AI and the Future of Education

For two centuries, formal education has been organised around one scarce resource, the attention of an expert teacher. Artificial intelligence is loosening that constraint. The institutions that work out what to do with abundant, personalised support, and what to keep firmly in human hands, will shape the next era of Australian education.

This paper sets out where artificial intelligence is taking education and what leaders in vocational and higher education should do about it. It draws on peer reviewed research, on Australian policy and regulation, and on international evidence. It is written for the people who have to make decisions now, rather than for those who can wait for the picture to settle.

The starting point is an uncomfortable one. Education already has the highest rate of generative AI adoption of any industry, yet a large share of teachers and students have had no training in how to use it well (Microsoft Education, 2025). Adoption has run ahead of readiness, and the gap between what learners are doing and what institutions have actually governed continues to widen.

This is not a story about machines replacing people. The most thorough Australian study to date finds that generative AI augments far more work than it automates, and that demand is rising for exactly the higher order human skills that education exists to build (Jobs and Skills Australia, 2025). The opportunity is to move education away from a model built on the scarcity of expert attention and towards one of abundant, personalised and well governed support.

Australia is well placed to lead. It has a mature vocational sector, a respected university system, a national qualifications architecture, and live reform through the Universities Accord. The advantage is real, but it is not permanent. It rewards the institutions that move with purpose, and it punishes drift.

The reasoning behind this paper is simple. An institution that governs AI well will graduate people the labour market is short of, protect the integrity of its qualifications, and free its teachers to do their most valuable work. An institution that does not will watch usage spread without guardrails, integrity erode quietly, and trust in its credentials come into question. The difference between those two futures is not budget or technology. It is the quality and the speed of the decisions taken now.

This paper sets the direction. It stops short of the detailed design, because that design depends on each institution's mission, students and appetite for risk, and because it is the work Origae does in partnership with leaders. What follows is the map, not the territory.

Something unusual has happened. A general purpose technology has reached classrooms, lecture theatres and training workshops faster than any institution planned for. Students adopted generative AI within months. Governance frameworks are still being drafted.

The numbers tell the story plainly. Education leads every other industry in generative AI adoption, yet roughly 45% of educators worldwide, and 52% of students in the United States, report receiving no AI training at all (Microsoft Education, 2025). The tools are in everyone's hands. The capability to use them well is not.

Formal systems are responding, but unevenly. Two thirds of countries now offer or plan to offer school computer science, double the share in 2019, and 81% of United States computer science teachers agree that AI should be part of foundational computer science education. Fewer than half feel equipped to teach it (Stanford HAI, 2025). Intent has run ahead of capability at every level of the system.

Regulation is further behind still. UNESCO issued the first global guidance on generative AI in education in 2023, and warned that the absence of national regulation in most countries leaves the data privacy of users unprotected and institutions exposed (UNESCO, 2023). When the OECD examined the question in 2023, none of the countries it reviewed had introduced specific regulation for generative AI in education, and only a small number had published voluntary guidance (OECD, 2023a).

There is a deeper pattern worth naming. New technologies in education have always been adopted by individuals long before they are governed by institutions, and generative AI has compressed that lag from years into months. Students reach for the most capable tool available. Policy committees meet on a slower clock. The result is a period in which the most consequential decisions about AI inside an institution are being made informally, by thousands of individual users, in the absence of clear direction.

This is the moment to be honest about the gap. The risk is not that AI arrives, because it has already arrived. The risk is that institutions allow usage, policy and capability to drift apart until a crisis forces a defensive and reactive response. A single high profile integrity failure, a privacy breach through an ungoverned tool, or a cohort of graduates who never learned to use AI responsibly can each undo years of reputation. Leadership here means closing the gap on your own terms, while there is still time to design rather than react.

To govern AI well, leaders need a clear picture of what it actually does in education, beyond the noise. On this point the research is consistent.

A systematic review of 138 studies found that the dominant applications gather into a few areas, namely personalised learning, intelligent tutoring, assessment, and administration (Crompton & Burke, 2023). An earlier and foundational review reached the same conclusion, and added a warning that still holds. Educators are largely absent from the research itself, which means the technology has often been designed around learners and systems rather than around the professionals who teach (Zawacki-Richter et al., 2019).

Large language models change the texture of all of these. They make personalised feedback, adaptive tutoring and learning support available at a scale that was simply not possible before, while introducing real risks around misinformation, academic integrity and equal access (Kasneci et al., 2023). The same capability that helps a struggling student at midnight can also produce a confident falsehood, and the two cannot be separated. They are properties of the same technology, to be managed rather than removed.

What is genuinely new about this generation of AI is its breadth. Earlier educational technologies were narrow, a tutoring system for algebra, or an adaptive platform for one course. Generative AI is general. The same tool drafts an essay, explains a concept, writes code, summarises a reading and answers a question, across every discipline at once. That breadth is why it spread so quickly, and it is why a piecemeal response that treats it as just another tool, in just one subject, will always be a step behind.

It helps to think in two horizons. In the near term, AI shows up as an assistant. It drafts feedback, generates practice material, answers routine questions and eases administration. It sits beside existing practice and makes it faster. In the medium term it becomes more capable of acting on its own, through adaptive learning environments that respond to each student in real conditions, tutors available outside class hours, and an emerging AI native institution whose curriculum, assessment and support are designed around abundant intelligence rather than bolted onto scarcity.

The distinction matters for investment. Near term gains are real but incremental, and most institutions are already capturing some of them informally. The larger prize, and the larger risk, sits in the medium term, where AI reshapes not only how tasks are done but what a course, an assessment, or a qualification fundamentally is. Institutions that optimise only the near term will end up with faster versions of practices that the medium term has already made obsolete.

Neither horizon is fixed. The shape AI takes in any institution is a choice about which tasks to delegate, which to support, and which to protect. The research is clear that AI can tutor, mark, personalise and predict. It is far less clear, and far more important, what an institution decides to keep as the work of people. That choice is the substance of strategy, and it is where direction matters more than tooling.

For students, AI is best understood as support rather than substitution. Used well, it offers personalised feedback, explanation on demand, and adaptive help that a single teacher across thirty learners could never provide (Kasneci et al., 2023). Used carelessly, it offers a shortcut around the very thinking that learning is meant to develop.

This makes a new set of literacies essential. Students now need to prompt a model clearly, to check its output against trustworthy sources, and to weigh confident machine written text with a critical eye. These are not technical niceties. They are the contemporary form of critical thinking, and they belong in the core curriculum rather than in an optional workshop.

It is worth being precise about what these literacies are. Prompting is the skill of framing a problem clearly enough for a machine to help with it, which is itself a form of disciplined thinking. Verification is the habit of treating AI output as a draft to be checked rather than an answer to be trusted, and it matters most when the model is at its most fluent and most confident. Critical evaluation is the judgement to know when AI is the right tool at all, and when a task calls for unaided human reasoning. None of these can be taught once. They are developed through repeated and supported practice across a whole program, in the same way that academic writing or numeracy is built.

There is also a comfortable assumption that needs dispelling. Students who are fluent consumers of digital media are not automatically careful users of AI. Ease with a tool is not the same as wisdom in its use, and treating young people as already AI literate is one of the more dangerous assumptions an institution can make, because it leaves the most important capability unexamined and untaught.

Integrity is the hardest edge. A survey of 1,217 people across 76 countries found that concern about the potential of generative AI to enable academic dishonesty features strongly, and that views are shaped heavily by cultural context rather than settling into a single global consensus (Yusuf et al., 2024). Detection on its own is not the answer, because it is difficult and unreliable, so the response has to combine clear policy, genuine training, and assessment that is harder to fake (Cotton et al., 2024). The goal is not to win an arms race against the tools. It is to make authentic learning the path of least resistance.

Equal access deserves the same attention. Where capable AI is spread unevenly, by device, subscription, language or confidence, it can widen the very gaps that education exists to close. A student who can afford the best tools, and who arrives already confident in using them, pulls further ahead of a student who has neither. Left unmanaged, AI becomes another axis of advantage rather than a leveller. The UNESCO guidance is explicit that generative AI has to be deployed in a way that puts people at the centre and protects learners rather than exposing them (UNESCO, 2023). Access is not a footnote to the AI strategy. It is part of its purpose, and an institution that provides fair access to capable and well governed tools turns a potential source of inequality into a genuine equaliser.

If there is a single story in education AI that is seldom told, it is the opportunity for teachers and trainers. Much of the profession's load is not the irreplaceable human work of teaching. It is the surrounding effort of drafting, marking, documenting and administering. This is precisely where AI offers relief.

Feedback is the clearest example. Timely and specific feedback is one of the most powerful drivers of learning, and one of the hardest things to deliver at scale. AI makes it possible to give students more frequent and more detailed feedback as they work, which frees teachers to apply their judgement where it counts, on the conversations, the mentoring, and the design work that only a person can do.

The economics here are genuinely new. For most of the history of mass education, the cost of personal attention forced a trade off, in which feedback was either timely, or detailed, or able to scale, but rarely all three. AI relaxes that trade off for the routine layers of feedback, the surface errors, the structure, and the fit to a brief, so that a teacher's scarce attention can be reserved for the layers that need human judgement, such as originality, reasoning, and the needs of a particular learner. The aim is not to remove the teacher from feedback. It is to move their effort to where it is worth the most.

This reframes workload rather than simply cutting it. The opportunity is to redirect the hours now consumed by repetitive marking and documentation towards the parts of teaching that are hardest to automate and most valuable to students. Institutions that bank the saved time as a pure efficiency, without reinvesting it in the human work, will capture the smaller share of the benefit and miss the larger one.

Assessment is where this matters most. The expert review commissioned by the higher education regulator concluded that generative AI has made existing assessment problems worse, and that reform has to be systemic and run across whole programs, rather than a series of patches to individual tasks (Lodge et al., 2023). For teachers, that reframes the work. The task is not to police every assignment. It is to redesign assessment so that human judgement is built into the system.

None of this works without keeping people in the loop. The evidence has long noted that educators are too often designed out of AI systems rather than into them (Zawacki-Richter et al., 2019). The institutions that get this right will treat teachers as the designers and decision makers of their AI supported practice, and will put real investment in capability behind them.

Education does not train people for today's economy. It trains them for the one they will enter. The signals from the labour market are now clear enough to act on, and they point in a consistent direction.

Worldwide, the World Economic Forum projects that by 2030 around 170 million jobs will be created and 92 million displaced, a net gain of about 78 million, while 39% of the core skills workers need will change and 85% of employers plan to prioritise upskilling their workforce (WEF, 2025). The signal is not mass unemployment. It is a mass transformation of what work requires.

The Australian picture is more reassuring still, and more useful for setting strategy. A whole of labour market study by Jobs and Skills Australia found that only around 4% of Australian roles face a high risk of automation, that nearly 80% face a low risk, and that about 49% face moderate augmentation, with generative AI far more likely to support human work than to replace it (Jobs and Skills Australia, 2025). The same study records rising demand for digital literacy and for higher order human skills such as critical thinking, communication and adaptability.

This is an important correction to the anxiety that dominates the public conversation. For Australian institutions, the evidence does not support a defensive crouch against a wave of automation. It supports a confident investment in the capabilities that make people more effective alongside AI. The real risk is not that machines take the jobs students are training for. It is that graduates enter those jobs unable to work well with the tools their employers already use.

For teachers and curriculum designers, this resolves a strategic question. The durable graduate capabilities are the ones that complement AI, namely judgement, communication, ethical reasoning, collaboration, and the ability to direct and check AI rather than be replaced by it. Policymakers will increasingly need consistent measures to track these human capabilities against advancing AI (OECD, 2023b), and the institutions that build them on purpose will graduate the people the labour market is actively short of.

There is a particular lesson in the Australian finding on augmentation. Because the main effect of generative AI on Australian work is to support rather than replace, the most employable graduate is not the one who avoids AI, nor the one who defers to it, but the one who can direct it, set the task, judge the output, and take responsibility for the result. That capability is built on disciplinary knowledge, not in place of it. You cannot judge AI output in a field you do not understand.

The implication is not to abandon disciplinary depth in favour of generic future skills. It is to fuse them, so that deep knowledge is exercised through distinctly human capability and amplified by the fluent and responsible use of AI. For curriculum designers, the practical question becomes which capabilities each program should build deliberately, and how to make the responsible use of AI a visible and assessed part of the discipline rather than an unspoken workaround.

The Australian qualifications architecture is a genuine national asset, and it was built for a world of longer and more stable credentials. As skills change faster, the architecture itself has to adapt without losing the rigour that makes Australian qualifications trusted.

The review of the Australian Qualifications Framework was direct. The framework was not designed to accommodate shorter forms of credential, and the review recommended recognising micro credentials for credit towards qualifications, and revisiting how the framework describes contemporary capabilities such as digital literacy (Noonan et al., 2019). That recommendation has only become more pressing.

Micro credentials are central to the answer. The National Microcredentials Framework defines them as short and certified units of learning aimed at a specific skill or capability, sitting below the scale of a full qualification, with national standards for recognition, portability and credit (Department of Education, Skills and Employment, 2022). Done properly, they let learners build and stack verified capability as the world changes, which makes them the natural unit for a faster moving skills economy.

The reason this matters now is timing. When the useful life of a specific skill shortens, the full multi year qualification remains essential for deep formation, but on its own it is too slow to keep a whole workforce current. Micro credentials fill that gap. They let working Australians add verified and recognised capability in months rather than years, and they let institutions respond to emerging demand without waiting for a full qualification to be designed and accredited. The risk to manage is credibility, because a micro credential is only valuable if it is genuinely recognised and genuinely rigorous, which is exactly what the national framework exists to protect.

The Universities Accord places this within a system wide ambition. It contains 47 recommendations, a target of 80% of working age Australians holding a tertiary qualification by 2050, up from around 60% today, modular and micro credential pathways, and a new Australian Tertiary Education Commission to steer the reform (Australian Universities Accord Panel, 2024). Qualifications reform is not a side project. It is national policy.

The comparison below sets out how the three pillars of post secondary recognition relate to one another. The point is not hierarchy but complement. Each does something the others cannot.

Integrity sits underneath all of it. In vocational education, the 2025 Standards for Registered Training Organisations, in effect from 1 July 2025, require that people make the competency decisions. Providers cannot rely on AI to determine competency, and they have to verify that learners have completed their own work (ASQA, 2025). In higher education, the regulator's position is that assessment reform has to be systemic and run across whole programs (Lodge et al., 2023). The two regulators have, in effect, drawn the same line. AI may assist the process, but human judgement certifies the outcome.

This is a more workable principle than it first appears. It does not ask institutions to ban AI from assessment, which is neither possible to enforce nor desirable. It asks them to be deliberate about where in a program the load bearing judgements sit, and to protect those points, through assessment design, supervision, oral defence, or evidence of process, so that a qualification continues to mean what it claims. The work is to find those points across a whole program and protect them by design, rather than to treat every individual task as a frontier to be policed.

Recognition of prior learning deserves a specific mention, because it sits exactly at this intersection. Done by hand, it is slow and inconsistent. Done with AI support against the relevant standards, it can be faster and more consistent, provided the competency decision remains a human one, as the Standards require. It is a clear example of the broader pattern. AI carries the analysis, and the qualified assessor carries the decision.

What should an institution actually do? At the level of direction, though not yet at the level of detailed design, the evidence points to a clear and disciplined response.

Start with posture. An analysis of the top 100 universities in the United States found that none banned generative AI outright, and that the common stance was open but cautious, with ethical use, data privacy and accuracy as the governing concerns (Wang et al., 2024). That is the right default for Australian institutions as well. Neither prohibition, which fails, nor a free for all, which exposes, but governed enablement.

The framing matters as much as the policy. Generative AI is at once a disruptive threat and a chance to reform, and institutions that treat it only as a threat to be contained will forfeit the gains (Lim et al., 2023). The leadership task is to hold both truths at once, to manage the genuine risks to integrity and privacy without giving up the chance to teach, assess and support students better than was ever possible before.

In practice, institutions tend to fail in one of two predictable ways. The first is prohibition, a defensive policy that drives AI use underground, where it cannot be observed, taught or governed, while competitors build capability in the open. The second is a free for all, a permissive posture that mistakes the absence of rules for the presence of strategy, and discovers the cost only when integrity or privacy fails in public. Governed enablement is the deliberate middle path, and it is harder than either extreme, because it requires actual design rather than a single decision to allow or forbid.

From this, four principles follow at the level of direction.

• Governance. Set clear policy across the institution on acceptable use, data privacy and accountability, and make sure it is lived rather than merely published.
• Capability. Invest in the AI literacy of staff and students as deliberately as in any other strategic capability, because the training gap, not the technology, is the binding constraint.
• Assessment security. Treat the redesign of assessment across whole programs as a core institutional project, protecting the judgements that certify a qualification.
• Partnership. Move faster and more safely by working with partners who have done this before, rather than rebuilding every capability in house.

A word on capability, because it is the most commonly underestimated of the four. The binding constraint on an institution's AI strategy is rarely the technology and rarely the policy. It is whether staff and students can actually use AI well, and whether they trust the institution's intent in deploying it. The training gap visible in the adoption data is not a temporary inconvenience to be waited out. It is the central obstacle, and it closes only through deliberate and sustained investment in people. Tools without capability produce noise. Capability without tools produces frustration. The two have to advance together.

The honest message to leaders is this. The direction is now knowable, but the design is not generic. Two institutions with the same principles will need very different implementations, shaped by their student mix, their disciplines, their exposure to regulation, their culture, and their appetite for risk. Getting that implementation right, quickly, safely, and in a way that people will actually adopt, is the real work. It is also the work most likely to go wrong when it is attempted from first principles, without the benefit of having designed it before.

Look out three to five years and a credible picture comes into view. AI literacy is embedded across the curriculum. Assessment has been redesigned across whole programs, so that qualifications stay trustworthy. Micro credentials are a recognised and stackable pathway. Teachers spend more of their time on the human work of teaching, and students receive personalised support as a matter of course. None of this needs science fiction. It needs deliberate execution.

A high level roadmap helps leaders sequence the journey without pretending it is a formula.

• Foundations. Establish governance, build a baseline of AI literacy for staff and students, and form an honest assessment of where usage and exposure already sit.
• Integration. Embed AI into teaching, feedback and support, redesign assessment across whole programs, and stand up micro credential pathways with real recognition.
• Transformation. Move towards an operating model in which the curriculum, the assessment and the support are designed around abundant and well governed intelligence.

The sequence matters more than the speed. Institutions that rush to transformation without first establishing governance and capability tend to stall, because they have built sophisticated practice on a foundation that cannot bear scrutiny. Institutions that invest only in governance, and never progress to integration and transformation, end up with policies that limit risk but capture none of the upside. The art is to move through the stages deliberately, fast enough to lead and careful enough to last, and the right pace depends on each institution's starting point.

Success has a recognisable shape. For an Australian institution, it looks like graduates the labour market is short of, qualifications that hold their integrity under scrutiny, staff who feel supported rather than displaced, and a reputation for getting AI right while peers are still debating it. It also looks like an institution that has made its choices explicit, one that can say clearly what it has delegated to machines, what it has chosen to support, and what it has protected as the work of people. The international evidence and the Australian policy settings point the same way. The conditions for leadership are in place.

The deeper truth is that the institutions which lead will not be the ones with the most tools. They will be the ones that made the clearest choices about what to delegate to machines and what to keep as the work of people, and then carried those choices out with discipline.

The rest of this paper sets out the evidence. What follows is more personal. These are the strategies I would put in place if the decision were mine, drawn from years of building AI products for Australian classrooms. They are my own views, offered as a practitioner rather than as settled policy.

The first thing I would accept is that the ground is moving. AI tools change week by week, and any strategy that assumes a fixed set of them will be out of date before it is approved. So I would build a strategy that works as a foundation now and is designed to evolve, with clear principles that hold steady even as the technology underneath them changes.

I would begin with an adoption phase whose only job is to bring everyone along, students and staff together. Capability and trust are built, not announced. Before an institution asks its people to redesign assessment or rethink a course, it has to give them the time, the training and the safe space to use these tools well (Microsoft Education, 2025).

For teachers, I would put AI agents to work on the parts of the job that drain time without developing students. A well prompted agent can draft a lesson plan, propose an assessment aligned to the learning outcomes, generate practice questions and build a marking guide in minutes, which leaves the teacher to do the shaping, the judgement and the teaching (Crompton & Burke, 2023).

I would also trial what I think of as classroom cobots, AI assistants that work alongside a teacher rather than replace them. Their value is clearest in large classes, where one teacher cannot give thirty students attention at once. A cobot can field routine questions, surface the students who are struggling, and let the teacher move from broadcasting towards something much closer to assisted, individual learning.

Online learning is where I would be most ambitious. The next generation of online classes should respond to the learner, not simply deliver content to them. That means designing for emotion and for neurodiversity, so that a class can sense when a student is disengaged or overloaded and adjust, and so that learners who think and process differently are planned for rather than accommodated as an afterthought. This is the thinking behind our own work on Affectly.

I would hand the repetitive load to AI so that teachers can spend their hours on the creative parts of the classroom. Faster grading and quicker administration are not the goal in themselves. They are how we buy back the time for the experiences only a person can create. Our app GradiumX exists for exactly this, producing rubric aligned feedback in seconds while a qualified assessor keeps control of the decision (Lodge et al., 2023).

Integrity has to be part of the same system. Where a piece of work has to be the student's own, I would give teachers practical tools to assess whether AI has likely been used, treated as one signal among many rather than as a verdict. Our app PenForensic was built for this, grounded in peer reviewed markers, to support a human judgement and not to replace it (Cotton et al., 2024).

The direction is shared, but adoption looks different in each sector. These are two priorities for each major teaching area, chosen because they matter most for what graduates can do and for the integrity of the qualification.

The general principles hold across the whole institution. Where they meet a particular discipline, though, they take on a specific shape. A creative arts faculty and a nursing school are both protecting human judgement, but they are protecting very different judgements, in very different ways. The two moves below are where each area should start.

This paper sets the direction. Turning that direction into something that works for your institution is a different kind of work, and it is where Dr Nick Patterson and Origae can help. He builds solutions and strategies to support your institution, grounded in products made for the realities of Australian education.

Each product below was built to solve a real problem in vocational and higher education, and each can be shaped to fit your institution's context, mission and appetite for risk.

• Affectly, adaptive online learning that designs for emotion and neurodiversity so classes respond to the learner.
• PenForensic, academic integrity analysis, which supports the authenticity judgements that assessment reform depends on.
• GradiumX, an AI grading assistant for vocational and higher education that keeps human assessors in control of the decision.
• PriorLeap, AI assisted recognition of prior learning against ASQA aligned qualifications.
• TradeItUp, a platform for TAFE apprentices and the people who train them.
• Timba, intelligent timetabling that removes administrative load from the institution.

About the author. Dr Nick Patterson is an experienced education sector expert and technologist. He holds a PhD in cyber security focused on AI anomaly detection, and is an Alfred Deakin medallist, a published author, and an AI strategist and builder. He works with the education sector, industry and government to shape an AI plan or product.

Methodology. The evidence in this paper is drawn from peer reviewed research and from primary policy sources, including Australian regulators, Australian government reviews, and major international bodies. Sources were chosen for their authority and their relevance to the Australian vocational and higher education context, and each claim is attached to a source that supports it. Every figure was checked against the original source. References follow the Harvard system, and the full list appears below.

1. Australian Skills Quality Authority (2025) Standards for Registered Training Organisations (RTOs) 2025. Australian Skills Quality Authority (ASQA). https://www.asqa.gov.au/for-providers/standards-for-RTOs
2. Australian Universities Accord Panel (2024) Australian Universities Accord: Final Report. Department of Education, Australian Government. https://www.education.gov.au/australian-universities-accord/resources/final-report
3. Cotton, D.R.E., Cotton, P.A. & Shipway, J.R. (2024) Chatting and cheating: ensuring academic integrity in the era of ChatGPT. Innovations in Education and Teaching International, 61(2), pp. 228 to 239. https://doi.org/10.1080/14703297.2023.2190148. https://www.tandfonline.com/doi/full/10.1080/14703297.2023.2190148
4. Crompton, H. & Burke, D. (2023) Artificial intelligence in higher education: the state of the field. International Journal of Educational Technology in Higher Education, 20, Article 22. https://doi.org/10.1186/s41239-023-00392-8. https://link.springer.com/article/10.1186/s41239-023-00392-8
5. Department of Education, Skills and Employment (2022) National Microcredentials Framework. Australian Government, Canberra. https://www.education.gov.au/higher-education-publications/resources/national-microcredentials-framework
6. Jobs and Skills Australia (2025) Our Gen AI Transition: Implications for Work and Skills. Australian Government, Canberra. https://www.jobsandskills.gov.au/publications/generative-ai-capacity-study-report
7. Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., Krusche, S., Kutyniok, G., Michaeli, T., Nerdel, C., Pfeffer, J., Poquet, O., Sailer, M., Schmidt, A., Seidel, T., Stadler, M., Weller, J., Wendler, J. & Kasneci, G. (2023) ChatGPT for good? On opportunities and challenges of large language models for education. Learning and Individual Differences, 103, Article 102274. https://doi.org/10.1016/j.lindif.2023.102274. https://www.sciencedirect.com/science/article/abs/pii/S1041608023000195
8. Lim, W.M., Gunasekara, A., Pallant, J.L., Pallant, J.I. & Pechenkina, E. (2023) Generative AI and the future of education: Ragnarök or reformation? A paradoxical perspective from management educators. The International Journal of Management Education, 21(2), Article 100790. https://doi.org/10.1016/j.ijme.2023.100790. https://doi.org/10.1016/j.ijme.2023.100790
9. Lodge, J.M., Howard, S., Bearman, M., Dawson, P. & Associates (2023) Assessment Reform for the Age of Artificial Intelligence. Tertiary Education Quality and Standards Agency (TEQSA), Melbourne. https://www.teqsa.gov.au/guides-resources/resources/corporate-publications/assessment-reform-age-artificial-intelligence
10. Microsoft Education (2025) 2025 AI in Education: A Microsoft Special Report. Microsoft Corporation. https://www.microsoft.com/en-us/education/blog/2025/08/ai-in-education-report-insights-to-support-teaching-and-learning/
11. Noonan, P., Blagaich, A., Kift, S., Lilly, M., Loble, L., More, E. & Persson, M. (2019) Review of the Australian Qualifications Framework: Final Report. Department of Education, Skills and Employment, Australian Government, Canberra. https://www.education.gov.au/quality-and-legislative-frameworks/resources/review-australian-qualifications-framework-final-report-2019
12. OECD (2023) AI and the Future of Skills, Volume 2: Methods for Evaluating AI Capabilities. OECD Publishing, Paris. https://doi.org/10.1787/a9fe53cb-en. https://www.oecd.org/en/publications/ai-and-the-future-of-skills-volume-2_a9fe53cb-en.html
13. OECD (2023) OECD Digital Education Outlook 2023: Towards an Effective Digital Education Ecosystem. OECD Publishing, Paris. https://doi.org/10.1787/c74f03de-en. https://www.oecd.org/en/publications/oecd-digital-education-outlook-2023_c74f03de-en.html
14. Stanford Human-Centered Artificial Intelligence (2025) AI Index Report 2025. Stanford University, Stanford CA. https://hai.stanford.edu/ai-index/2025-ai-index-report/education
15. UNESCO (2023) Guidance for Generative AI in Education and Research. United Nations Educational, Scientific and Cultural Organization, Paris. https://www.unesco.org/en/articles/guidance-generative-ai-education-and-research
16. Wang, H., Dang, A., Wu, Z. & Mac, S. (2024) Generative AI in higher education: seeing ChatGPT through universities' policies, resources, and guidelines. Computers and Education: Artificial Intelligence, 7, Article 100326. https://doi.org/10.1016/j.caeai.2024.100326. https://www.sciencedirect.com/science/article/pii/S2666920X24001292
17. World Economic Forum (2025) The Future of Jobs Report 2025. World Economic Forum, Geneva. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
18. Yusuf, A., Pervin, N. & Román-González, M. (2024) Generative AI and the future of higher education: a threat to academic integrity or reformation? Evidence from multicultural perspectives. International Journal of Educational Technology in Higher Education, 21, Article 21. https://doi.org/10.1186/s41239-024-00453-6. https://educationaltechnologyjournal.springeropen.com/articles/10.1186/s41239-024-00453-6
19. Zawacki-Richter, O., Marín, V.I., Bond, M. & 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, Article 39. https://doi.org/10.1186/s41239-019-0171-0. https://link.springer.com/article/10.1186/s41239-019-0171-0