The review’s central conceptual move is to distinguish two kinds of assistance. Instrumental or representational assistance helps learners generate options, translate, format, retrieve, summarise or re-express material. Judgement-bearing assistance goes further: it evaluates correctness, relevance, quality, persuasiveness, ethical acceptability or evidential sufficiency. The two categories can overlap, since even a summary selects what matters, but judgement-bearing assistance is normatively far more demanding because it delegates not only the production of work but the standards by which that work is assessed. A learner may use AI repeatedly yet remain epistemically agentic if the interaction prompts questioning, comparison and revision; another may use it once but treat its answer as decisive in a high-stakes task. The analytic boundary, the authors insist, is not use versus non-use but the relation between assistance and judgement.

To operationalise that boundary, the review proposes six diagnostic criteria separating productive reliance from harmful dependence: contestability, recoverability, transfer, traceability, distributed responsibility and epistemic plurality. Dependence becomes problematic when the relation is difficult to contest, obscures its evidential basis, weakens the learner’s capacity to reconstruct or transfer judgement, or allocates responsibility to people who lack meaningful control. Conversely, frequent AI use can remain productive when it provokes comparison, makes uncertainty visible and leaves the learner more capable of independent and collaborative judgement. The criteria are offered as diagnostic questions for research, design and pedagogy rather than a psychometric scale, and no single criterion is decisive in every context; the pattern matters, as does the importance of the delegated judgement.

The heart of the paper is an analysis of four sociotechnical pathways through which AI affordances and institutional conditions can slide into harmful dependence. The first is fluent authority. Large language models produce grammatically polished, coherent and often confident responses, and fluency is epistemically persuasive because users can mistake ease of processing for reliability. Anthropomorphism, social presence and personalisation can amplify this effect: conversational systems occupy roles associated with human epistemic others, such as respondent, tutor, editor and evaluator, and a supportive tone with apparent memory may encourage the transfer of interpersonal trust to a system that lacks human understanding, responsibility or commitment to the learner. The risk is not error alone but the alignment of error, confidence and immediacy, especially for learners with limited domain knowledge.

The second pathway is frictionless delegation. Generative AI compresses searching, reading, comparing, synthesising and drafting into a single request-response cycle. Cognitive offloading can be adaptive when it frees limited resources for planning, reflection or complex problem solving, but adjacent research on internet search suggests that persistent access to external information can alter memory, confidence and future search behaviour. When intermediate epistemic actions disappear from the workflow, learners may complete tasks without practising the very actions the tasks were intended to develop. Assessment incentives intensify the problem: if institutions reward polished products while leaving process and justification invisible, delegating epistemic work becomes rational, and AI shifts from scaffold to substitute, particularly in feedback and evaluation.

The third pathway is opaque synthesis. Conversational AI often presents an integrated answer without making its source selection, weighting, exclusions or uncertainty inspectable. Traditional search was never transparent or neutral, but it commonly exposed multiple documents, authors and domains, even if learners evaluated them poorly. Conversational synthesis can hide that plurality behind one voice, transforming the learner’s task from selecting among visible sources to recovering the evidential structure of an answer that appears already complete. This can weaken verification while improving the surface quality of the product, making epistemic deficits harder for teachers to detect. It also threatens disciplinary reasoning, since a generic synthesis may flatten the distinct standards by which historical, scientific and philosophical claims are warranted.

The fourth pathway is institutionalised dependence. Universities and schools shape reliance through procurement, platform integration, assessment design, timetabling, policy and professional development, while commercial systems encode objectives concerning engagement, speed, cost and data capture. Once AI is built into learning management systems, writing environments and feedback workflows, it may become the default route into academic work rather than a discrete tool chosen by the learner. Institutionalisation also redistributes authority and responsibility: a student may be held accountable for claims generated through an institutionally licensed but opaque system, a teacher may be expected to police use without access to system logs, and a university may depend on vendor assurances that cannot be independently audited. The review argues that responsibility should track control, knowledge and benefit across learners, educators, institutions and providers, rather than being dumped on the end user.

This institutional dimension brings epistemic justice to the centre of the analysis. Generative systems draw on unequal knowledge infrastructures and may reproduce dominant languages, classifications and perspectives while marginalising local, minoritised or experiential knowledge. Drawing on Miranda Fricker’s account of epistemic injustice and on recent work on formative epistemic injustice, the authors argue that learning arrangements can wrong students by restricting the knowledge, practice and accurate self-assessment through which they develop as knowers. Repeated reliance may also reshape learner identity, fostering a self-conception of being unable to write, understand or judge without AI. Productive reliance should expand participation and capability over time; harmful dependence makes learners and institutions more fragile when the system is absent, changes its terms or fails particular communities.

Against these risks, the review proposes relational epistemic agency as the normative aim: the capacity to question, verify, compare, justify and take responsibility for knowledge claims within human, technological and institutional relations. This is not independence from the machine, and it is not an anti-dependence position. Education has always involved dependence on teachers, peers, texts, instruments and institutions, and ideals of self-sufficient knowing are both unrealistic and exclusionary. The relevant test is functional and developmental: does the human-technology relation enlarge the learner’s capacity to participate in epistemic practice, or merely deliver a product? A calculator supports mathematical agency when the learner understands when and why its operations are appropriate; a generative model supports inquiry when it expands hypotheses, reveals alternatives and prompts verification. The same tools bypass agency when they become non-contestable authorities.

The practical implications are concrete. Designers should provide claim-level provenance where feasible, distinguish retrieved evidence from model-generated synthesis, represent uncertainty through alternatives and explicit unknowns, preserve user control over prompts and outputs, and surface disagreement and culturally diverse sources. Pedagogy should move from policing AI use to teaching AI-mediated judgement, with routines for lateral reading, source triangulation and claim verification, and assignments that require learners to annotate AI responses, identify unsupported assumptions and document why they accepted or rejected particular suggestions. Assessment should make process, judgement and transfer visible through staged drafts, oral defence, source maps and reflective decision logs, without becoming surveillance-heavy. Institutions, meanwhile, should evaluate procurement for provenance, bias, auditability and data governance, and specify which learning outcomes must remain demonstrably human. The empirical agenda that follows is equally clear: researchers must track which epistemic actions are preserved, transformed or displaced, using process evidence such as interaction traces and think-aloud protocols, and must examine organisations and markets, not only learners. The future of knowing, this review suggests, depends less on how often students use AI than on whether the systems and institutions around them are configured so that judgement survives the delegation.

Subject of Research: Epistemic dependence and learner agency in AI-mediated education

Article Title: Epistemic dependence in AI-mediated learning

Article References: Du, Y., & Yuan, Y. (2026). Epistemic dependence in AI-mediated learning. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03294-1

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03294-1

Keywords: generative AI, epistemic dependence, epistemic agency, AI in education, cognitive offloading, epistemic justice, evaluative judgement, large language models, human-AI interaction, assessment, educational governance, AI & Society

The study, led by Zachery Quince of Southern Cross University’s Centre for Learning and Teaching together with Emily Faulconer of Monash University, examined a second-year physics unit serving engineering and education students. Before the term began, the university’s policy had been overhauled, collapsing a five-tier model of permitted AI use into three categories: no use, purpose-specific use, and open use. Because none of the unit’s existing assessments complied with the new rules, the course coordinator was forced to redesign all three tasks, creating a natural experiment in how policy-driven assessment change reshapes student behaviour.

The redesign was revealing in itself. The first assessment, an oral presentation worth 20 percent of the unit, operated under purpose-specific rules: students could use AI for background research, idea generation, slide drafting and script development, but the final recorded presentation had to be their authentic selves, not an AI-generated avatar. The second assessment, a laboratory workbook worth half the grade, and the third, a take-home quiz worth 30 percent, both ran under open-use rules that permitted unrestricted AI engagement, provided students submitted a mandatory declaration describing exactly which tools they used and why.

Those declarations became the study’s raw data. Across the three assessments, students reported more than 120 unique use cases, which the researchers coded into four functional categories: learning, formatting, doing the task, and feedback. The pattern that emerged was striking. In the first assessment, use was dominated by learning and formatting, together accounting for over 95 percent of reported uses, as students brainstormed ideas, checked grammar and structured their presentations. By the final take-home quiz, feedback had exploded to 66.7 percent of all reported uses, with students describing how they checked calculations, verified answers and clarified concepts such as the viability of formulas.

To interpret this shift, the researchers mapped each category onto the SAMR framework, a well-known model that classifies technology integration into four levels: substitution, augmentation, modification and redefinition. Learning and formatting uses sat mostly at the lower levels, where AI merely replaces or improves existing tools without changing the task. Doing the task, where AI generated step-by-step solutions to unfamiliar problems, reached modification and redefinition, the levels where technology genuinely transforms what students can do. Feedback proved the most interesting category of all: it spanned every SAMR level, from simple error-spotting at the bottom to an iterative, student-controlled dialogue at the top, in which students could question, test and immediately apply advice in real time, something no previous feedback mechanism offered at comparable speed or scale.

The performance data added a cautious footnote to the behavioural story. Students who declared AI use scored slightly higher than non-users on all three assessments, with the largest gap appearing in the scaffolded presentation, where users averaged 15.76 out of 20 against 14.80 for non-users, a moderate effect size. But none of the differences reached statistical significance, and the researchers are explicit that this trend should be read as a tentative observation rather than evidence that AI boosts grades. What the data does show is that declared AI use was not associated with poorer achievement in a course where assessments required students to explain reasoning and take final responsibility for their work.

That caveat matters because the wider literature is deeply divided. Some studies report that AI feedback improves performance; others find that heavy ChatGPT use correlates with worse exam results, procrastination and memory loss. Research on high school mathematics has shown that generative AI without guardrails can harm learning, while other work finds stronger outcomes when students use AI to construct knowledge rather than merely execute procedures. The Australian authors argue that much of this contradiction stems from studies conflating fundamentally different behaviours, such as substitution-level use that bypasses cognitive effort with modification-level use that supports it, and from unclear policy contexts that make results impossible to compare.

The study’s central insight is what the researchers call task contingency. Students did not use AI simply because it was available; they used it for the kinds of support each task made useful. A presentation task made planning, language and communication support relevant. A workbook and quiz made checking, clarification and verification relevant. This means a policy label such as purpose-specific or open use can define what is permitted but cannot determine what students will actually do. The assessment design creates the demand. Where a task rewards polished communication, students reach for AI to polish. Where it demands problem-solving, they reach for AI to test their reasoning.

The findings also complicate the academic integrity narrative that dominated early coverage of AI in education. Far from being passive cheaters, students in this cohort reported using AI for preparation, clarification, refinement and verification, not only for answer production. The authors acknowledge the limits of self-reported data, which may be incomplete or strategically framed, and note that the single-marker design and small cohort of 36 students restrict generalisability. Still, they argue that integrity responses built purely on restriction and suspicion are too blunt, and that the real design question is whether an assessment requires students to demonstrate judgement and reasoning that AI cannot simply declare on their behalf.

The practical recommendations are concrete. Where students use AI for feedback, the authors suggest requiring them to highlight the feedback received and explain how it did or did not shape their revisions, making evaluative judgement visible and building feedback literacy. Greater restriction remains appropriate where AI could perform the intended cognitive work without students demonstrating their own understanding. The broader lesson is that generative AI has not invalidated authentic assessment; it has made it urgent. When tools can instantly solve standard textbook problems, educators are pushed toward tasks demanding synthesis, justification and communication, the very skills that define professional practice in physics and engineering. In that sense, the technology that threatened to hollow out assessment may end up forcing higher education to design better ones.

Subject of Research: How changing institutional generative AI policies influence student AI use and academic performance in undergraduate physics assessment

Article Title: Student GenAI use under changing institutional policies: a mixed-methods case study in undergraduate physics

Article References: Quince, Z., & Faulconer, E. (2026). Student GenAI use under changing institutional policies: a mixed-methods case study in undergraduate physics. International Journal of STEM Education, 13(1), Article 61. https://doi.org/10.1186/s40594-026-00651-w

Image Credits: AI Generated

DOI: 10.1186/s40594-026-00651-w

Keywords: generative AI, higher education, physics education, assessment design, SAMR framework, academic integrity, institutional policy, feedback, evaluative judgement, STEM education, authentic assessment, constructive alignment