Resisting cognitive laziness: a DASHBOARD to make AI use visible and transform critical thinking

Reboot Mental

ARPASANU-SUSOI Nicoleta-Roxana ; nicoleta-roxana.arpasanu-susoi@etu.univ-montp3.fr ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ARAMBURU Josela ; josela-ines.aramburu@etu.univ-montp3.fr ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ MIRANDA Anthony ; anthony.miranda@etu.umpv.fr ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀All students are from Université Paul Valéry

Training / education / pedagogy

Our use case is an 𝗼𝗽𝘁-𝗶𝗻 𝗰𝗼𝗴𝗻𝗶𝘁𝗶𝘃𝗲 𝗱𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱 designed for students who already use generative AI in their academic work, but want to do so without losing intellectual ownership of the task. We built it around our learner persona, (imagine!) Emma , a 23-year-old Master’s student who uses AI to cope with dense readings, tight deadlines and academic overload. Emma does not want AI to replace her thinking; she wants to know whether she is still genuinely learning, verifying, interpreting and reformulating. Our solution therefore does not focus on detecting AI use, but on making the student’s 𝗶𝗻𝘁𝗲𝗹𝗹𝗲𝗰𝘁𝘂𝗮𝗹 𝗽𝗿𝗲𝘀𝗲𝗻𝗰𝗲 visible. It documents what the student checked, reformulated, triangulated or accepted too quickly, and turns an opaque AI interaction into a transparent learning process. In its current form, the project is no longer just a dashboard: it is a guided workflow built around a “human first, AI support, verification, reflection” sequence, ending in a student-controlled cognitive report. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Please click here to see our PERSONA: https://docs.google.com/presentation/d/1l2K-71jnhJrg3tbBQQ1gT5D0Mf2NhYgD5qsY2u49nBo/edit?usp=sharing

This use case matters because generative AI is no longer a future scenario in higher education; it is already part of students’ everyday academic life. Recent evidence shows that AI use is now almost universal among students: 95% report using AI in at least one way, and 94% say they use generative AI to help with assessed work. AI is therefore not an external or occasional tool anymore. It is already woven into reading, writing, structuring ideas, explaining concepts and preparing assignments. At the same time, this normalisation creates a serious educational tension: students may become more efficient, but not necessarily more intellectually present. OECD evidence warns that when AI is used mainly to deliver direct answers, it can improve short-term task performance while reducing active engagement and weakening deeper learning. In other words, the problem is no longer whether students use AI, but whether their use of AI still supports thought, judgement and real understanding. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀This is why the issue cannot be reduced to student behaviour alone. It concerns the whole triangle of student, university and Big Tech. On the student side, learners like Emma use AI because of academic overload, time pressure and the need for support, not simply because they want to avoid effort. But once AI produces fluent and plausible outputs, students can lose sight of what they actually verified, understood, reformulated or accepted too quickly. Our project starts from this precise point: the current problem is one of cognitive opacity. The student receives an apparently clean final answer, but the intellectual path becomes blurred. The dashboard matters because it makes visible what is usually invisible: source checking, detection of hallucinations, bias identification, reformulation, triangulation and reflection. It turns AI use from a hidden shortcut into a documented learning process. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀On the university side, the problem is equally urgent. Many institutions are trying to respond, but often in fragmented or inconsistent ways. The HEPI survey shows that institutional encouragement around AI remains mixed, that students still feel uncertainty about acceptable use, and that there is a real anxiety around false accusations of cheating. At the same time, fewer than half of students feel their teaching staff are helping them develop AI skills for the future. This suggests a gap between the speed of AI adoption and the pedagogical support currently available. My reading of this is that many teachers are not simply resistant; they are often left without sufficiently clear, discipline-sensitive tools to guide students toward critical and pedagogical uses of AI. OECD work reinforces this concern: general-purpose systems such as ChatGPT or Claude are not designed around curriculum, pedagogy, learner modelling or teacher autonomy. So the challenge for universities is not just to regulate AI, but to reclaim a formative role: teaching students how to verify, question, compare, and think with AI rather than merely through it. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀The Big Tech dimension is also central. Today, AI companies need better human feedback if they want to reduce hallucinations, improve factual reliability, and limit biased or stereotyped outputs. Our project is important because it imagines a more intelligent and more ethical relationship between students and AI companies. Instead of extracting value invisibly from user activity, the system would allow students to contribute voluntarily and under opt-in control the corrections they have genuinely produced: detected hallucinations, verified errors, identified biases, and domain-specific reformulations. In the project, these corrections are not random clicks; they are contextualised acts of critical work that could enrich RLHF pipelines with higher-quality educational feedback. This is especially valuable because student corrections are anchored in real disciplinary tasks, not abstract benchmark settings. Properly governed, this could help improve model quality while recognising students as contributors rather than passive consumers. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀At the same time, this must not be framed naively. Big Tech does not just need more data; it needs better data and better human judgement. Hallucinations and stereotyped outputs are not fixed simply by scale. They require critical feedback, contextual interpretation, and human oversight. That is why our proposal insists on an ethical model: student control over data, explicit consent, pseudonymisation, and the possibility of keeping the cognitive report private rather than sharing it. The value of the project is precisely that it does not imagine the university surrendering its mission to platforms. On the contrary, it repositions the university as the place that trains critical supervisors of AI, while allowing scientifically useful feedback to flow toward model improvement only under conditions that protect human agency, privacy and pedagogical purpose. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀So, this use case matters because it responds to a real structural transformation already underway. Students are already living with AI. Universities are under pressure to move beyond prohibition and build critical AI literacy. AI companies need high-quality human feedback to reduce errors and harmful biases. Our project sits exactly at this intersection. It proposes that the answer is not surveillance, punishment or blind adoption, but a new infrastructure of visibility: one that documents intellectual effort, strengthens critical thinking, supports teachers pedagogically, and channels human correction into AI improvement without sacrificing student sovereignty. That is why this is not just a dashboard. It is a proposal for a new social and pedagogical contract around AI in higher education.

Our team’s motivation has evolved significantly throughout this project. At first, we imagined a relatively simple dashboard that would make students’ AI use more visible. As our work progressed, however, we realised that visibility alone was not enough. The real challenge is not simply to visualise interaction with AI, but to protect and strengthen students’ cognitive agency, reduce the drift toward passive dependence, and rethink the relationship between learning, assessment and AI within higher education. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀This evolution was strongly shaped by the interdisciplinary nature of our team. We bring together three complementary perspectives: a mathematics and computer science student focused on the project’s technical feasibility; a psychology student concerned with cognitive indicators and pedagogical methodology; and a communication student interested in RLHF, feedback dynamics and the broader communicational implications of AI use. Our group is also heterogeneous in terms of generation and academic level, with one member pursuing a bachelor’s degree and two pursuing master’s degrees. Rather than seeing this diversity as a limitation, we consider it a strength: it allows us to approach student AI use from a broader and more realistic range of experiences and expectations. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀We are all deeply interested in AI and in helping students use it as a tool for learning rather than as a substitute for thinking. We believe this is the right moment for such a project, precisely because current uses are still being shaped and universities still have the opportunity to prevent harmful habits from becoming normalized. This is why our project gradually moved from a tool-centred idea to a broader pedagogical and ethical architecture: an opt-in system, a cognitive coach, a one-page report controlled by the student, discipline-specific university training, and a governance model that preserves the university’s formative role instead of outsourcing it to platforms. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Our learning objectives are therefore fourfold: to understand how AI reshapes students’ cognition and emotions; to design indicators that capture verification, transformation and reflection rather than mere productivity; to build an ethically robust model grounded in human agency, privacy and fairness; and to move from a punitive logic of suspicion toward a culture of documented, critical and teachable AI use. In that sense, our project is no longer just “a dashboard”: it is a proposal for how higher education can transform AI use from opacity into intellectual accountability and learner sovereignty.

𝗪𝗵𝗮𝘁 𝗶𝘀 𝗮 𝗱𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱 𝗶𝗻 𝗼𝘂𝗿 𝗽𝗿𝗼𝗷𝗲𝗰𝘁? In our project, a dashboard is a simple visual interface, similar to a concise infographic, designed to make the learning process more visible and understandable. It brings together key data and a short synthetic interpretation of how a student interacted with AI during a summarisation task. Instead of showing raw technical information, it presents readable indicators of the student’s intellectual activity: what was asked, what was generated by AI, what was verified, what was reformulated in the student’s own words, and what was accepted, corrected or potentially lost in the transformation of the original content. The dashboard therefore acts as a pedagogical mirror: it translates an opaque interaction into a transparent and interpretable representation of the student’s work. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀𝗢𝘂𝗿 𝗽𝗿𝗼𝗷𝗲𝗰𝘁 𝗵𝗮𝘀 𝗯𝗲𝗲𝗻 𝘀𝘂𝗯𝘀𝘁𝗮𝗻𝘁𝗶𝗮𝗹𝗹𝘆 𝗶𝗺𝗽𝗿𝗼𝘃𝗲𝗱 𝗮𝗻𝗱 𝗲𝘅𝗽𝗮𝗻𝗱𝗲𝗱 𝗱𝘂𝗿𝗶𝗻𝗴 𝘁𝗵𝗲 𝗰𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲 (please check our last version : https://docs.google.com/document/d/1vVdMwtJcyJizCM9h0Ano0S4z2nf9ApL18Lb_V4TWDyw/edit?usp=sharing). What started as a simple idea for making AI use visible has become a more complete model structured around four pillars: social, emotional, material and intellectual. These pillars allow us to address not only the technical side of the dashboard, but also the human, pedagogical and institutional dimensions of AI use in higher education. We also added a clear ethical framework, ensuring that the dashboard supports autonomy, transparency and fairness rather than control or punishment. In parallel, we developed a practical recipe for improvement, with simple actions that help students progress when their interaction with AI remains too passive or insufficiently critical. Finally, we made our social commitment more explicit: this project is designed to strengthen critical thinking, support responsible AI use and help universities remain active educational actors rather than passive observers of technological change.⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀𝟭. 𝗦𝗼𝗰𝗶𝗮𝗹 𝗽𝗶𝗹𝗹𝗮𝗿 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀The social pillar emphasizes that AI use in higher education is not only an individual practice, but a relational and institutional process. The dashboard reshapes the roles of the main actors in the educational ecosystem: the student becomes a more active and critical auditor of information; the teacher moves away from a logic of suspicion and toward a more transparent evaluation of the learning process; the university strengthens its formative role by teaching critical AI literacy; and AI companies may benefit from higher-quality feedback grounded in real academic verification practices. This perspective is also inspired by Alex Mucchielli’s systemic approach to communication, according to which behaviours and meanings can only be understood through the network of interactions, roles and contexts in which they occur. In this sense, the dashboard is not just a tool for one user: it is a device that reorganises the educational relationship around visibility, responsibility and shared understanding. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀𝟮. 𝗜𝗻𝘁𝗲𝗹𝗹𝗲𝗰𝘁𝘂𝗮𝗹 𝗽𝗶𝗹𝗹𝗮𝗿 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀The intellectual pillar gives the dashboard its pedagogical and methodological depth. Its purpose is not simply to record AI use, but to interpret how learning is taking place through that use. For this reason, the dashboard is structured around three analytical layers. The first layer concerns the quality of the input, that is, the way the student formulates the prompt before the exchange really begins. Here, the interaction is examined through the 6C framework — clarity, complexity, coherence, creativity, consistency and contextuality — in order to assess whether the request made to AI is explicit, demanding, coherent and sufficiently situated. The second layer concerns the interactional process itself. It focuses on observable traces of behaviour during the exchange: returns to the source text, requests for justification, comparisons between versions, correction of the first AI answer, evolution in prompt formulation, and signs of passive acceptance. These traces do not yet prove critical thinking by themselves, but they provide the empirical basis from which a pedagogical interpretation becomes possible. The third layer is therefore the level of pedagogical and cognitive interpretation. It transforms those observable traces into meaningful indicators such as content fidelity, level of transformation, rigour of verification, reflective activation, automation bias, cognitive fatigue, AI overdependence, weak information literacy, insecurity or impostor feeling, and verification perfectionism. In this sense, the dashboard does not confuse what the student asked, what the student did, and what those behaviours mean educationally; it distinguishes these levels in order to produce a more rigorous reading of the learning process. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀The indicators are evaluated through a mixed logic. Some can be inferred automatically from the interaction itself: for example, fidelity to the source, degree of textual transformation, or part of the verification behaviour can be detected through comparison between the original material, the AI output and the student’s revised version. Other dimensions, however, cannot be inferred reliably from behavioural traces alone. This is especially true for reflective understanding, insecurity, or some finer cognitive biases. To address this, our project draws on the MentS scale as an important conceptual basis. Rather than treating mentalization as a separate questionnaire disconnected from the task, we adapt its logic to the AI-supported summarisation process in order to capture whether the student is simply reacting to the output or is actually engaging in reflective, self-aware and cognitively mediated learning. For these indicators, the dashboard uses micro-reflective prompts integrated into the task, asking the student to explain a concept, justify a correction, identify what nuance the AI omitted, or reflect on why a given answer felt convincing at first sight. In this way, the system combines trace-based indicators, brief reflective elicitation, and a mentalization-oriented framework inspired by MentS, so that the evaluation remains pedagogically meaningful rather than merely technical. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Methodologically, this pillar is coherent with the PMAISE model because PMAISE treats pedagogy as the active mediator between AI affordances and meaningful student engagement. In that model, AI has educational value only when it is embedded in instructional design mechanisms such as structured feedback, scaffolding and alignment with learning needs. That is exactly the role of the three-layer architecture in our project: the first layer clarifies the student’s demand, the second observes the real dynamics of the interaction, and the third interprets these dynamics in pedagogical terms so that feedback can support improvement. The pillar is also consistent with a constructivist and constructive alignment logic: learning depends less on what the tool produces than on what the student actually does with it, and sound pedagogy requires alignment between objectives, learning activities and assessment. Our dashboard follows this logic by linking prompt quality, interaction traces and interpretive indicators into one coherent learning design. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀𝟯. 𝗘𝗺𝗼𝘁𝗶𝗼𝗻𝗮𝗹 𝗽𝗶𝗹𝗹𝗮𝗿 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀The emotional pillar addresses a central but often overlooked dimension of AI use in higher education: the student’s emotional relationship to the tool. Our project starts from the idea that many students do not experience generative AI in a neutral way. On the contrary, AI use can generate discomfort, uncertainty and moral tension. In our framework, this emotional state is informed by the notion of “AI guilt” discussed by Chan (2024) and further developed by Qu and Wang (2025), who link AI use to feelings of imposture, cognitive dissonance, stress and doubts about the legitimacy of one’s own work. This unease is reinforced by the broader academic climate: recent student evidence also shows anxiety about being falsely accused of cheating and uncertainty about what counts as acceptable AI use. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Our dashboard is designed to transform that emotional experience. Instead of leaving the student alone with a fluent but opaque tool, it makes visible the concrete traces of intellectual work: what was verified, what was reformulated, what was corrected, and where critical judgement was exercised. This visibility matters emotionally because it allows the student to see that the work is not based on passive delegation, but on active cognitive engagement. The aim is therefore to move from opacity, fear and impostor feeling toward confidence, relief and intellectual sovereignty. In this sense, the dashboard does not only measure behaviour; it helps restore a sense of authorship and legitimacy in AI-supported learning. At the same time, it also reduces pressure on teachers by shifting the pedagogical relationship away from suspicion and toward documented critical practice. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀This emotional transformation is not automatic: it depends on pedagogical mediation. As the PMAISE model shows, AI supports engagement most meaningfully when it is embedded in structured feedback, interactive scaffolding and clear instructional alignment. That logic is essential for the emotional pillar as well, because reassurance does not come from AI alone, but from the way the institution frames and guides its use. This is why the university has a central role in our model: by offering explicit guidance, assessment-specific clarity, and training in critical AI literacy, it can reduce uncertainty and replace a culture of guilt and fear with one of confidence, reflective use and legitimate intellectual agency. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀𝟰. 𝗜𝗻𝘁𝗲𝗹𝗹𝗲𝗰𝘁𝘂𝗮𝗹 𝗽𝗶𝗹𝗹𝗮𝗿 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ The intellectual pillar explains how the dashboard works in practice, step by step, as a simple pedagogical recipe. First, the student activates the dashboard at the beginning of the task. From that moment, the system records the interaction and later displays it in a clear visual format, similar to a compact infographic. The dashboard presents a set of readable indicators showing how the student worked with AI during the summarisation task: the quality of the prompt, the degree of transformation between the AI output and the student’s final text, the level of source verification, the presence of reformulation in the student’s own words, and possible signs of passive acceptance or over-reliance. In this way, the student does not only receive an answer from AI, but also a visible representation of the intellectual process behind the task. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀The second step is the comprehension evaluation. The dashboard does not assume that a fluent interaction means real understanding. For this reason, it introduces short reflective checks integrated into the workflow. The student may be asked to explain a concept in their own words, justify why a passage was considered unreliable, identify what nuance the AI forgot, or clarify why a correction was necessary. This part is essential because it helps distinguish between superficial task completion and genuine understanding. Our approach is also informed by the 𝗠𝗲𝗻𝘁𝗦 𝘀𝗰𝗮𝗹𝗲, which provides a useful conceptual basis for thinking about reflective and mentalization-related dimensions of learning. The goal is therefore not only to analyse what the student did, but also to understand whether the student remained cognitively engaged and aware of their own reasoning. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀The third step is the activation protocol, which acts as an improvement recipe. If the dashboard detects weak points, it does not stop at diagnosis. Instead, it proposes concrete pedagogical actions. For example, if source triangulation is weak, it can suggest comparing the AI response with Google Scholar, PubMed, or institutional resources. If the interaction shows too much passive acceptance, it can encourage the student to ask AI only for key concepts and then write the synthesis independently. If reformulation is poor, it can invite the student to rewrite the answer in a different structure or with discipline-specific vocabulary. Each weak indicator therefore becomes a practical entry point for improvement. The dashboard functions not only as a mirror, but also as a cognitive coach that guides the student toward a more critical and autonomous use of AI. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀This practical sequence is grounded in two explicit frameworks. On the methodological and pedagogical side, it is aligned with the 𝗣𝗠𝗔𝗜𝗦𝗘 𝗺𝗼𝗱𝗲𝗹 (𝑷𝒆𝒅𝒂𝒈𝒐𝒈𝒊𝒄𝒂𝒍 𝑴𝒆𝒅𝒊𝒂𝒕𝒊𝒐𝒏 𝒐𝒇 𝑨𝑰 𝒇𝒐𝒓 𝑺𝒕𝒖𝒅𝒆𝒏𝒕 𝑬𝒏𝒈𝒂𝒈𝒆𝒎𝒆𝒏𝒕), because the dashboard does not treat AI as educationally valuable by itself, but only when it is combined with structured feedback, scaffolding and reflective engagement. The sequence follows that logic step by step: first, the dashboard makes the interaction visible through indicators; second, it checks whether comprehension is actually present; third, it activates targeted pedagogical support when needed. On the ethical side, this pillar is framed by the 𝗔𝗟𝗧𝗔𝗜 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 (𝑨𝒔𝒔𝒆𝒔𝒔𝒎𝒆𝒏𝒕 𝑳𝒊𝒔𝒕 𝒇𝒐𝒓 𝑻𝒓𝒖𝒔𝒕𝒘𝒐𝒓𝒕𝒉𝒚 𝑨𝑰), which is why the dashboard is not designed as a surveillance or punishment tool, but as a system that protects human agency, transparency, intelligibility, fairness and data control. Its purpose is to make AI use more visible without becoming intrusive, and to ensure that the information displayed always serves a pedagogical function: helping the student learn better, reflect more deeply, and remain in control of their own cognitive work. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀𝗧𝗵𝗲 𝘄𝗮𝗹𝗹𝗲𝘁 𝘀𝘆𝘀𝘁𝗲𝗺 is part of a broader redistribution model linking Big Tech, the university, students and teachers. Once students have identified and documented hallucinations, biases or factual errors, and once these detections have been validated within the university, the process does not rely on isolated individual submissions. Instead, the university centralises the validated contributions and sends them to partner AI companies as qualified data packages for RLHF, organised according to their academic quality and disciplinary specialisation. The university then negotiates funding on the basis of the value of these certified data packages. Before any transfer takes place, the data are anonymised or pseudonymised so that the identities of both students and teachers remain protected. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Once funding is received from Big Tech, the university uses the wallet system to redistribute the value generated by this critical intellectual work. For students, this takes the form of campus credits that can be used for meals, printing, books or other university services, thereby giving tangible recognition to their critical engagement with AI. For teachers, a percentage of the redistributed value may take the form of research funding, expertise bonuses or additional pedagogical resources, recognising the time and academic labour involved in validating the quality of the data. In this way, the wallet is not just a student reward mechanism, but part of a wider ethical and institutional model of value redistribution. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀From an ethical perspective, this system is designed to remain consistent with the ALTAI framework. It respects human agency because participation remains voluntary and based on opt-in consent; it promotes transparency because each actor can understand how the data circulate and how the value is redistributed; it protects privacy and data governance because only relevant intellectual contributions are transferred and only after anonymisation; and it supports fairness by recognising both student critical work and teacher expertise. The university acts as a trusted intermediary, ensuring that the system rewards intellectual contribution without turning learning into surveillance. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀𝗕𝗶𝗯𝗹𝗶𝗼𝗴𝗿𝗮𝗽𝗵𝘆 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Al-Khalifa, H., Almansour, R., Alhuasini, L. A., Alsaleh, A., Temsah, M.-H., & Alruwaili, A. R. S. (2025). The prompting brain: Neurocognitive markers of expertise in guiding large language models. arXiv. https://doi.org/10.48550/arXiv.2508.14869 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Biggs, J. (1996). Enhancing teaching through constructive alignment. Higher Education, 32, 347–364. https://doi.org/10.1007/BF00138871 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Chan, C. K. Y. (2024). Exploring the factors of “AI guilt” among students: Are you guilty of using AI in your homework? arXiv. https://doi.org/10.48550/arXiv.2407.10777 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Dimitrijević, A., Hanak, N., Altaras Dimitrijević, A., & Jolić Marjanović, Z. (2018). The Mentalization Scale (MentS): A self-report measure for the assessment of mentalizing capacity. Journal of Personality Assessment, 100(3), 268–280. https://doi.org/10.1080/00223891.2017.1310730 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (2024). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. British Journal of Educational Technology, 56(2), 489–530. https://doi.org/10.1111/bjet.13544 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Gonsalves, C., & Lin, Z. (2025). Clear in advance to whom? Exploring ‘transparency’ of assessment practices in UK higher education institution assessment policy. Studies in Higher Education, 50(7), 1454–1470. https://doi.org/10.1080/03075079.2024.2381124 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Long, D. Y., Wang, S., Md Rashid, S., & Lu, X. T. (2026). Artificial intelligence in higher education: A systematic review of its impact on student engagement and the mediating role of teaching methods. Frontiers in Education, 10, Article 1648661. https://doi.org/10.3389/feduc.2025.1648661 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Mamełka, K. I., & Fabjanowicz, A. (2023). Organizacja osobowości a zdolność do mentalizowania w grupie osób uzależnionych od alkoholu. Psychiatria i Psychologia Kliniczna, 23(4), 260–267. https://doi.org/10.15557/PiPK.2023.0033 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Mucchielli, A. (2004). Étude des communications: Approche systémique dans les organisations. Armand Colin. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Mucchielli, A. (2016). L’approche communicationnelle. In J.-F. Dortier (Ed.), La communication: Des relations interpersonnelles aux réseaux sociaux (pp. 184–195). Éditions Sciences Humaines. ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Nugumanova, A., Alzhanov, A., Mansurova, A., & Mansurova, M. (2025). Exploring the impact of students’ prompt engineering on GPT’s performance: A blockchain-focused automatic term extraction experiment. Electronics, 14(11), Article 2098. https://doi.org/10.3390/electronics14112098 ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀OECD. (2026). OECD digital education outlook 2026: Exploring effective uses of generative AI in education. https://doi.org/10.1787/062a7394-en ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Poulain, H. (réal.), Casilli, A. A., & Goetz, J. (scén.) (2025). Les sacrifiés de l’IA [documentaire]. France Télévisions / StoryCircus. Diffusé le 11 février 2025 sur France 2. Disponible sur https://www.france.tv ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Qu, Y., & Wang, J. (2025). The impact of AI guilt on students’ use of ChatGPT for academic tasks: Examining disciplinary differences. Journal of Academic Ethics, 23, 2087–2110. https://doi.org/10.1007/s10805-025-09643-x ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Stephenson, R., & Armstrong, C. (2026). Student generative AI survey 2026 (Report 199). Higher Education Policy Institute. https://www.hepi.ac.uk/reports/student-generative-ai-survey-2026/ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀Tian, J., & Zhang, R. (2025). Learners’ AI dependence and critical thinking: The psychological mechanism of fatigue and the social buffering role of AI literacy. Acta Psychologica, 260, Article 105725. https://doi.org/10.1016/j.actpsy.2025.105725