AI and the
Future of Education

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.1 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.3 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.4 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.5

There is a deeper pattern here 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.

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.7

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.8

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 not possible before, while introducing real risks around misinformation, academic integrity and equal access.9 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.

It helps to think in two horizons. In the near term, AI is an assistant. It drafts feedback, generates practice material, answers routine questions and eases administration. In the medium term it becomes more capable of acting on its own, through adaptive learning environments, 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 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.

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.9 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.

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.

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.10 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.11

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. UNESCO's 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.4 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.

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.12 For teachers, 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.8 The institutions that get this right will treat teachers as the designers of their AI supported practice, with real investment in capability behind them.

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.6 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.2 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. 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 train for. It is that graduates enter those jobs unable to work well with the tools their employers already use.

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. That capability is built on disciplinary knowledge, not in place of it. You cannot judge AI output in a field you do not understand. Policymakers will increasingly need consistent measures to track these human capabilities against advancing AI.19

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.13

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.14 Done properly, they let learners build and stack verified capability as the world changes.

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.15 Qualifications reform is not a side project. It is national policy.

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.16 In higher education, the regulator's position is that assessment reform has to be systemic and run across whole programs.12

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.

Recognition of prior learning 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. AI carries the analysis, and the qualified assessor carries the decision.

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.17 That is the right default. Neither prohibition, which fails, nor a free for all, which exposes, but governed enablement.

Generative AI is at once a disruptive threat and a chance to reform, and institutions that treat it only as a threat will forfeit the gains.18 Institutions tend to fail in one of two ways. Prohibition drives AI use underground, where it cannot be observed, taught or governed. A free for all mistakes the absence of rules for the presence of strategy. Governed enablement is the deliberate middle path, and it is harder than either extreme.

From this, four principles follow at the level of direction. Governance, clear policy across the institution on acceptable use, data privacy and accountability, lived rather than merely published. Capability, investment in the AI literacy of staff and students, because the training gap, not the technology, is the binding constraint. Assessment security, the redesign of assessment across whole programs. Partnership, working with those who have done this before rather than rebuilding every capability in house.

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 students, their disciplines, their exposure to regulation, and their appetite for risk. Getting that implementation right, quickly and safely and in a way people will adopt, is the real work.

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.1

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.7

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.12

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.11

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.