It begins, often, with an email from above: a new AI-enabled marking assistant is being piloted to ease workload pressures. A tool that can provide instant formative feedback. A detection system designed to safeguard academic integrity. A dashboard to help “flag” suspicious patterns in student submissions. The language is careful, optimistic, managerial - “supporting staff,” “enhancing consistency,” “freeing up time for teaching.” For overburdened educators, the promise is tempting: let the system shoulder the repetitive load so that teachers can focus on more “human” aspects of their work.

This is a familiar script in contemporary higher education, where the growing administrative and emotional burdens of massified systems, funding constraints, and performance pressures converge (Brown and Carasso, 2013; Yorke, 2003). As Yorke (2003) argues, formative assessment must be embedded meaningfully within pedagogic practice - not treated as a technical add-on - if it is to genuinely support student learning and curriculum coherence. Assessment, when framed only as a managerial obligation or measured through automated throughput, loses its potential as a site of pedagogical dialogue and professional judgement. Assessment is increasingly framed not as a relational and pedagogical act, but as a site of technical optimisation. In this context, artificial intelligence is presented not only as a tool but as a fix - a way to scale judgement.

But what happens when the processes of assessment, feedback, and academic integrity are increasingly delegated to computational systems? What does it mean when the work of making meaning from student work is handed over to an algorithm? In the name of efficiency, we risk eroding the very heart of education: the interpretive, situated, and dialogic encounter between learner and teacher.

To mark is to judge. And to judge is not merely to measure, but to attend, to interpret, to care. As Gert Biesta (2010) reminds us, educational judgement is not reducible to correctness or compliance; it requires the exercise of pedagogical wisdom, the ability to respond to the unpredictable and the particular. When assessment is automated - whether through AI detection tools, feedback generators, or risk prediction dashboards - this mode of judgement is flattened into something that can be computed.

This shift from judgement to scoring reflects a deeper transformation in how education is conceptualised. It signals a move toward what Williamson (2017) describes as the “datafication” of learning - where knowledge, engagement, and progress are all rendered as digital traces, ready for analysis. In this paradigm, the value of assessment lies not in understanding the learner, but in producing data points that can be tracked, compared, and acted upon. AI systems, in this view, are not merely tools for learning but instruments of governance.

To be clear, the use of digital tools to support assessment is not inherently problematic. There are contexts in which automation can be useful, especially for large-scale tasks involving clearly defined knowledge domains. But when the logic of automation becomes the default mode - when AI systems are rolled out without sustained pedagogical dialogue - then educators risk becoming mere overseers of systems that increasingly determine what counts, what matters, and what can be known.

The adoption of AI in assessment is not just a technical decision; it is a pedagogical and political one. It asks us to reflect on what we are willing to delegate, and at what cost. Are we supporting teachers, or supplanting them? Are we making assessment more meaningful, or more mechanised? These are not questions that can be answered by efficiency metrics or vendor guarantees. They demand educational judgement.

The rise of AI in assessment is often introduced through the logic of institutional efficiency. In university press releases, vendor pitches, and senior leadership emails, the story is familiar: artificial intelligence will ease workload, accelerate feedback, reduce marking burdens, improve consistency, and help overstretched academics cope with large student cohorts. In a sector under considerable pressure, these promises are hard to resist.

The structural challenges are real. Higher education systems around the world have undergone a process of massification, enrolling more students than ever before without a commensurate increase in teaching capacity or academic staffing (Tight, 2019).

Marketisation has imposed new forms of accountability and performance management, while funding pressures have driven institutions to pursue cost-saving measures, often through automation (Brown and Carasso, 2013). Added to this are rising student expectations for timely feedback, detailed assessment criteria, and individualised support - all of which create mounting demands on staff time and energy.

Within this context, AI tools are positioned not as pedagogical interventions, but as operational solutions. Automated feedback systems promise rapid turnaround. Detection tools promise integrity at scale. Predictive analytics promise proactive support. Crucially, these technologies are not just tools for teachers - they are tools for systems. They offer a vision of assessment that is more uniform, more trackable, and more readily managed by central units. In this framing, AI is not merely helping teachers assess - it is helping institutions administer assessment.

Yet this framing deserves closer scrutiny. Efficiency is never neutral. It is always defined in relation to particular goals and values. As Holmes and Porayska-Pomsta (2022) argue, the pursuit of efficiency in educational systems - particularly through AI - can displace pedagogical values such as dialogue, care, and contextual judgement, privileging automation and standardisation over relational practice. When efficiency becomes the overriding metric for adopting new tools, we risk instrumentalising education - treating assessment not as a dialogue, but as a process to be completed.

Moreover, the push for automation often obscures what is lost when human judgement is replaced with computational scoring. Tools designed to save time may erode opportunities for nuanced engagement, marginalise atypical learners, or reinforce what Knox, Williamson, and Bayne (2020) call “machine behaviourism” - a reductive logic that treats learning as observable output, and good student work as that which aligns with pre-programmed norms. They can also reconfigure teaching as a form of system oversight, where the academic’s role is to monitor dashboards and check alerts rather than to engage directly with students’ ideas.

The question, then, is not whether universities should seek efficiencies - many already operate under unsustainable conditions - but who defines what counts as efficient, and to what end. Is a tool “efficient” because it reduces labour, or because it supports better learning? Is a process “efficient” because it produces quick feedback, or because it cultivates understanding?

These are not rhetorical questions. They are at the heart of a broader pedagogical and ethical debate about what universities are for, and what assessment ought to be. The danger is not the use of AI itself, but the uncritical acceptance of a narrative in which human judgement becomes the bottleneck, and automation becomes the answer.

Among the most rapidly adopted uses of AI in higher education are tools designed to detect misconduct - plagiarism, contract cheating, and more recently, the use of generative AI in student writing. These tools are often implemented under the banner of academic integrity, with the promise of safeguarding standards in a time of digital uncertainty. But beneath this protective framing lies a more troubling shift: from trust to suspicion, from pedagogy to policing.

The original wave of plagiarism detection tools, such as Turnitin, offered institutions a sense of control over increasingly digital submissions. But as students gained access to more sophisticated content-generation tools - such as ChatGPT and similar large language models - a new generation of detection systems has emerged to identify “AI-written” work. These tools, however, are built not on direct evidence but on statistical inference. They rely on probabilistic indicators such as word frequency, sentence structure, and other linguistic patterns - features that correlate with but do not prove AI authorship. As Liang et al. (2023) demonstrate, such detectors are particularly unreliable when applied to writing by non-native English speakers, who are more likely to be falsely flagged as using AI-generated text, raising serious equity and discrimination concerns.

The epistemic status of such tools is inherently fragile. They produce scores, not certainties. They cannot trace the process by which a text was composed. They cannot distinguish between student revision and machine intervention. And yet, in institutional settings, these tools often take on the appearance of authority. A “high probability” output can be interpreted - by staff or panels - as a sufficient basis for punitive action, especially when policy frameworks have not evolved to reflect the limitations of probabilistic detection. As Bittle and El-Gayar (2025) highlight, the integration of generative AI into academic settings has outpaced the development of institutional policy and staff training, raising concerns that detection tools are being adopted faster than the ethical and procedural frameworks required to govern their use.

This dynamic raises profound concerns for both due process and the pedagogical climate. If students are presumed to be deceivers unless proven otherwise, the trust that underpins meaningful learning relationships is compromised. The classroom becomes a site of surveillance, and the writing process is reduced to a series of forensic signals. As Selwyn and Gilliard (2023) note, such systems embed a logic of platform-driven suspicion: learners are rendered visible primarily to be monitored, and the role of technology becomes one of detection rather than support.

This shift has wider cultural implications. In an environment saturated with detection technologies, students may feel compelled to write defensively, producing work that is designed to pass AI scrutiny rather than express original thought. Pedagogically, this can result in a narrowing of discourse: fewer risks, less experimentation, and a focus on compliance over creativity. Teachers, too, may experience a crisis of confidence - unsure whether to trust student work, or whether they themselves might be held accountable for failing to “catch” misconduct.

It is important to recognise that institutions often adopt such tools under considerable pressure: to maintain standards, to respond to public concern, to assure external regulators. But an overreliance on detection technologies risks reshaping academic integrity as an adversarial posture. Instead of cultivating shared values, we reinforce a culture of defence and evasion.

The work of academic integrity is ultimately relational. It involves creating conditions where students feel ownership of their ideas, where the writing process is visible and supported, and where assessment is meaningful beyond its policing function. AI detection tools may have a role in extreme or clear-cut cases, but when they become central to institutional responses, we must ask: what kind of pedagogical culture are we building? And at what cost?

In the name of speed, scale, and standardisation, artificial intelligence is increasingly being used to automate the generation of feedback on student work. From language models trained to detect writing patterns, to algorithmic scoring based on rubrics and keyword detection, AI systems promise educators rapid, consistent, and data-rich insights into student performance. For universities under pressure to deliver timely feedback and reduce academic workload, the appeal is considerable.

Yet these systems carry significant pedagogical implications. They are often built upon a narrow conception of writing and learning - one that treats texts not as sites of meaning, argument, or voice, but as data structures to be analysed for compliance with predefined norms. Feedback becomes less an act of interpretation and dialogue, and more an exercise in feature extraction. A student’s reasoning is not read; it is parsed. Their argument is not engaged; it is pattern-matched.

This reduction is evident in systems that “score” writing using linguistic proxies such as sentence length, vocabulary sophistication, or syntactic complexity. While these may correlate with certain indicators of academic writing, they are poor substitutes for the messy, situated, and often unpredictable process of constructing an argument. As Boud and Dawson (2021) note, effective feedback is relational and context-sensitive - it depends on the student’s prior learning, their intentions, and their ability to act on what is offered. An algorithm trained on past submissions cannot engage in this kind of pedagogical reasoning.

Similarly, AI-generated comments often rely on keyword detection or rubric alignment - highlighting whether particular concepts were “mentioned,” whether structure follows a recognisable pattern, or whether surface features of academic style have been adhered to. But as Tsai (2022) argues, feedback systems driven by learning analytics often prioritise efficiency and surface-level indicators over student understanding. Without active engagement from learners, these systems can reduce feedback to a transactional exercise - reinforcing compliance rather than fostering critical reflection or academic growth. Students quickly learn what the system values, and shape their work accordingly - sometimes writing to the algorithm rather than to the reader.

This is not a hypothetical concern. Research by Perelman (2014) and later by Zawacki-Richter et al. (2019) found that automated writing evaluation systems often reward verbosity, formulaic structure, and superficial sophistication over clarity or originality. In doing so, they may inadvertently encourage students to adopt strategic behaviours - focusing on style over substance, on pattern over argument.

Such systems also raise concerns about the feedback loop between writing and reflection. Good feedback does more than point out errors or omissions - it opens space for dialogue, for rethinking, for growth. It is not just evaluative but generative. As Wiliam (2018) emphasises, formative assessment must be responsive and contingent on students’ real-time thinking - not pre-scripted or automated. By contrast, many AI-based systems offer judgement without interpretation: comments that are generic, decontextualised, and difficult to act upon. As Carless and Boud (2018) argue, the most meaningful feedback arises from sustained interaction - not one-off annotations from a black box.

To be clear, AI-assisted feedback may have a role in some contexts - particularly as a supplement in large cohorts or for early-stage formative tasks. But when it becomes a substitute for relational engagement, it risks undermining one of the most pedagogically rich moments in the learning process. It reframes feedback not as an invitation to dialogue, but as a data product - efficient, standardised, and ultimately alienating.

The question is not whether AI can generate grammatically correct comments, but whether it can help students develop judgement, voice, and critical awareness. And here, the evidence is far from conclusive. Education is not merely the transmission of standards - it is the cultivation of capacity. And that requires human connection.

At the heart of any meaningful assessment lies an interpretive act. To assess is not merely to score; it is to make a situated judgement about the quality, relevance, coherence, and potential of a student’s work. This act is not mechanical. It draws on disciplinary expertise, contextual understanding, and an attuned sense of the learner’s developmental trajectory. It involves ambiguity, responsiveness, and care. It is, in short, pedagogical judgement.

Gert Biesta (2012) defines pedagogical judgement as the capacity of the educator to make wise, context-sensitive decisions in the service of students’ educational growth. It cannot be reduced to technical procedures or standardised metrics. Instead, it involves what Aristotle called phronesis - practical wisdom grounded in experience and ethics. For Biesta, this is precisely what distinguishes education from training or information transfer: teaching is always a moral and relational encounter, not just a delivery mechanism.

Yet AI systems designed for assessment often operate on an entirely different epistemology. Rather than engaging in interpretation, they function through recognition: identifying patterns in large datasets and assigning probabilistic scores based on correlations. Whether detecting stylistic markers, tagging rubric criteria, or predicting learning outcomes, AI reframes assessment as a computational problem to be optimised. This logic is not necessarily wrong - but it is limited, and when left unexamined, it risks displacing deeper forms of pedagogical reasoning.

The displacement is not merely symbolic. As automation expands, educators increasingly find themselves in the role of system overseers: reviewing algorithmic outputs, adjudicating alerts, and validating auto-generated feedback. Their interpretive labour is not eliminated - but it is restructured and often marginalised. As Knox, Williamson, and Bayne (2020) argue, the logic of “machine behaviourism” reduces education to the management of observable actions, positioning teachers as system supervisors who intervene only when algorithmic processes falter or generate anomalies.

The danger here is twofold. First, the erosion of professional autonomy: educators are asked to conform to the logics of tools they did not design and whose inner workings are often opaque. Second, the deskilling of academic labour: as institutions invest in systems that promise consistency and efficiency, they may devalue the nuanced, slow, and sometimes messy work of making pedagogical judgements.

Moreover, when judgement is outsourced to systems, the educational relationship is reframed. The student is no longer writing to the teacher, but for the algorithm. Assessment becomes less a conversation and more a transaction. This dynamic not only undermines trust but can distort what students learn to value: producing work that aligns with machine-readable standards rather than engaging in deep inquiry or ethical reflection.

This is not to romanticise human judgement. It can be biased, inconsistent, and shaped by institutional norms. But unlike algorithmic scoring, pedagogical judgement can be made accountable through dialogue, revision, and professional reflection. It is rooted in responsibility and relationship. To automate it is not to improve it - but to transform it into something else entirely.

The challenge, then, is not simply to critique AI systems, but to insist on the irreplaceable value of human judgement in education. This means protecting space for interpretation, creating institutional cultures that recognise the expertise of educators, and ensuring that automation enhances rather than erodes professional agency.

If the story of AI in assessment is partly about tools and technologies, it is equally a story of policy, procurement, and institutional governance. The systems now reshaping the everyday practices of teaching and assessment do not enter the university through pedagogical conversation. They arrive through strategic frameworks, digital roadmaps, procurement cycles, and external pressures - often well-meaning, sometimes opaque, and rarely deliberated on by those who must live with their consequences.

University leaders operate in a challenging political economy. They are tasked with balancing budgetary constraints, student satisfaction metrics, quality assurance audits, national performance frameworks, and the reputational demands of international rankings (Brown and Carasso, 2013; Shore and Wright, 2015). The appeal of automation - particularly at scale - is understandable in this context. AI systems promise cost savings, risk reduction, compliance with regulatory standards, and demonstrable outputs in key performance indicators. At a strategic level, they offer a vision of control and consistency in an increasingly unpredictable landscape.

But this is precisely the problem. AI systems are often adopted not as pedagogical tools but as governance technologies (Williamson, 2017). Their function is not simply to support educators, but to align institutional practice with externally defined metrics of productivity and performance. And because these systems are procured and implemented through centralised processes - often led by IT services, digital strategy boards, or commercial partnerships - academic staff may find themselves on the receiving end of decisions they had little opportunity to influence. They are then tasked with integrating these tools into their practice, with or without adequate training, consultation, or ethical reflection.

This can create a form of pedagogical complicity - not in the sense that educators support these systems uncritically, but in the sense that they are positioned as enactors of policy decisions made elsewhere. As Ball (2016) argues, policy is not simply something that happens to institutions; it is enacted through the daily practices of those working within them. When assessment becomes a matter of platform configuration rather than professional dialogue, and when compliance with tool usage becomes a condition of quality assurance, the role of the educator is subtly redefined - from a practitioner of judgement to a manager of process.

It is important to be clear: this is not a critique of individuals. Many leaders, IT specialists, and learning technologists are doing their best within constrained, pressured environments. Their decisions are shaped by multiple, sometimes conflicting, demands. But this makes it all the more important to examine the logics underpinning these decisions. What pedagogical values are being encoded in the tools we buy? What assumptions about learning, judgement, and student behaviour are being baked into our platforms? And who is given a voice in shaping these answers?

There are signs of hope. Some institutions are beginning to include academic staff and students in the procurement process, piloting tools with reflective evaluation, and embedding AI literacy in strategic planning. But these practices are not yet the norm. Too often, the implementation of AI systems is driven by vendor promises and institutional aspirations, rather than critical scrutiny or grassroots dialogue.

To resist the reductive logics of automation, then, is not to dismiss AI outright. It is to insist that any use of AI in assessment must be accountable to pedagogical judgement, institutional ethics, and the lived realities of those who teach and learn. It is to call for governance models that are participatory, transparent, and dialogic - not technocratic, centralised, and opaque. And it is to remember that systems shape practice not only through their features, but through the policies that sustain them.

The challenge posed by AI in assessment is not merely one of technological adoption, but of ethical alignment. If current systems often risk displacing human judgement, flattening feedback, and embedding surveillance logics into the learning process, then what would it mean to design AI for education differently? What would it mean to develop systems that support, rather than constrain, the deep values of teaching and learning?

The first step is to reframe the purpose of AI in assessment. Rather than treating automation as a means of replacing teachers or scaling uniformity, ethically aligned AI begins with the premise that technology should augment rather than substitute pedagogical relationships. This principle - sometimes called human-centred AI or human-in-the-loop design - emphasises that educational decision-making must remain a matter of professional judgement, contextual interpretation, and moral responsibility (Selwyn, 2022; Fjeld et al., 2020).

At the end of this exploration into AI and assessment, one truth remains clear: no matter how sophisticated the technology becomes, it cannot - and should not - replace the ethical, relational, and contextual work of teaching. Pedagogical judgement is not a computational task. It is a human act grounded in responsibility, trust, and attentiveness to learners as whole people, not just data points.

To teach is to judge - not in the narrow, punitive sense, but in the expansive sense of making considered, situated decisions about what is meaningful, what is possible, and what a learner needs in a given moment. As Biesta (2013) reminds us, judgement is the opposite of automation: it is the moment when a teacher chooses to respond, rather than merely react. It is how education maintains its integrity in the face of standardisation and abstraction.

And yet, many current deployments of AI in assessment move us in the opposite direction. They promise efficiency, objectivity, and scale - but often at the cost of relational depth, professional autonomy, and educational nuance. When scoring replaces feedback, when pattern recognition replaces interpretation, and when compliance replaces dialogue, we must ask whether we are still engaged in education, or merely in its simulation.

This is not to suggest that AI has no role to play. On the contrary, it may well support educators by identifying patterns, flagging inconsistencies, or facilitating student reflection. But these roles must remain subordinate to the pedagogical process, not supplant it. AI must serve learning - not subsume it.

Reclaiming the art of judgement, then, is both a professional and civic responsibility. For educators, it means asserting the value of expertise, dialogue, and care in an age that often values automation and scalability. For institutions, it means designing systems and policies that protect time for reflection, reward interpretive work, and create space for pedagogical deliberation. For both, it means asking hard questions about what kinds of learning futures we are building - and who gets to shape them.

This is a moment for collective reflection. How do we want to assess learning? What kind of relationships do we want technology to support? What would it mean to build infrastructures that are pedagogically just, not merely administratively efficient?

As Knox et al. (2020) contend, the ongoing integration of AI into education reflects a broader shift toward “learnification” and “datafication” - paradigms that prioritise measurable behaviours over reflective understanding. Reclaiming judgement, then, becomes a political and pedagogical necessity.

AI will not stop evolving. But neither should our commitment to education as a deeply human endeavour. It is time to reclaim not just control over the tools we use, but clarity about the values we hold. In doing so, we reassert that judgement is not a flaw to be engineered out of assessment - it is its most vital feature.

• Ball, S.J. (2016). The Education Debate (3rd ed.). Bristol: Policy Press.
• Biesta, G. (2021). World-Centred Education: A View for the Present (1st ed.). New York: Routledge. https://doi-org.manchester.idm.oclc.org/10.4324/9781003098331
• Biesta, G.J.J. (2010). Good Education in an Age of Measurement: Ethics, Politics, Democracy (1st ed.). Routledge. https://doi.org/10.4324/9781315634319
• Biesta, G.J.J. (2012). Giving Teaching Back to Education: Responding to the Disappearance of the Teacher. Phenomenology & practice, 6(2), pp.35–49. https://doi.org/10.29173/pandpr19860
• Biesta, GJJ 2013, Beautiful Risk of Education, Taylor & Francis Group, Oxford. Available from ProQuest Ebook Central. [26 May 2025].
• Binns, R. et al. (2018). ‘It’s Reducing a Human Being to a Percentage’; Perceptions of Justice in Algorithmic Decisions. https://doi.org/10.48550/arxiv.1801.10408
• Bittle, K. and El-Gayar, O. (2025). Generative AI and Academic Integrity in Higher Education: A Systematic Review and Research Agenda. Information (Basel), 16(4), p.296. https://doi.org/10.3390/info16040296
• Boud, D. and Dawson, P. (2021) ‘What feedback literate teachers do: an empirically-derived competency framework’, Assessment & Evaluation in Higher Education, 48(2), pp. 158–171. https://doi.org/10.1080/02602938.2021.1910928
• Boyd, D. & Crawford, K. (2012). Critical Questions for Big Data: Provocations for a cultural, technological, and scholarly phenomenon. Information, Communication & Society, 15(5), 662–679. https://doi.org/10.1080/1369118X.2012.678878
• Brown, R. and Carasso, H. (2013). Everything for Sale? The Marketisation of UK Higher Education. London: Routledge.
• Carless, D. and Boud, D. (2018). The development of student feedback literacy: enabling uptake of feedback. Assessment & Evaluation in Higher Education, 43(8), pp. 1315–1325. https://doi.org/10.1080/02602938.2018.1463354
• Fjeld, J., Achten, N., Hilligoss, H., Nagy, A.C. and Srikumar, M. (2020). Principled Artificial Intelligence: Mapping Consensus in Ethical and Rights-Based Approaches to Principles for AI. Berkman Klein Center Research Publication No. 2020-1, Available at SSRN: https://ssrn.com/abstract=3518482 or http://dx.doi.org/10.2139/ssrn.3518482
• Gilliard, C. and Selwyn, N. (2023). “Automated Surveillance in Education”. Postdigital science and education, 5(1), pp.195–205. https://doi.org/10.1007/s42438-022-00295-3
• Holmes, W., & Porayska-Pomsta, K. (Eds.). (2022). The Ethics of Artificial Intelligence in Education: Practices, Challenges, and Debates (1st ed.). Routledge. https://doi.org/10.4324/9780429329067
• Holstein, K. et al. (2022). Equity and Artificial Intelligence in education. In The Ethics of Artificial Intelligence in Education. Routledge, pp. 151–173. https://doi.org/10.4324/9780429329067-9
• Knox, J., Williamson, B. and Bayne, S. (2019) ‘Machine behaviourism: future visions of ‘learnification’ and ‘datafication’ across humans and digital technologies’, Learning, Media and Technology, 45(1), pp. 31–45. https://doi.org/10.1080/17439884.2019.1623251.
• Liang W, Yuksekgonul M, Mao Y, Wu E, Zou J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7) p. 100779. https://doi.org/10.1016/j.patter.2023.100779
• Perelman, L. (2014). When “the state of the art” is counting words. Assessing Writing, 21, pp.104–111. https://doi.org/10.1016/j.asw.2014.05.001
• Selwyn, N. (2019). Should Robots Replace Teachers? AI and the Future of Education. Cambridge: Polity Press.
• Selwyn, N. (2022). Education and Technology: Key Issues and Debates. London: Bloomsbury Academic. Retrieved May 26, 2025, from http://dx.doi.org/10.5040/9781350145573
• Shore, C. and Wright, S. (2015). Governing by Numbers: Audit Culture, Rankings and the New World Order. Social Anthropology, 23(1), pp. 22–28. https://doi.org/10.1111/1469-8676.12098
• Tight, M. (2019). Mass higher education and massification. Higher education policy, 32(1), pp.93–108.
• Tsai, Y.S., 2022. Why feedback literacy matters for learning analytics. arXiv preprint arXiv:2209.00879.
• Wiliam, D. (2018). Embedded formative assessment. Second edition. Bloomington, Indiana: Solution Tree Press.
• Williamson, B 2017, Big Data in Education : The Digital Future of Learning, Policy and Practice. SAGE Publications, Limited, London. ProQuest Ebook Central.
• Yorke, M. Formative assessment in higher education: Moves towards theory and the enhancement of pedagogic practice. Higher Education 45, 477–501 (2003). https://doi.org/10.1023/A:1023967026413
• Zawacki-Richter, O., Marín, V.I., Bond, M. and Gouverneur, F. (2019). Systematic review of research on artificial intelligence applications in higher education – where are the educators?. International Journal of Educational Technology in Higher Education 16, 39. https://doi.org/10.1186/s41239-019-0171-0