportantly, AfL also operates as a form of pedagogical governance. Decisions about task design, feedback practices, and opportunities for reflection shape how power, responsibility, and accountability are negotiated within learning environments (Carless, 2009). When assessment systems emphasise transparency, dialogue, and formative use of evidence, learners are more likely to engage actively and ethically with assessment processes. Conversely, when assessment is experienced as opaque or punitive, trust is eroded and learner agency is constrained, regardless of the tools employed. Within this framework, AI does not determine assessment practice; rather, its role is conditioned by the AfL-oriented design within which it is embedded. AI tools may amplify, support, or undermine AfL principles depending on how they are positioned in relation to human judgment and learner participation.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eFeedback Literacy and Reflective Sense-Making\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eFeedback literacy forms a second core component of the conceptual framework. Feedback literacy refers to learners\u0026rsquo; capacities to interpret, evaluate, and use feedback productively over time, rather than merely receiving feedback as information (Carless, 2017). This perspective reframes feedback as an active, judgment-based process that requires learners to make sense of criteria, assess the credibility and relevance of feedback, and decide how to act upon it. Reflection plays a central role in this process. Reflective practice enables learners to connect feedback with experience, values, and professional goals, supporting deeper learning and the development of evaluative judgment (de Sousa Nunes Vieira et al., 2024). In postgraduate contexts, reflection is particularly significant because it mediates the formation of professional identity and ethical reasoning. Reflection thus becomes not only a learning strategy but also an ethical practice through which learners articulate boundaries, responsibilities, and standards of quality. Within this study, students\u0026rsquo; reflective accounts are treated as sites of sense-making where feedback literacy is enacted. Rather than assessing whether AI-generated feedback is \u0026ldquo;correct,\u0026rdquo; the framework attends to how learners interpret usefulness, negotiate trust, and assert agency in relation to AI-supported feedback.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eTrust, Agency, and Ethical Engagement in Assessment\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eTrust is a critical relational condition underpinning both AfL and feedback literacy. Trust influences whether learners feel able to disclose uncertainty, engage critically with feedback, and take responsibility for improvement (Carless, 2009; Dolan et al., 2019). In assessment contexts, trust is shaped by transparency of criteria, clarity of purpose, and perceptions of fairness and care. Ethical engagement with assessment therefore emerges through relationships rather than rules alone. Research on ethical dilemmas in assessment highlights that tensions often arise when institutional accountability pressures conflict with commitments to learner development and inclusion (Pope et al., 2009; Elwood, 2013). AfL-oriented environments seek to mitigate these tensions by foregrounding learning, dialogue, and shared responsibility. Within this framework, learner agency is not defined as independence from guidance, but as the capacity to exercise informed judgment within supportive structures. AI tools, when introduced into assessment, may either support or constrain this agency depending on whether they are perceived as authoritative evaluators or as resources for reflection and sense-making.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eHuman-Centred AI in Assessment Contexts\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eEthical discussions of AI in education have increasingly emphasised system-centred principles such as fairness, transparency, explainability, and accountability (Li, 2025). While these principles are essential, a human-centred perspective highlights that ethical AI use is ultimately realised through pedagogical practice. AI systems do not operate in isolation; their ethical implications are mediated by how learners and educators interpret, trust, and act upon AI-generated outputs (Mohamed, 2025). From a human-centred perspective, ethical AI in assessment is not achieved solely through technical safeguards but through pedagogical design that preserves human judgment and responsibility. When AI is positioned as a dialogic support rather than an evaluative authority, it can be integrated into AfL practices without displacing ethical accountability. Conversely, when AI outputs are treated as definitive judgments, they risk undermining trust, agency, and reflective engagement.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eMusic Education as Contextual Anchor\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eMusic education provides a particularly salient context for examining these dynamics. Assessment in music education frequently relies on qualitative judgment, contextual interpretation, and reflective reasoning, rather than solely on standardised metrics. As such, issues of trust, transparency, and ethical judgment are especially pronounced. The integration of AI into assessment within this discipline therefore raises critical questions about how technological tools intersect with professional judgment and pedagogical values. Within this study, music education is not treated as an exceptional case but as a context that renders visible broader tensions inherent in AI-supported assessment. By examining postgraduate music education students\u0026rsquo; reflective accounts, the framework illuminates how ethical and trustworthy AI use is shaped through pedagogical governance, feedback literacy, and relational trust.\u003c/p\u003e"},{"header":"Methodology","content":"\u003cp\u003eThis study is situated within a qualitative, interpretive research tradition grounded in reflective inquiry. Its purpose is not to evaluate the technical performance of an artificial intelligence (AI) system, nor to measure learning outcomes in a causal or experimental sense. Instead, the study examines how learners make sense of AI-supported assessment within an Assessment for Learning (AfL) framework, with particular attention to issues of usefulness, trust, agency, and ethical responsibility. Learners\u0026rsquo; reflective accounts are therefore treated as primary sites through which pedagogical, relational, and ethical meanings are articulated.\u003c/p\u003e\n\u003cp\u003eQualitative approaches are especially appropriate for investigating assessment practices understood as socially embedded and ethically situated phenomena. Assessment involves judgment, interpretation, and negotiation of values, processes that cannot be meaningfully reduced to quantitative indicators alone (Elwood, 2013; Pollard \u0026amp; Filer, 2009). By focusing on reflective writing produced within authentic assessment contexts, this study seeks to capture how learners interpret feedback, negotiate authority, and articulate ethical boundaries when engaging with AI-supported tools.\u003c/p\u003e\n\u003cp\u003eThe study is informed by principles of Assessment for Learning, which emphasise the formative use of evidence, learner participation, and reflective sense-making rather than summative measurement or certification (Weeden \u0026amp; Simmons, 2017; Davison, 2013). Within this orientation, reflection is not treated as supplementary data, but as a central pedagogical and methodological resource through which evaluative judgment and ethical reasoning become visible. Importantly, AI is not conceptualised as an independent variable exerting causal influence on learners. Rather, it is understood as a mediating artefact embedded within a specific pedagogical design. This positioning aligns with human-centred approaches to AI ethics, which argue that ethical implications emerge through use, interpretation, and governance rather than through technical properties alone (Mohamed, 2025; Li, 2025).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eContext and pedagogical design\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe study was conducted within a Master\u0026rsquo;s-level course on assessment in music education offered at a Malaysian public university. The course is designed to develop students\u0026rsquo; assessment literacy, with a particular emphasis on formative assessment principles, feedback practices, and reflective evaluation of pedagogical decision-making. Participants were primarily in-service and pre-service music educators teaching in primary and secondary schools, as well as in specialised contexts such as military music education. This professional positioning is analytically significant, as it situates participants not only as postgraduate learners but also as practitioners engaged in the development of assessment practices for real-world teaching contexts.\u003c/p\u003e\n\u003cp\u003eThe course explicitly adopts an Assessment for Learning orientation. Assessment activities are designed to foreground process, judgment, and improvement rather than grading or product evaluation alone. Students engage in ongoing reflective tasks intended to support conceptual understanding of assessment, ethical considerations, and application to authentic teaching situations. Within this pedagogical framework, an AI-based Assessment and Feedback Assistant was introduced as a support tool rather than an evaluative authority. Students were permitted to engage with the AI tool only after submitting their reflective writing to the learning forum. This sequencing was deliberately designed to preserve authorship, authenticity, and learner responsibility, while enabling students to use AI-generated feedback as a secondary input for sense-making. Explicit guidance was provided to ensure ethical use, including clear statements that AI should not be used to generate reflective content, but only to support understanding of assessment criteria, feedback processes, and reflective quality. AI was therefore positioned as dialogic and provisional rather than authoritative, embedded within AfL governance structures that prioritise trust, learner agency, and human responsibility (Carless, 2009; Dolan et al., 2019).\u003c/p\u003e\n\u003cp\u003eMusic education provides a particularly relevant disciplinary context for this inquiry. Assessment in music frequently relies on qualitative judgment, contextual interpretation, and reflective evaluation, rendering issues of trust, transparency, and ethical responsibility especially salient. The integration of AI into this context thus offers a productive lens for examining broader tensions surrounding AI-supported assessment without reducing assessment to technical scoring or optimisation.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eParticipants\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe participants comprised 18 postgraduate students enrolled in the course during the semester in which the study was conducted. All students engaged in the learning activities from which the data were drawn. Participation in the research was non-intrusive and pedagogically embedded; students were not recruited into a separate intervention, and no additional research tasks were introduced beyond normal course requirements. This approach aligns with ethical guidance for educational research that emphasises minimal disruption to learning and avoidance of coercion in assessed contexts (Elwood, 2013). The relatively small cohort size is appropriate for qualitative inquiry focused on depth of interpretation rather than statistical generalisation. The study does not seek representativeness across populations but rather examines how learners articulate experiences of AI-supported assessment within a specific pedagogical design. Given participants\u0026rsquo; professional engagement with assessment in their own teaching contexts, their reflections offer conceptually rich insights into issues of judgment, agency, and ethical responsibility.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eData sources\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe primary data source consisted of written reflective forum posts produced by students as part of the course\u0026rsquo;s Assessment for Learning design. These reflections were submitted in response to structured prompts inviting students to reflect on assessment practices and, subsequently, on their experiences of using an AI-based Assessment and Feedback Assistant. Students were required to submit their initial reflections independently, without AI assistance. Only after submission where they permitted to engage with the AI tool to receive formative feedback related to clarity, alignment with assessment criteria, and depth of reflection. Following interaction with the AI tool, students were invited, on a voluntary basis, to post additional reflective comments in the forum addressing the perceived usefulness, limitations, trustworthiness, and ethical implications of AI-supported feedback. These reflective commentaries constitute the core dataset analysed in this study.\u003c/p\u003e\n\u003cp\u003eIn addition to forum reflections, a secondary source of qualitative data was obtained through a voluntary Google Form completed by 17 participants. The form invited students to reflect explicitly on similar dimensions, including usefulness of AI feedback, understanding of assessment rubrics, impact on reflective quality, and perceived challenges or ethical concerns. These responses were treated not as survey data but as short-form reflective narratives. Analysed alongside forum posts, they served to corroborate, extend, and refine emerging interpretive themes. Reflective writing is an established data source in qualitative research on assessment, feedback literacy, and professional learning. Reflection enables access to learners\u0026rsquo; interpretive processes, revealing how individuals make sense of feedback, articulate values, and negotiate responsibility (Carless, 2017; de Sousa Nunes Vieira et al., 2024). In this study, both forum reflections and Google Form responses are treated as naturally occurring pedagogical data, enhancing ecological validity by capturing sense-making as it unfolds within authentic assessment conditions.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAnalytic approach\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eData was analysed using reflexive thematic analysis informed by an interpretive qualitative orientation. The analysis did not seek to quantify frequencies of opinion or evaluate system effectiveness, but to examine how students articulated meaning around AI-supported assessment within an AfL context. Analytic attention focused on how participants described usefulness, negotiated trust, asserted agency, and articulated ethical boundaries in relation to AI-generated feedback. The analysis followed an iterative, multi-stage process. All reflective texts were read repeatedly to develop familiarity with the dataset, with attention to evaluative language, expressions of judgment, and ethical positioning. Initial codes were generated inductively, focusing on meaning units rather than isolated keywords. Codes were then examined for patterns of convergence and tension and grouped into candidate themes that captured recurring ways in which students made sense of AI within the pedagogical design. Themes were refined through iterative comparison with the dataset to ensure conceptual coherence and analytic depth. Reflexive memos were maintained throughout to document analytic decisions and surface researcher assumptions. Although primarily inductive, the analysis was theoretically sensitised by concepts drawn from Assessment for Learning, feedback literacy, trust, and human-centred AI ethics. Theory functioned as an interpretive lens rather than a coding template, providing a vocabulary for articulating patterns emerging from the data without constraining analytic openness.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eEthical considerations\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eEthical considerations were addressed with particular attention to the dual role of the researcher as course instructor and investigator, the use of assessed learning activities as data, and the integration of AI within assessment-related practices. The data analysed were drawn from naturally occurring pedagogical activities, and no additional research interventions were introduced. Students were informed that anonymised reflections might be used for research purposes, and participation in reflective discussions about AI use was voluntary. Students were assured that non-participation would have no impact on grades or academic standing.\u003c/p\u003e\n\u003cp\u003eAnonymity and confidentiality were maintained through removal of identifying information and the use of pseudonyms. Data was stored securely and accessed only by the researcher. Ethical use of AI was an explicit component of both the pedagogical design and the research ethics. AI was framed as a supportive, non-authoritative tool, and students were not monitored or evaluated on their AI use. Ethical engagement with AI was examined through students\u0026rsquo; own reflective articulations rather than through surveillance or compliance checking, aligning with AfL principles that prioritise trust and learner agency (Carless, 2009). Finally, the study recognises assessment itself as an ethical practice. By foregrounding reflection, transparency, and dialogue around AI use, the research seeks not only to investigate ethical issues but also to enact ethically responsive assessment practices within the course.\u003c/p\u003e"},{"header":"Findings","content":"\u003cp\u003eThe analysis of postgraduate students\u0026rsquo; reflective accounts reveals a coherent yet nuanced picture of how AI-supported feedback was experienced within an Assessment for Learning (AfL) design. Across both forum reflections and Google Form responses, participants did not frame the AI-based Assessment and Feedback Assistant as an authoritative assessor. Instead, they positioned it as a provisional resource whose value depended on learners\u0026rsquo; judgment, contextual knowledge, and ethical restraint. Five interrelated themes capture how students made sense of AI-supported assessment in practice.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAI as a Clarifier of Criteria Rather Than an Evaluator\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eA dominant theme across participants\u0026rsquo; reflections was the use of AI as a tool for clarifying assessment expectations rather than evaluating performance. Students consistently described the AI assistant as helping them understand rubrics, criteria, and levels of achievement in more concrete and accessible terms. Rather than accepting feedback as verdicts, learners used AI output to interpret what constituted quality work within the course\u0026rsquo;s assessment framework.\u003c/p\u003e\n\u003cp\u003eMany participants noted that AI feedback helped them \u0026ldquo;see\u0026rdquo; assessment criteria more clearly, particularly in relation to abstract dimensions such as depth of reflection, alignment between theory and practice, and clarity of argument. For several students, this clarification supported self-assessment by making implicit expectations more explicit. Importantly, this function did not replace students\u0026rsquo; own evaluative judgment. Instead, AI feedback operated as a reference point against which students compared their own interpretations of the rubric.\u003c/p\u003e\n\u003cp\u003eThis pattern was evident across both longer forum reflections and shorter Google Form responses. Even when students expressed strong appreciation for AI feedback, they framed its usefulness in interpretive terms, emphasising understanding rather than decision-making. AI was valued for illuminating criteria, not for determining grades or judgments.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eReflective Deepening Through Dialogic Feedback\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eParticipants frequently described AI-supported feedback as prompting deeper reflection rather than final answers. Rather than treating feedback as corrective instruction, students engaged with it dialogically, using suggestions and questions generated by the AI to re-examine their own assumptions, arguments, and professional experiences.\u003c/p\u003e\n\u003cp\u003eSeveral students reported that AI feedback encouraged them to move beyond descriptive reflection towards more analytical and evaluative writing. This shift was often framed as a process of refinement rather than transformation. Students did not describe AI as producing better reflections for them, but as helping them recognise gaps, tensions, or underdeveloped reasoning within their own work.\u003c/p\u003e\n\u003cp\u003eThis dialogic engagement was particularly evident among participants who were practising teachers. These students frequently described adapting AI feedback to align with their specific teaching contexts, such as rural schools, ensemble-based instruction, or military music settings. In doing so, they asserted the primacy of contextual judgment over generic feedback, treating AI output as a stimulus for reflection rather than an endpoint.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAgency Maintained Through Selective Acceptance and Adaptation\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eA striking feature of the dataset is the extent to which participants articulated active control over how AI feedback was used. Rather than expressing dependency, students consistently emphasised the need to filter, adapt, and sometimes reject AI-generated suggestions. This selective engagement functioned as a form of agency, reinforcing learners\u0026rsquo; responsibility for judgment rather than undermining it.\u003c/p\u003e\n\u003cp\u003eParticipants described several strategies for maintaining agency. These included cross-checking AI feedback against course requirements, aligning suggestions with personal teaching experience, and revising outputs to ensure authenticity of voice. In some cases, students explicitly cautioned against over-reliance on AI, noting that feedback could be overly general or insufficiently sensitive to disciplinary and contextual nuances.\u003c/p\u003e\n\u003cp\u003eThis theme was reinforced in Google Form responses, where students frequently acknowledged both benefits and limitations in the same reflection. The coexistence of appreciation and caution suggests that agency was not experienced as threatened by AI use, but as something that required conscious enactment. AI-supported assessment was therefore framed as a space of negotiation rather than submission.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eTrust as Pedagogically Framed Rather Than Technically Assumed\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eTrust emerged as a relational and pedagogical issue rather than a technical one. Participants did not describe trust in AI as inherent or automatic. Instead, trust was contingent upon how the AI tool was introduced, constrained, and positioned within the assessment design. Students repeatedly referenced the importance of clear guidance regarding ethical use. Knowing that AI was intended as a support tool rather than an assessor contributed to learners\u0026rsquo; willingness to engage with it critically. The sequencing of reflection before AI use was particularly salient, as it reassured students that authorship and responsibility remained with them. At the same time, participants expressed conditional trust in AI output. Trust was extended to AI feedback when it aligned with students\u0026rsquo; own judgments or clarified existing understanding. Conversely, when feedback appeared disconnected from context or overly generic, trust was withheld. This conditionality suggests that trust was exercised as judgment rather than granted by default.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eEthical Boundaries and Professional Responsibility\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAcross the dataset, participants demonstrated explicit ethical awareness in relation to AI-supported assessment. Students did not treat ethical considerations as abstract policy concerns but as practical questions embedded in everyday academic and professional practice. Many reflections emphasised the importance of preserving originality, avoiding dependency, and ensuring that final decisions remained human.\u003c/p\u003e\n\u003cp\u003eSeveral participants extended this ethical reasoning to their future roles as educators. They expressed interest in adapting AI tools for their own assessment practices while simultaneously asserting that AI should not replace teacher judgment. This forward-looking orientation indicates that ethical engagement with AI was not limited to the course context but integrated into participants\u0026rsquo; broader professional identities.\u003c/p\u003e\n\u003cp\u003eEthical reflection was particularly pronounced among participants working in high-responsibility contexts, such as military education or high-stakes school environments. These students framed AI as a potential support for efficiency and consistency, but only under conditions of human oversight, contextual sensitivity, and ethical restraint.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eSummary of Findings\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eTaken together, the findings indicate that AI-supported feedback, when embedded within an Assessment for Learning design, was experienced as a reflective resource rather than an evaluative authority. Students used AI to clarify criteria, deepen reflection, and interrogate their own judgments, while maintaining agency and ethical responsibility. Trust in AI was not assumed but constructed through pedagogical framing, sequencing, and explicit ethical guidance. Rather than displacing human judgment, AI engagement in this context appeared to foreground it. Ethical and trustworthy use of AI emerged not from technical safeguards alone, but from pedagogical governance that positioned learners as responsible agents in assessment practices.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study set out to examine how postgraduate music education students made sense of AI-supported assessment within an explicitly framed Assessment for Learning (AfL) design. Rather than evaluating AI accuracy or efficiency, the analysis foregrounded learners\u0026rsquo; reflective articulations of usefulness, trust, agency, and ethical responsibility. The findings suggest that ethical and trustworthy engagement with AI in assessment contexts is not primarily a function of technological sophistication, but of pedagogical governance, on how AI is positioned, constrained, and interpreted within learning-oriented assessment practices.\u003c/p\u003e\n\u003cp\u003eAcross participants\u0026rsquo; reflections, AI was consistently described as supportive but provisional. Students valued AI-generated feedback for clarifying assessment criteria, improving structural coherence, and prompting deeper reflection. At the same time, they resisted positioning AI as an authoritative judge of quality. This selective engagement aligns closely with AfL principles, which emphasise learner participation, evaluative judgment, and feedback as a dialogic rather than transmissive process (Weeden \u0026amp; Simmons, 2017; Davison, 2013; Carless, 2017). In this sense, AI functioned less as an assessor and more as a mediating artefact that enabled learners to engage more actively with criteria, standards, and reflective judgment.\u003c/p\u003e\n\u003cp\u003eImportantly, students\u0026rsquo; accounts demonstrate that usefulness was not equated with correctness. AI feedback was valued precisely because it could be questioned, adapted, or rejected. This finding complicates dominant narratives of AI efficiency and automation in education, which often frame usefulness in terms of speed, consistency, or objectivity (Mohamed, 2025; Kaswan et al., 2024; Monika Singh et al., 2024). Instead, the present study suggests that usefulness in assessment contexts is relational and interpretive, emerging through learners\u0026rsquo; capacity to evaluate feedback against their own professional knowledge, contextual realities, and ethical commitments. This reinforces critiques of system-centred approaches to AI ethics that prioritise algorithmic properties over lived pedagogical practice (Marques-Cobeta, 2024; Li, 2025).\u003c/p\u003e\n\u003cp\u003eTrust emerged as a central condition shaping students\u0026rsquo; engagement with AI-supported feedback. However, trust was neither unconditional nor na\u0026iuml;ve. Students expressed confidence in AI when it aligned with their own judgments and course expectations, while simultaneously articulating caution regarding over-reliance, contextual mismatch, and potential erosion of professional responsibility. This calibrated trust reflects a sophisticated ethical stance: learners acknowledged AI\u0026rsquo;s affordances while maintaining human accountability for assessment decisions. Such positioning resonates with research on trust in assessment reform, which emphasises that trust is sustained not through surveillance or control, but through transparency, dialogue, and shared responsibility (Carless, 2009; Dolan et al., 2019; Sutherland-Harris et al., 2025).\u003c/p\u003e\n\u003cp\u003eThe AfL-oriented design of the course appears to have been instrumental in enabling this ethical positioning. By sequencing AI use after initial submission and framing AI explicitly as non-authoritative, the pedagogical design preserved learner authorship and reduced the risk of dependency. Students\u0026rsquo; reflections indicate that this design supported agency rather than undermining it. Rather than deferring to AI judgments, learners used AI feedback as a resource for reflection, comparison, and refinement. These findings challenge deterministic accounts that frame AI as inherently displacing human judgment, suggesting instead that displacement is a consequence of design choices rather than technological inevitability (Mohamed, 2025; Li, 2025).\u003c/p\u003e\n\u003cp\u003eFeedback literacy provides a useful lens for interpreting these findings. Students demonstrated emerging capacities to interpret feedback, judge its relevance, and decide how or whether to act on it. AI-supported feedback did not replace these processes but appeared to scaffold them, particularly in relation to abstract criteria such as depth of reflection and theoretical alignment. However, the need for careful adaptation and contextualisation was repeatedly emphasised, underscoring that feedback literacy remains a human capability that cannot be automated (Carless, 2017; Dolan et al., 2019). AI may amplify feedback availability, but it does not resolve the interpretive work required to make feedback meaningful.\u003c/p\u003e\n\u003cp\u003eReflection itself functioned as both a pedagogical and ethical practice within the study. Students\u0026rsquo; reflections moved beyond surface description to consider responsibility, authorship, and professional integrity. Several participants explicitly articulated boundaries around ethical AI use, rejecting the idea that AI should generate reflective content or replace teacher judgment. These reflections suggest that ethical engagement with AI is not merely a matter of compliance with institutional policy, but of developing ethical discernment through practice. In this respect, reflection operates as a site where ethical reasoning is enacted rather than merely declared (Elwood, 2013; de Sousa Nunes Vieira et al., 2024).\u003c/p\u003e\n\u003cp\u003eThe disciplinary context of music education further sharpens the significance of these findings. Assessment in music education relies heavily on qualitative judgment, contextual interpretation, and professional discretion. Students\u0026rsquo; insistence that final assessment decisions should remain with teachers reflects an acute awareness of the limits of standardisation and automation in such contexts. Rather than resisting AI outright, participants articulated a nuanced vision in which AI supports assessment literacy while human educators retain interpretive authority. This disciplinary sensitivity highlights the risks of importing generic AI assessment models into contexts where judgment cannot be meaningfully reduced to metrics (Pollard \u0026amp; Filer, 2009; Thompson \u0026amp; Erickson, 2025).\u003c/p\u003e\n\u003cp\u003eTaken together, the findings suggest that ethical AI use in assessment is less about regulating systems and more about cultivating pedagogical conditions that support agency, trust, and reflective judgment. Human-centred AI ethics, when operationalised through AfL design, shifts the focus from what AI can do to how learners and educators ought to engage with it. This reframing has implications for both research and practice. For researchers, it underscores the need to study AI-in-use rather than AI-in-the-abstract. For educators, it highlights the importance of designing assessment environments that foreground dialogue, transparency, and ethical responsibility.\u003c/p\u003e\n\u003cp\u003eAt the same time, this study does not suggest that AI use in assessment is unproblematic. Students\u0026rsquo; reflections surfaced tensions related to over-generality, contextual misalignment, and the risk of dependency. These tensions should not be viewed as failures, but as productive sites for ethical learning. Indeed, the presence of ambivalence may be a marker of ethical engagement rather than uncertainty or resistance. By making these tensions visible through reflection, AfL-oriented designs can support learners in developing critical relationships with AI rather than uncritical acceptance or rejection (Carless, 2009; Dolan et al., 2019).\u003c/p\u003e\n\u003cp\u003eIn sum, this study contributes to emerging conversations on trustworthy AI in education by demonstrating that trustworthiness is enacted through pedagogy rather than guaranteed by technology. When AI is embedded within learning-oriented assessment designs that preserve authorship, encourage reflection, and distribute responsibility transparently, it can support rather than undermine ethical assessment practice. The challenge for future work lies not in perfecting AI systems, but in refining the pedagogical and ethical frameworks through which they are engaged.\u003c/p\u003e"},{"header":"Conclusion and Implications","content":"\u003cp\u003eThis study sets out to explore how postgraduate music education students experienced and interpreted the use of an AI-based Assessment and Feedback Assistant within an explicitly framed Assessment for Learning (AfL) design. Rather than approaching AI as a technical intervention whose value could be judged in terms of efficiency or accuracy, the study examined how ethical, pedagogical, and relational meanings were constructed through learners\u0026rsquo; reflective engagement. The findings demonstrate that ethical and trustworthy AI use in assessment is not inherent in the technology itself, but is enacted through pedagogical design, learner agency, and reflective judgment.\u003c/p\u003e\n\u003cp\u003eThe analysis shows that students did not position AI as an authoritative assessor. Instead, they treated AI feedback as provisional, dialogic, and open to contestation. This positioning is significant because it counters dominant efficiency-driven narratives that frame AI as a solution to assessment workload or consistency. In contrast, participants\u0026rsquo; reflections indicate that the value of AI lay in its capacity to prompt sense-making: clarifying criteria, supporting structural coherence, and stimulating reflective questioning. Usefulness, in this context, was defined not by correctness or automation, but by how well AI feedback could be interpreted, adapted, or rejected in relation to professional judgment and contextual realities.\u003c/p\u003e\n\u003cp\u003eTrust emerged as a negotiated and conditional practice rather than a fixed disposition. Students expressed confidence in AI when it aligned with course expectations and their own evaluative reasoning, while simultaneously articulating caution about over-reliance and decontextualised feedback. This calibrated trust reflects a mature ethical stance in which responsibility for assessment decisions remained firmly human. Rather than eroding agency, the AfL-oriented design appeared to strengthen it by making the limits of AI visible and by preserving authorship through deliberate sequencing and explicit ethical guidance.\u003c/p\u003e\n\u003cp\u003eThe findings also highlight the central role of feedback literacy and reflection in mediating AI use. Students demonstrated increasing capacity to interpret feedback, judge its relevance, and decide how to act on it, capabilities that cannot be automated. Reflection functioned not merely as a reporting mechanism but as a site of ethical reasoning, where boundaries around acceptable AI use were articulated and negotiated. In this sense, ethical engagement with AI was enacted through practice rather than compliance, reinforcing the view that ethics in assessment is relational and situated rather than procedural.\u003c/p\u003e\n\u003cp\u003eThe disciplinary context of music education further underscores the importance of these insights. Assessment in music education relies heavily on qualitative judgment, contextual interpretation, and professional discretion. Participants\u0026rsquo; insistence that final assessment decisions should remain with teachers reflects an acute awareness of the limits of standardisation and algorithmic judgment in such contexts. This suggests that AI-supported assessment models must be sensitive to disciplinary epistemologies rather than applied generically across fields.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eImplications for Practice\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor educators, the study highlights the importance of pedagogical governance when integrating AI into assessment practices. AI tools should be positioned as supportive resources rather than evaluative authorities, with clear guidance that preserves learner authorship, agency, and responsibility. Sequencing matters: requiring students to produce work independently before engaging with AI feedback helps prevent dependency and reinforces ethical use. Embedding AI within AfL designs that emphasise dialogue, reflection, and transparency can transform AI from a risk to assessment integrity into a resource for developing assessment literacy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eImplications for Research\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor researchers, the findings suggest a need to shift attention away from AI performance metrics and towards AI-in-use within authentic pedagogical contexts. Qualitative, interpretive approaches are particularly well suited to examining how ethical and trustworthy AI use is enacted through learners\u0026rsquo; sense-making and reflective judgment. Future research might explore how such designs function across different disciplines, levels of study, or institutional contexts, and how tensions around trust and agency evolve over time.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLimitations and Future Directions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study is situated within a single Master\u0026rsquo;s-level course with a relatively small cohort, and the findings are not intended to be statistically generalisable. Instead, they offer theoretically informed insight into how ethical AI engagement can be pedagogically cultivated. Future studies could extend this work through longitudinal designs, comparative disciplinary analyses, or by examining educators\u0026rsquo; perspectives alongside students\u0026rsquo; reflections.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConcluding Remarks\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOverall, this study contributes to emerging discussions on trustworthy and ethical AI in education by demonstrating that trustworthiness is not guaranteed by technology but is constructed through pedagogical design and reflective practice. When AI is embedded within Assessment for Learning frameworks that foreground agency, trust, and ethical responsibility, it can support rather than undermine principled assessment practices. The challenge ahead lies not in perfecting AI systems, but in designing learning environments that enable educators and learners to engage with AI critically, responsibly, and humanely.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eEthics Approval and Consent to Participate\u003c/h2\u003e\n\u003cp\u003eThis study used anonymised data derived from routine pedagogical activities in a postgraduate course. Students were informed that anonymised excerpts might be used for research and publication. Participation was voluntary and had no impact on assessment outcomes. Informed consent was obtained from all participants.\u003c/p\u003e\n\u003ch2\u003eCompeting Interests\u003c/h2\u003e\n\u003cp\u003eThe authors declare that they have no known financial or non-financial competing interests that could have influenced the work reported in this paper.\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eThis research received no direct research funding. Open access publication was supported through the Springer Nature transformative agreement facilitated by the Konsortium Sumber Elektronik Pendidikan Tinggi (KONSEPt), with institutional support from Sultan Idris Education University, Malaysia.\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eC.A.N.M. conceptualised the study, designed the methodology, conducted the analysis, and drafted the manuscript. M.A.O. contributed to conceptual refinement, critical review of the manuscript, and provided scholarly feedback throughout the writing process. Both authors approved the final version of the manuscript.\u003c/p\u003e\n\u003ch2\u003eAcknowledgement\u003c/h2\u003e\n\u003cp\u003eThe author would like to acknowledge the postgraduate students enrolled in the Master\u0026rsquo;s-level course on assessment in music education (cohort M251PM2) for their thoughtful engagement and reflective contributions, without which this study would not have been possible. Their willingness to critically articulate their experiences, professional judgments, and ethical considerations provided valuable insight into the pedagogical enactment of Assessment for Learning and the responsible use of artificial intelligence in educational assessment. The author also acknowledges the institutional support of Universiti Pendidikan Sultan Idris, within which the course was designed and conducted. The study benefited from an academic environment that encourages reflective practice, ethical inquiry, and pedagogical innovation in music education. The author further acknowledges the support provided by the Konsortium Sumber Elektronik Pendidikan Tinggi (KONSEPt) and the participating institutions for facilitating access to scholarly resources and enabling open access publication. Gratitude is also extended to Springer Nature for supporting open access initiatives through the transformative agreement with KONSEPt. Finally, the author acknowledges the broader scholarly community whose work on Assessment for Learning, feedback literacy, trust in assessment, and human-centred approaches to artificial intelligence informed the conceptual framing of this study.\u003c/p\u003e\n\u003ch2\u003eData Availability\u003c/h2\u003e\n\u003cp\u003eThe datasets generated and analysed during the current study are not publicly available due to ethical considerations and the use of assessed student work but are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAkib, I., Akib, E., Ghafar, M. N. A., \u0026amp; Latif, A. A. (2016). Measuring assessment for learning. \u003cem\u003eMan in India, 96\u003c/em\u003e(1\u0026ndash;2), 267\u0026ndash;277. https://www.researchgate.net/publication/298639911_Measuring_assessment_for_learning\u003c/li\u003e\n\u003cli\u003eArnold, J., \u0026amp; Holden, M. (2024). Designing inclusive Assessment for Learning in English Language Arts. In \u003cem\u003eDesigning Inclusive Assessment in Schools\u003c/em\u003e (pp. 27\u0026ndash;38). Routledge. https://doi.org/10.4324/9781003463184-5\u003c/li\u003e\n\u003cli\u003eBraun, V., \u0026amp; Clarke, V. (2006). Using thematic analysis in psychology. \u003cem\u003eQualitative Research in Psychology, 3\u003c/em\u003e(2), 77\u0026ndash;101. https://doi.org/10.1191/1478088706qp063oa\u003c/li\u003e\n\u003cli\u003eBraun, V., \u0026amp; Clarke, V. (2021). One size fits all? What counts as quality practice in reflexive thematic analysis? \u003cem\u003eQualitative Research in Psychology, 18\u003c/em\u003e(3), 328\u0026ndash;352. https://doi.org/10.1080/14780887.2020.1769238\u003c/li\u003e\n\u003cli\u003eCarless, D. (2009). Trust, distrust and their impact on assessment reform. \u003cem\u003eAssessment \u0026amp; Evaluation in Higher Education, 34\u003c/em\u003e(1), 79\u0026ndash;89. https://doi.org/10.1080/02602930801895786\u003c/li\u003e\n\u003cli\u003eCarless, D. (2017). Scaling up assessment for learning: Progress and prospects. In \u003cem\u003eThe Enabling Power of Assessment\u003c/em\u003e (pp. 3\u0026ndash;17). Springer. https://doi.org/10.1007/978-981-10-3045-1_1\u003c/li\u003e\n\u003cli\u003eDavison, C. (2013). Innovation in assessment: Common misconceptions and problems. In \u003cem\u003eInnovation and Change in English Language Education\u003c/em\u003e (pp. 263\u0026ndash;276). Routledge. https://doi.org/10.4324/9780203362716-29\u003c/li\u003e\n\u003cli\u003ede Sousa Nunes Vieira, M., da Silva Campos Costa, N. M., \u0026amp; Pereira, E. R. S. (2024). Assessment of learning in health higher education. \u003cem\u003eMedicina (Ribeir\u0026atilde;o Preto), 57\u003c/em\u003e(2), e-214928. https://doi.org/10.11606/issn.2176-7262.rmrp.2024.214928\u003c/li\u003e\n\u003cli\u003eDolan, B. M., Arnold, J., \u0026amp; Green, M. M. (2019). Establishing trust when assessing learners: Barriers and opportunities. \u003cem\u003eAcademic Medicine, 94\u003c/em\u003e(12), 1851\u0026ndash;1853. https://doi.org/10.1097/ACM.0000000000002982\u003c/li\u003e\n\u003cli\u003eElwood, J. (2013). Educational assessment policy and practice: A matter of ethics. \u003cem\u003eAssessment in Education: Principles, Policy \u0026amp; Practice, 20\u003c/em\u003e(2), 205\u0026ndash;220. https://doi.org/10.1080/0969594X.2013.765384\u003c/li\u003e\n\u003cli\u003eKaswan, K. S., Dhatterwal, J. S., \u0026amp; Ojha, R. P. (2024). AI in personalized learning. In \u003cem\u003eAdvances in Technological Innovations in Higher Education\u003c/em\u003e (pp. 103\u0026ndash;117). CRC Press. https://doi.org/10.1201/9781003376699-9\u003c/li\u003e\n\u003cli\u003eLi, S. (2025). Trustworthy AI meets educational assessment: Challenges and opportunities. \u003cem\u003eProceedings of the AAAI Conference on Artificial Intelligence, 39\u003c/em\u003e(27), 28637\u0026ndash;28642. https://doi.org/10.1609/aaai.v39i27.35089\u003c/li\u003e\n\u003cli\u003eMarques-Cobeta, N. (2024). Artificial intelligence in education: Unveiling opportunities and challenges. \u003cem\u003eLecture Notes in Educational Technology\u003c/em\u003e, 33\u0026ndash;42. https://doi.org/10.1007/978-981-97-2468-0_4\u003c/li\u003e\n\u003cli\u003eMiranda, S. (2025). Artificial intelligence in education: An exploratory survey. \u003cem\u003eEducation Sciences, 15\u003c/em\u003e(8), 975. https://doi.org/10.3390/educsci15080975\u003c/li\u003e\n\u003cli\u003eMohamed, L. K. (2025). Implementation of AI in education: Promises and challenges. In \u003cem\u003eEducation, Future Jobs and Smart Systems in the Age of Artificial Intelligence\u003c/em\u003e (pp. 51\u0026ndash;66). Emerald. https://doi.org/10.1108/978-1-83608-904-920251004\u003c/li\u003e\n\u003cli\u003eMonika Singh, T., Reddy, C. K. K., Murthy, B. V. R., Nag, A., \u0026amp; Doss, S. (2024). AI and education: Bridging the gap to personalized, efficient, and accessible learning. In \u003cem\u003eInternet of Behavior-Based Computational Intelligence for Smart Education Systems\u003c/em\u003e (pp. 131\u0026ndash;160). IGI Global. https://doi.org/10.4018/979-8-3693-8151-9.ch005\u003c/li\u003e\n\u003cli\u003ePollard, A., \u0026amp; Filer, A. (2009). Social practices in school assessment and their impact on learner identities. In \u003cem\u003eInternational Encyclopedia of Education\u003c/em\u003e (3rd ed., pp. 512\u0026ndash;517). Elsevier. https://doi.org/10.1016/B978-0-08-044894-7.00300-6\u003c/li\u003e\n\u003cli\u003ePope, N., Green, S. K., Johnson, R. L., \u0026amp; Mitchell, M. (2009). Examining teacher ethical dilemmas in classroom assessment. \u003cem\u003eTeaching and Teacher Education, 25\u003c/em\u003e(5), 778\u0026ndash;782. https://doi.org/10.1016/j.tate.2008.11.013\u003c/li\u003e\n\u003cli\u003ePortilla, J. E. N., Cede\u0026ntilde;o, J. K. Z., J\u0026aacute;come, G. O. L., \u0026amp; Gallegos, L. A. M. (2025). Artificial intelligence in Education 4.0: A systematic review. \u003cem\u003eJournal of Educators Online, 22\u003c/em\u003e(3). https://doi.org/10.9743/JEO.2025.22.3.13\u003c/li\u003e\n\u003cli\u003eSutherland-Harris, R., Ali, A., \u0026amp; Elkhoury, E. (2025). Practicing trust between academic developers and faculty for equitable assessments. \u003cem\u003eInternational Journal for Academic Development, 30\u003c/em\u003e(1), 132\u0026ndash;136. https://doi.org/10.1080/1360144X.2024.2445617\u003c/li\u003e\n\u003cli\u003eThalji, N. J., \u0026amp; Alkhasawneh, S. (2025). How can artificial intelligence shape the future of sustainable education? \u003cem\u003eJournal of Theoretical and Applied Information Technology, 103\u003c/em\u003e(9), 3836\u0026ndash;3850.\u003c/li\u003e\n\u003cli\u003eThompson, W. C., \u0026amp; Erickson, J. D. (2025). Voice, recognition, and exit: A rights-based response to cultural bias in educational assessments. In \u003cem\u003eCulturally Responsive Assessment\u003c/em\u003e (pp. 52\u0026ndash;69). Routledge. https://doi.org/10.4324/9781003392217-5\u003c/li\u003e\n\u003cli\u003eWeeden, P., \u0026amp; Simmons, M. (2017). Formative assessment. In \u003cem\u003eDebates in Geography Education\u003c/em\u003e (pp. 140\u0026ndash;155). Routledge. https://doi.org/10.4324/9781315562452\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Artificial intelligence in education, Assessment for Learning, ethical AI, feedback literacy, learner agency, music education, reflective practice, trustworthy AI","lastPublishedDoi":"10.21203/rs.3.rs-8554027/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8554027/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe integration of artificial intelligence (AI) into educational assessment has intensified ethical and pedagogical debates concerning learner agency, trust, and responsibility. While AI-enabled systems are frequently promoted for efficiency and personalization, their use in assessment contexts raises concerns that extend beyond technical performance into questions of judgment, power, and ethical governance. Framed within an Assessment for Learning (AfL) perspective, this qualitative interpretive study examines how postgraduate music education students make sense of an AI-based Assessment and Feedback Assistant embedded within a Master\u0026rsquo;s-level course at a Malaysian public university. Drawing on reflective forum posts produced as part of routine coursework, the study explores how learners articulate perceptions of usefulness, negotiate trust, assert agency, and establish ethical boundaries around AI-supported feedback. Data was analysed using reflexive thematic analysis, informed by concepts of AfL, feedback literacy, trust, and human-centred AI ethics. The findings indicate that students do not experience AI as an authoritative assessor, but as a provisional and dialogic resource that supports reflective sense-making when embedded within an AfL-oriented pedagogical design. Trust in AI emerged as conditional and negotiated, grounded in alignment with human judgment and contextual relevance rather than technological authority. The study argues that ethical and trustworthy AI use in assessment is enacted through pedagogical governance rather than guaranteed by system design alone. By foregrounding reflection, transparency, and learner agency, AfL-oriented environments can enable AI to function as support for assessment literacy without displacing human responsibility. The study contributes empirical insight into human-centred approaches to AI ethics in assessment, particularly within qualitative, judgment-intensive disciplines such as music education.\u003c/p\u003e","manuscriptTitle":"AI-Supported Feedback as Assessment for Learning: Learner Agency, Trust, and Ethical Sensemaking in a Postgraduate Music Education Context","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-28 13:45:11","doi":"10.21203/rs.3.rs-8554027/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"566fd133-80e9-49fe-b1fa-7670e4564d3c","owner":[],"postedDate":"January 28th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-10T08:42:28+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-28 13:45:11","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8554027","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8554027","identity":"rs-8554027","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}