AI and Learning: Five Themes Emerging from the Latest Research
Artificial intelligence is moving quickly from novelty to infrastructure in education. The conversation is also beginning to mature.
Recent research and thought leadership from the National Catholic Education Commission (NCEC), the Castlereagh Statement, the Barker Institute and University of Sydney’s rapid literature review, and Pope Leo XIV’s Magnifica Humanitas approach the issue from different directions. One is concerned particularly with Catholic education and national coordination. Another asks how Australia’s education and training system needs to change. The Barker review examines the emerging research on how generative AI actually affects young people’s learning. Magnifica Humanitas begins somewhere deeper: with the dignity and formation of the human person.
Yet read together, some striking common themes emerge.
A central question that emerges or education is no longer simply whether students should use AI. It is what students should continue doing, knowing and becoming in a world where AI is readily available.
The performance paradox
One of the most useful concepts in the Barker review is the “performance paradox”: the possibility that generative AI can improve a student’s immediate performance while weakening the durable learning or independent capability beneath it.
A student might produce a stronger essay, solve more mathematics problems or complete a coding task more successfully with AI. But a better output does not necessarily equate to a more capable learner.
Research reviewed by Barker illustrates the problem. In one mathematics study, students using an unrestricted GPT-4 interface performed better while they had access to AI. Many, however, used it to obtain or copy solutions. When the AI was removed, they performed worse than students who had practised without it. A version configured as a tutor, withholding answers and instead providing hints and responding to students’ attempts, produced a different pattern.
This is a critical distinction.
It suggests that measuring the quality of the finished product may become an increasingly unreliable proxy for learning. The Barker review argues that educators need evidence of understanding, reasoning and judgement, and of whether students can transfer their learning to different contexts.
It encourages educators to ask what cognitive work should the student still be doing.
Not all cognitive offloading is bad
Humans have always offloaded cognitive work. We use calculators, notebooks, maps, search engines and other people.
The Barker review makes an important distinction between beneficial and detrimental cognitive offloading. Beneficial offloading removes unnecessary or peripheral demands so a learner can focus on what matters. Detrimental offloading transfers to the tool the very cognitive work through which learning was supposed to occur.
That distinction moves the debate beyond simplistic positions of either embracing or banning AI.
An AI tool that removes an administrative hurdle, provides an alternative explanation or helps a student practise may create more room for meaningful learning. An AI tool that does the reasoning, evaluating or composing that the student needs to learn may do the opposite.
The educational challenge is to become much more deliberate about what we allow students to offload.
Some friction is good
AI’s great attraction is its ability to make difficult things easier. But education has never been solely about making things easy.
Some difficulty is the mechanism through which learning happens.
The Barker review calls this “productive friction”: challenge that prompts sustained engagement, restructuring of thought and deeper learning. Its implications for educators include deciding which cognitive and metacognitive friction should be removed, which should be preserved, and where AI might introduce new forms of productive challenge.
The Castlereagh Statement reaches a similar conclusion. Its near-term recommendations call for teaching and learning strategies that retain “desirable difficulties”, while using AI to “support, extend, but not supplant” learning. It also proposes shifting assessment away from product alone and towards greater recognition of process and capability.
The goal should not be to make learning frictionless. It should be to remove the wrong friction while preserving the struggle through which learning happens.
That could mean an AI tutor that asks another question rather than supplying the answer. It might mean requiring students to critique an AI response before improving it. Or it could mean deliberately completing parts of a learning sequence without AI.
The technology itself does not determine the educational value. Pedagogical design does.
From critical thinking to critical judgement
There is also significant convergence around what students will need.
The Castlereagh Statement argues that education systems designed for an era of scarce information and cognitive labour need to place greater emphasis on discernment, relationships, dispositions and learning how to learn.
The Barker review concludes that AI literacy alone is insufficient. Students still require foundational knowledge alongside the ability to question, critique, synthesise, exercise ethical judgement and solve complex problems.
NCEC places these ideas explicitly within a Catholic conception of education, arguing that AI can never replace the capacities education seeks to cultivate: conscience, dignity, responsibility, relationship, ethical decision-making and the search for truth in community.
Critical thinking can sometimes be discussed as though it were a generic skill that exists independently of knowledge. AI makes that assumption even more problematic.
To recognise that an AI-generated explanation is wrong, incomplete, biased or inappropriate, a learner needs something against which to judge it.
In an AI-rich environment, knowledge may become more important as the foundation for judgement, not less.
A new dimension of educational equity
Access to AI is one form of inequality, but it may not be the only - or even the most consequential - one.
Students who already possess strong knowledge, self-regulation and metacognitive skills may be better equipped to use AI as an amplifier. They can interrogate outputs, recognise weaknesses, ask better questions and decide when the technology is useful.
Novice learners may be more likely to accept answers uncritically or outsource the very processes through which expertise would otherwise develop.
The Barker review frames equity as more than access to technology. Students need opportunities to develop the knowledge, judgement and agency required to participate critically in AI-rich environments.
NCEC extends the equity question to families, noting the importance of supporting those with lower digital confidence, limited technology access or less understanding of AI.
Equitable AI policy cannot simply ensure every student has access to the same tool. It must ensure every student develops the capability to use - and sometimes refuse to use - that tool intelligently.
Adding the Catholic perspective
The Barker review primarily asks: Under what conditions does AI support learning?
Castlereagh asks a broader systemic question: What should Australian education value and how should our structures change in response to AI? Its three goals centre on defining valued human capabilities, building coherent lifelong learning pathways, and ensuring Australians can engage confidently, critically and creatively with AI.
NCEC asks how Catholic education should respond collectively, including through national guidance, collaboration, teacher support, parent engagement, assessment and advocacy.
What is education ultimately for?
In Magnifica Humanitas, Pope Leo warns against the temptation to render human thought “seemingly superfluous precisely when it is most needed”. NCEC draws on this to argue for restraint alongside capability: technology should remain subordinate to human dignity and the common good.
That takes the discussion beyond whether AI produces better educational outcomes.
A system could conceivably become more efficient while becoming less relational. Students might produce better work while exercising less agency. Teachers might generate resources faster while losing opportunities for professional judgement. Schools might personalise learning more precisely while inadvertently narrowing what they understand a person to be.
The question is not simply whether AI works, but what its use is forming us towards.
The teacher becomes more important, not less
The Barker review argues that effective AI use actually places greater demands on teacher expertise. Teachers remain responsible for relationships, motivation, recognising misunderstanding, sense-making, judgement, ethical reasoning and deciding when students are surrendering too much responsibility to the technology.
Castlereagh similarly argues for substantial investment in educator capability and says AI should support human flourishing across the education community.
Reducing unnecessary workload is valuable. But perhaps the most interesting opportunity is not using AI to make teachers do the same work faster. It is allowing teachers to spend more of their time on the things technology cannot adequately replicate: relationships, professional judgement, feedback, encouragement, noticing, questioning and formation.
Questions worth exploring together
None of these papers provides a finished blueprint. Together, they suggest a rich set of questions for Catholic education:
What knowledge and capabilities do students need to retain independently of AI?
Which cognitive tasks can safely be offloaded, and which are essential to learning?
Where should schools deliberately preserve productive friction?
How should assessment change when the quality of a finished product tells us less about the capability of the student who produced it?
How do we prevent AI from creating a new divide between students who can critically direct it and those who become dependent on it?
What should responsible AI use look like at different stages of a child’s development?
How do we involve parents meaningfully in these decisions?
What distinctly human work should teachers have more time to undertake?
What does human flourishing look like in an education system where increasingly capable machines are always available?
Collaboration will be key, exploring these questions and others in an ongoing national dialogue.
The questions AI presents are too large - and changing too quickly - for every school or system to solve independently.
Explore the thinking behind the discussion
These papers approach AI and education from different perspectives — research, policy, system leadership and Catholic thought. Together, they provide a useful foundation for deeper discussion.
NCEC – AI in Catholic Education: Explore a distinctly Catholic approach to AI, including human dignity, cognitive offloading, equity, assessment and opportunities for national collaboration.
The Castlereagh Statement: Consider what Australians should know and be able to do in an AI-enabled world, and how education might preserve human capability while embracing technological change.
Barker Institute / University of Sydney – Rapid Literature Review: Dive into the emerging research on generative AI and learning, including the performance paradox, productive friction and the conditions under which AI helps or hinders learning.
Pope Leo XIV – Magnifica Humanitas: Explore the deeper question of what technological change means for human dignity, thought, responsibility and the formation of the person.
