What Must Education Preserve?

Sir Mufti Hamid Patel CBE

He teaches them the Book and wisdom, and purifies them.

Al-Qur’an, 2:151

Introduction

Artificial Intelligence has brought an age-old question into sharper focus: “What is education for?”

Much of the debate begins, understandably, with what AI can do. The list is already impressive. Powerful, large language models (LLMs) such as ChatGPT, Claude, Copilot and Gemini can complete tasks that once demanded significant human effort. They are widely accessible to teachers and pupils for free. A student can produce polished work in seconds. A teacher can use the same technology to assess it and generate feedback. We are approaching the absurd point where machines can produce work that other machines then evaluate.

But polished output is not the same as learning. Schools know this. Leaders are already adapting curriculum, assessment, academic integrity guidance, and expectations around how pupils work. Yet those decisions cannot begin with the capabilities of the technology. They must begin with the purpose of education.

Those of us who work with young people see why. Some of the most important outcomes of education are not visible in a final piece of work. Pupils grow by mastering difficult knowledge, persevering when success is not immediate, taking responsibility for others, learning from trusted adults, performing, and serving. These experiences do more than improve performance. They help form the person.

That idea has deep resonance within the Islamic tradition. Education is not simply about what a person can know or produce. Rather, it is about what kind of person they are becoming and whether they have understood their purpose. In an age of increasingly capable machines, the central question cannot simply be about what can be automated. It has to be about evaluating what education must deliberately preserve to fulfil its role in forming the human being.

Knowledge is the intellectual foundation of that work. Without it, judgment, wisdom and responsible action are all weakened. I propose that five important dimensions of human formation sit upon the foundation of knowledge. As machines grow more capable, these dimensions become more important. They are: effort, agency, embodiment, relationships and reflection. Before exploring these five dimensions, however, we need to begin with knowledge itself.

Knowledge in an age of information

“Where is the wisdom we have lost in knowledge?
Where is the knowledge we have lost in information?”
TS Eliot, ‘The Choruses’

In an age of information abundance, Eliot’s question feels remarkably contemporary. So too does the distinction between information and knowledge.

Critical thinking, creativity and problem solving do not happen in a vacuum.¹ Pupils need something substantive with which to think. Rich vocabulary, strong conceptual understanding, secure factual and deep disciplinary knowledge are all vital. For this reason, curriculum needs to be carefully sequenced, coherent and cumulative so that knowledge acquisition builds securely over time. The evidence here is well established. What pupils already know shapes what they are subsequently able to understand, question and create.²

AI makes the distinction between access and understanding harder to ignore. It can retrieve, organise and generate information at extraordinary speed. It can also produce novel combinations of existing ideas. But generating information does not create knowledge in the mind of a learner. Knowledge requires interpretation, the ability to understand significance, make connections, and exercise judgement.

We see the difference in classrooms. A pupil may find the right fact but miss its significance, or be unable to apply an explanation to a new problem. They may receive a convincing AI-generated answer but lack enough subject knowledge to recognise what is inaccurate, incomplete or misleading. The LLM tools themselves warn users to check important information, but evaluation goes further than spotting factual mistakes. Pupils need enough subject knowledge to recognise when an answer is implausible, to compare it with other sources, and to ask what assumptions or perspectives may be missing. AI systems learn from human-produced data and are shaped by human choices, so their outputs can reflect the limitations and biases of both. Remaining in command of these tools requires pupils to interrogate, rather than simply accept, what they are given.

Checking, however, is not a free-standing skill. It depends on knowing enough to spot the problem, ask the next question and know where to look. To interrogate information intelligently, pupils first need knowledge of their own. This is also why AI literacy cannot be reduced to learning how to prompt effectively. The OECD and European Union’s AI Literacy Framework places technical knowledge alongside the skills and attitudes needed to use AI critically, responsibly and ethically.³

The role of the teacher, therefore, remains central. Teachers do more than make information available. They select and sequence knowledge, explain difficult ideas, connect new learning to what pupils already know, identify misconceptions, and model how to question, verify and revise information. In an AI-rich classroom, this role becomes even more significant. A knowledgeable teacher can move pupils beyond accepting an answer to understanding why it is right, where it may be wrong, and what further evidence is needed.

The Islamic concepts of ilm—knowledge and hikmah (wisdom)—deepen the point. Ilm places knowledge at the heart of education; hikmah directs us towards its understanding and use. Information is not the end point. Knowledge should lead to judgement; judgement should move towards wisdom and responsible action. The Prophet (PBUH) would pray, “O Allah, I ask You for beneficial knowledge, good provision, and accepted deeds.” The Islamic tradition, therefore, asks not only whether knowledge has been acquired, but whether it benefits, directs and bears fruit in action.

The practical implication for schools is clear. Access to information is not a substitute for learning. We have to specify what pupils need to know for themselves. Assessment should then distinguish between what pupils genuinely know, understand and can apply independently, and what they can achieve when using AI intelligently, critically and responsibly. Both are important in the modern world, but we should not confuse one with the other. The growth of AI, therefore, strengthens, rather than weakens, the case for a knowledge-rich education.

If knowledge is the foundation of education, the question then becomes how that knowledge is acquired, secured and ultimately transformed into judgment and wisdom. It is here that the five dimensions of human formation outlined above become important. The first of these is effort.

Effort: ensuring productive struggle is preserved

Most of us can recall the moment when something we had struggled with for days or weeks finally clicked. That understanding rarely arrived immediately. It emerged through grappling with the problem, making mistakes, revisiting ideas and persisting through difficulty. The process was not separate from the learning. It was the learning.

Not every difficulty is educationally valuable, of course. Unnecessary obstacles should be removed wherever possible. However, productive struggle should be preserved.

Learning often takes place within a “Goldilocks” sweet spot. If a task is too easy, pupils have little reason to stretch their thinking or develop new capabilities. If it is too difficult, cognitive overload, frustration or disengagement can lead to an abandonment of trying. The most productive space lies between the two. This resonates with Vygotsky’s Zone of Proximal Development—the space between what a learner can currently achieve independently and what they can achieve with appropriate guidance. Good teaching continually makes judgements within this space—when to explain, when to give a hint, when to ask another question, and when to allow a pupil to stay with the difficulty for a little longer.

Effort is a necessary component of learning for at least two reasons. First, it develops perseverance and the capacity to continue when success is not immediate. Second, the mental work itself contributes to learning. Retrieving knowledge, attempting an answer, making an error, correcting a misconception, drafting and redrafting are more than inconveniences on the path to learning. They can be part of the process through which learning becomes secure. Research on retrieval practice, for example, shows that previously learned material, which is then actively and regularly recalled, can slow forgetting and strengthen later learning.⁴

AI creates a new challenge here because one of its great strengths is its ability to remove friction. Used well, this can be valuable. A pupil stuck on a mathematics problem might benefit from a carefully judged hint; a writer may get feedback that identifies a weakness without rewriting the paragraph; a reader could gain clarification of unfamiliar vocabulary before returning to a demanding text. Such support can keep the learner within a productive zone of challenge. The aim is to provide enough support for progress without doing the formative work on the learner’s behalf.

The problem is that AI can also make successful performance look like successful learning. The OECD’s Digital Education Outlook 2026 notes that generative AI (GenAI) can improve immediate task performance without necessarily producing equivalent learning gains when cognitive work is outsourced.⁵ A pupil may therefore produce an impressive answer while bypassing the retrieval, reasoning, error correction or drafting through which learning would otherwise take place. PISA 2025 provides further evidence of this distinction. Across OECD countries, pupils who used AI for specific schoolwork tasks such as summarising, researching or drafting tended to perform less well in science than those who did not. Yet a different pattern emerged for pupils who used AI to support their learning. Moderate use was associated with similar, and sometimes slightly stronger, performance. The OECD cautions that these findings do not establish causation, but they reinforce an important principle. AI is most useful educationally when it supports the learner’s thinking rather than replaces it.⁶

This is one of the great challenges education now faces. LLMs are easily accessible. We have to ensure pupils understand that learning when and how to exert effort is itself a skill worth cultivating, and will be a significant support for effective learning and for life.

Islamic notions of sabr and ihsan may provide frameworks to support such understanding. Sabr is not simply passive endurance, but steadfastness through difficulty; ihsan points towards excellence. Together they remind that worthwhile development often involves sustained effort, discipline and gradual improvement.

The practical question for schools is therefore: “What should pupils attempt before turning to AI?” In some areas, an “attempt first, AI later” approach may be appropriate, particularly where retrieval, reasoning, drafting or problem solving are themselves part of the learning objective. Technology should remove barriers without removing the effort through which learning is secured. But increasingly, pupils themselves will have to make that judgement. At that point, effort becomes a question of agency. Can the learner remain the person who decides, judges and takes responsibility for how the tool is used?

Agency: preserving the learner as thinker and actor

George Eliot wrote that “the strongest principle of growth lies in human choice.”⁷ In education, agency begins when young people understand that their decisions have implications. Agency is more than simply giving pupils choice though. It means getting them to understand that they are the author of their own thinking and actions, rather than merely the recipient of increasingly intelligent assistance. Young people develop agency when they are entrusted with genuine responsibility, and when they experience the consequences of their decisions. We see this every day in our schools. Pupils who may appear disengaged are often transformed when they are given ownership of something that matters. At Star we see this, whether helping in the school’s peace garden, supporting an assembly, or mentoring younger pupils, pupils frequently carry out these responsibilities with pride, and they grow through this.

In an age of AI, however, tools make it easier for students to delegate tasks as well as elements of judgment itself. But when too much of the decision-making is transferred to the tool, the pupil may cease to be the genuine author and agent of the work. The UNESCO AI Competency Framework for Students offers guidance around this. The framework focuses on pupils as responsible users and co-creators, selectively using AI as a tool. It emphasises human-centred thinking, critical judgment, ethics and citizenship responsibilities.⁸ AI literacy therefore has to include knowing when to use a tool, when to question it and, sometimes, when not to use it at all. Pupils need to learn that they will remain accountable for the judgment exercised, the work submitted, and the decisions made about how the technology contributed to it.

Islamic thinking adds important dimensions to this understanding of agency. Niyyah asks pupils to consider their intention, pushing them to ask questions such as, “Why am I doing this?” and “What am I trying to achieve?” This can help pupils understand their deeper motivations. Amanah—trust, places responsibility alongside capability, and asks pupils to think about what has been entrusted to them, and what duties follow from that trust? Khilafah—stewardship, widens agency further, connecting human action with responsibility towards the common good.

Seen through this lens, agency is not simply freedom to do what one wishes. It is the ability to exercise judgment and accept responsibility for one’s actions.

Schools in turn need to protect opportunities in which pupils can genuinely act. Leadership, service, mentoring, volunteering and civic participation should not be regarded as peripheral enrichment. They all provide formative experiences through which young people practise responsibility, contribution and judgment.

These self-same principles should shape expectations around AI. Pupils should be able to explain what role AI played in their work, acknowledge its use where appropriate, defend the choices they made, and ultimately take ownership of the final product.

Agency is not developed only through intellectual decisions. It is exercised through real actions, with real people, in real settings. Leadership, service and responsibility require presence and encounter. Agency therefore leads us naturally to embodiment.

Embodiment: preserving learning through lived experience

Learning is a lived and felt experience. It involves the mind, but also the body, emotion, uncertainty, risk, memory and encounter. Education needs both information and participation. Confidence, leadership, service, teamwork, resilience and public speaking are ultimately learned through experience, practice and participation. Understanding the principles of public speaking, for example, is one thing. Standing before an audience, heart pounding and learning to manage nerves, is a different kind of education. Likewise, leadership can be discussed in theory, but taking responsibility for an event, or serving the community, places very different physical and emotional demands upon the individual.

Embodied experiences such as wider youth activities⁹, and nature-specific outdoor learning¹⁰, demonstrate benefits across socio-emotional development, academic outcomes, and wellbeing. These findings reinforce what schools see in practice. Sport, performance, outdoor education, service and more, create opportunities for young people to encounter challenge in ways that are felt and lived, rather than merely understood conceptually.

As educational experiences become increasingly mediated through screens and intelligent systems, embodied learning experiences become ever more important. AI may be able to explain how to give a speech, or generate a plan for a community project. But it cannot stand nervously at the lectern, recover after forgetting a line, negotiate with a difficult teammate or experience the satisfaction of having served someone well. Some learning requires presence. As Aristotle observed simply, “We learn an art or craft by doing.”¹¹

Fitrah (innate disposition) and tarbiyyah (nurturing gradual development) help expand this understanding from an Islamic perspective. If education is concerned with the cultivation of the whole human being, then formation cannot concern intellect alone. Islam is an embodied way of life. Prayer, fasting and pilgrimage are examples of that. Human beings learn, act and develop as embodied persons. Physical, emotional and practical experiences in turn have an essential place within education as part of the wider work of formation.

Schools should, therefore, be cautious about allowing efficiency or technological convenience to narrow the range of experiences pupils receive. Sport, performance, service, outdoor education, public speaking and enrichment may sometimes look less efficient than screen-based alternatives; indeed, they may take more resources. That does not make them educationally inefficient. Their value lies partly in what cannot be downloaded, generated or simulated.

Relationships: preserving the human encounter

Those of us who work in schools know the importance of relationships.

Across our schools at Star, every child is connected deliberately to five trusted adults. For some young people, these are among the most stable and consistent relationships in their lives. These five adults often notice changes in behaviour long before anyone else does and intervene before a problem becomes a crisis. No algorithm can reproduce that human connection. It is through such relationships that young people are supported, challenged and ultimately flourish.

Relationships and rigour should not be set against one another. Being known, valued and supported does not require lowering expectations. Often it is precisely because a pupil trusts the adult in front of them that they are willing to accept correction, attempt something difficult, or persevere after failure. Care and challenge can reinforce one another.

Evidence supports the importance of these relational experiences. A recent UK government systematic review of youth provision found particularly consistent positive evidence around mentoring and peer mentoring, with studies reporting benefits including self-efficacy, connectedness, resilience and wellbeing.¹² Such evidence is, of course, important, but for those who work in schools, the human reality behind it is immediately recognisable.

This becomes especially significant as AI becomes more capable. AI can personalise explanations, generate feedback and sustain increasingly convincing conversations. These can be useful capabilities. But personalisation should not be mistaken for relationship. The OECD describes relationships, collaboration and trust as part of the “human core” of education and asks how that core can be strengthened as technological change accelerates.¹³

There is also an intergenerational dimension to education. Knowledge, culture, values and wisdom are often passed from one generation to another through human relationships. Teachers communicate more than subject content. Through their conduct, judgment and example, they also communicate something about what is worth knowing, how knowledge should be approached and how other people should be treated. Relationship is therefore not merely a vehicle through which education is delivered. It is part of the education itself.

Islamic concepts such as rahmah (compassion), akhlaq (etiquette), and suhbah (companionship) strengthen this understanding. Rahmah brings mercy and care into our dealings with others. Akhlaq reminds us that character is encountered and modelled as well as taught. Suhbah points to the formative influence people have upon one another. Together these concepts suggest an understanding of education in which human formation takes place through sustained relationships, example, challenge and care.

This has an important practical implication for the use of technology in schools. Educational technology should be evaluated partly by its effect on relational time. If automation reduces administrative burden and creates more time for teachers to teach, mentor, listen and know their pupils, it may strengthen the human work of education.¹⁴

So the question facing schools is more than whether AI can perform a particular task more efficiently. Rather, it is about examining whether AI use creates greater capacity for meaningful human encounter, and deepens the relational experience; as it is often within encounters with teachers, mentors and peers that young people are encouraged to reconsider what they think, examine their choices, and make sense of their experiences. That leads naturally to the next formative dimension: reflection.

Reflection: creating space for meaning and judgment

Not all of the important work of education is easy to measure. Every day, teachers help young people recognise the consequences of their actions, take responsibility when something has gone wrong rather than blaming others, and think more carefully about the choices they are making. These moments do not appear in performance tables, yet they rightly occupy a significant proportion of a school’s time, because they are part of how character and judgment are formed.

Reflection is one of the processes through which knowledge becomes judgement and, ultimately, wisdom. Pausing, interpreting experiences, examining assumptions, reconsidering evidence, all form part of reflection.

We see this in many forms in schools. A pupil revisits an argument after hearing a different point of view. They may assess what went well (and what did not) after leading an event. They may return to a piece of work after feedback and see something they had previously missed. Sustained reading, extended writing, discussion, mentoring and self-evaluation all create opportunities for young people to make sense of what they know and what they have experienced.

There is a strong evidence base for developing such habits deliberately. The Education Endowment Foundation identifies metacognition and self-regulation as important to effective learning, with pupils benefiting from being explicitly taught how to plan, monitor and evaluate their learning. Importantly, these approaches appear to work best when embedded within subject teaching rather than taught as detached ‘thinking skills’.¹⁵ Metacognition is one educational expression of reflection, but reflection is larger than a learning strategy. It includes making meaning from experience, examining oneself and revising judgement.

Reflection in an AI-rich environment requires more than checking whether an answer is correct. It means pausing to compare the response with one’s own thinking, understanding that the response may be built on inherent bias, asking whether it changes one’s judgment, and considering why. Developing habits of questioning, comparing and revising helps keep the learner intellectually active rather than becoming a passive recipient of generated answers.

AI introduces a particular challenge because of the speed with which it can respond. A question, uncertainty or half-formed thought can now move almost instantly to a polished answer. That may be useful on occasion. But if every gap is filled immediately, pupils may have fewer opportunities to predict, wonder, retrieve what they already know, test an idea, or remain with uncertainty long enough to think more deeply. The OECD’s Digital Education Outlook 2026 warns of risks from cognitive offloading to GenAI, including what it terms “metacognitive laziness” and disengagement. It argues that education should continue to develop independent thinking and foundational skills without GenAI, as well as learning how to work effectively with educational and general-purpose AI.

There is educational value in the pause before the answer.

This need not mean excluding AI from reflective learning. Used carefully, it can itself become something upon which pupils reflect. A pupil might first articulate what they know or predict, then consult AI, before comparing the response with their original thinking by reflecting on such questions as “What was different?” “What was convincing?” “What needs verifying?” “Has the evidence changed my view, and if so, why?” Such routines keep the learner intellectually active and make the development of judgment, rather than production of an answer, the main objective.

The Islamic traditions of tafakkur (deep reflection) and muhasabah (self-accountability) offer a particularly rich language for this dimension of education. The former directs us towards thoughtful contemplation, whilst the latter involves examining oneself, one’s actions and intentions. Both resist the idea that speed is always progress. Formation sometimes requires activity to give way to consideration. The question becomes not simply, “What did I produce?” but, “What have I understood?” “What have I learned?” “What should I reconsider?” and, “What should I now do differently?”

Schools, therefore, need to protect space for sustained attention and independent thought. Not every question should move immediately from prompt to generated answer. Extended reading and writing, thoughtful discussion, teacher questioning, self-evaluation and, at appropriate points, deliberately AI-free or device-free periods can all preserve the space in which pupils learn to reflect. Assessment too can value explanation of reasoning, reconsideration of evidence and revision of judgment, rather than evaluating the final product alone. Project files, for example, may take equal weight to an essay.

This returns us to the question with which we began. As technology becomes increasingly capable of producing answers and completing tasks, what experiences should education deliberately preserve because they contribute to the formation of the human being?

Conclusion: becoming more deliberately human

The challenge for schools is not to choose between embracing AI and protecting learning. It is to develop a clear educational rationale for deciding where technology helps and where something important may be lost. A useful human formation test is to consider what a technology enables or strengthens, what formative processes it may bypass, and what happens to the learner when those processes are removed.

Applied across school life, this has practical consequences. Curriculum should remain knowledge-rich and carefully sequenced, with AI literacy built upon secure subject knowledge. Assessment should distinguish between what pupils know, understand, and can do independently, and what they can produce with intelligent assistance. Pedagogy should use AI where it genuinely scaffolds learning, while preserving productive struggle, independent thinking, and teacher judgment. School culture should continue to create deliberate opportunities for responsibility, service, embodied experience, human relationship, and reflection. Technology should serve these educational purposes, not quietly redefine them.

Underlying these choices is the distinction between performance and formation. A stronger answer does not necessarily mean a more knowledgeable learner. A task completed more quickly does not necessarily mean that the effort through which learning becomes secure has taken place. Greater AI assistance does not necessarily develop greater agency. And something can be highly efficient while being educationally impoverished.

Knowledge is the intellectual foundation of education. Effort, agency, embodiment, relationships and reflection are dimensions of formation. They are not substitutes for knowledge, nor are they free-floating skills detached from it. Rather, they describe some of the ways in which knowledge is acquired, exercised, tested and turned towards judgement, responsibility, and wisdom.

Some of the experiences that appear inefficient from the perspective of technology may, therefore, be precisely those that are formative from the perspective of education. Wrestling with a difficult idea, standing nervously before an audience, taking responsibility when something goes wrong, being challenged by a trusted teacher, serving another person, or sitting with a question long enough to change one’s mind are all formative. These experiences take time because formation takes time.

This is where Islamic educational traditions have something important to contribute to the wider debate. They can offer a language in which education is concerned not simply with what a person can produce, but with what kind of person they are becoming. Knowledge should lead towards wisdom, capability towards responsibility, learning towards the formation of character, and the wise exercise of human agency.

AI may profoundly change what is possible in education. Schools and systems should therefore judge educational technologies not simply by the efficiencies they create or the quality of the outputs they generate, but by what they do to the learner and the kind of human being they help to form. The greater the capabilities of our machines, the more deliberate education must become about forming the human being.

1. For critical thinking see for example: Willingham, D. T. (2019), How to Teach Critical Thinking, NSW Department of Education, Education for a Changing World Occasional Paper Series, pp. 3, 6.
   For creativity see for example: Sun, M., Wang, M. and Wegerif, R. (2020), “Effects of divergent thinking training on students’ scientific creativity: The impact of individual creative potential and domain knowledge”, Thinking Skills and Creativity, 37, 100682, DOI: 10.1016/j.tsc.2020.100682.
   For problem solving see for example: National Academies of Sciences, Engineering, and Medicine (2018), How People Learn II: Learners, Contexts, and Cultures, Washington, DC: National Academies Press, Chapter 5, “Knowledge and Reasoning”, pp. 85–108, DOI: 10.17226/24783. ↩
2. OECD (2020), What Students Learn Matters: Towards a 21st Century Curriculum, OECD Publishing, Paris, https://doi.org/10.1787/d86d4d9a-en. p.57 cf p.17 ↩
3. OECD / European Union (2026). Empowering learners for the age of AI: An AI literacy framework for primary and secondary education. OECD Publishing, Paris. https://doi.org/10.1787/65cd27d4-en p.5 ↩
4. McDermott, K.B., (2021), Practicing Retrieval Facilitates Learning. Annual Review of Psychology 72:609-633. https://doi.org/10.1146/annurev-psych-010419-051019 ↩
5. OECD (2026), OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education, OECD Publishing, Paris, https://doi.org/10.1787/062a7394-en. ↩
6. OECD (2026), PISA 2025 Results (Volume I): Future-Ready Students, PISA, OECD Publishing, Paris, DOI: 10.1787/73451bc5-en. https://www.oecd.org/content/dam/oecd/en/publications/reports/2026/09/pisa-2025-results-volume-i_5265bfb1/73451bc5-en.pdf ↩
7. Eliot, G., (1876), Daniel Deronda, London, William Blackwood and Sons, Book Six: Revelations, Chapter 42. ↩
8. UNESCO (2024), AI Competency Framework for Students, UNESCO, Paris. ISBN 978-92-3-100709-5 https://doi.org/10.54675/JKJB9835 ↩
9. Department for Culture, Media and Sport (2024), Youth provision and life outcomes: systematic literature review, London. UK Government DCMS, 29 February 2024. https://www.gov.uk/government/publications/youth-provision-and-life-outcomes-research/youth-provision-and-life-outcomes-systematic-literature-review-executive-summary ↩
10. Mann J. et al, (2022), Getting Out of the Classroom and Into Nature: A Systematic Review of Nature-Specific Outdoor Learning on School Children’s Learning and Development. Front Public Health. 2022 May 16. https://pubmed.ncbi.nlm.nih.gov/35651851/ ↩
11. Aristotle, Nicomachean Ethics. ↩
12. Department for Culture, Media and Sport (2024), Youth provision and life outcomes: systematic literature review, London. UK Government DCMS, 29 February 2024. https://www.gov.uk/government/publications/youth-provision-and-life-outcomes-research/youth-provision-and-life-outcomes-systematic-literature-review-executive-summary ↩
13. OECD (2026), Reimagining Teaching in an Accelerating World, International Summit on the Teaching Profession, OECD Publishing, Paris, https://doi.org/10.1787/d0edfe8c-en. ↩
14. ibid. ↩
15. EEF, (2025), Metacognition and Self-Regulated Learning Guidance Report, London, Education Endowment Foundation, https://educationendowmentfoundation.org.uk/education-evidence/guidance-reports/metacognition ↩

---