[Paul Cézanne, Mont Sainte-Victoire Seen from the Bibémus Quarry]

AI can now write plausible essays, solve mathematics problems and explain almost any tricky subject-specific question within seconds. This presents an obvious challenge to schools and universities. If students can use AI to complete the work set for them then how do we know that they have learned anything? Behind this practical problem lies a more fundamental question. What were all those essays, problems and examinations supposed to achieve in the first place? If education amounted simply to producing correct answers, AI is just doing us all a favour by producing them more efficiently. Of course, we all know that education is more than this. But what exactly is that something more? What is it that makes education truly educational?

Cézanne’s doubt

The French philosopher Maurice Merleau-Ponty suggested that Paul Cézanne doubted whether conventional ways of painting could capture how we actually see the world, while continually questioning the truth of his own paintings. Conventional painting presented the world as already formed: objects had stable outlines, space followed geometrical perspective, and the viewer occupied a fixed position. Cézanne tried to reach beneath these conventions to perception as it is lived. In his paintings, contours emerge from colour, perspectives shift, and the familiar solidity of things appears in the process of taking shape.

This seems to me to be profoundly educational. Cézanne does not merely show us another collection of objects. He changes how we see. That movement from accepting a finished appearance to understanding the process through which it has been formed is, I think, central to education.

Reengineering reality

Cambridge Engineering has a global reputation not only for teaching students how existing technologies work but for enabling them to design things that work differently. Professor Nathan Crilly, who teaches engineering design and problem-solving, demonstrated this approach in a workshop for the Cambridge Digital Education Futures Initiative.

He showed how a problem can be opened up by comparing it with an apparently unrelated object, situation or story. The analogy loosens our initial definition of the problem. It enables us to see connections and possibilities that were previously hidden.

Like Cézanne’s painting, this approach disturbs the apparent solidity of the world. But engineering also adds something important. It shows that what has been formed in one way can be formed differently.

Students must, of course, acquire established knowledge and skills. An engineer cannot redesign a bridge without understanding the principles that keep it standing. But the educational purpose of that knowledge is not simply to reproduce bridges that have already been built. It enables students to understand why things have taken their present form, assess the good and the bad aspects of every design and then participate with others and with tools like AI in designing something that offers a different and possibly better answer to the challenges that are given to us by the world.

Thinking together

I arrived at the idea that education is essentially about expanding dialogic space through research with groups of children solving reasoning-test puzzles. The most successful groups were not those in which one child quickly supplied the correct answer. They were groups that could think together better than any individual member. We measured this by comparing the group’s performance on reasoning tests with the members’ performance when working alone.

Video recordings helped us see how this collective intelligence emerged. Before breakthroughs, there were often periods of shared silence in which everyone focused together on the problem. Children openly voiced uncertainty: “I don’t get this at all.” They reconsidered their initial judgments: “No, it can’t be that one.” They also displayed humility and openness by asking others for help: “I don’t see it—can you explain?” These were not signs of failure or confusion. They were ways of opening a shared space in which a better understanding could emerge.

This is what I mean by dialogic space: the space of possibility that opens between and around people when they listen to one another, hold back from immediate conclusions, and allow different perspectives to interact.

It opens when the first answer is not accepted but questioned. It widens as other voices and possibilities enter. It deepens when participants question their initial assumptions and listen for a different way of understanding the problem to emerge.

F. Scott Fitzgerald once wrote that the test of a first-rate intelligence was the ability to hold two opposed ideas in mind at once. Something similar was happening in these groups, but collectively, and their achievement went further than just tolerating contradiction. It was to be able to sustain the space between opposed ideas until a better understanding could emerge.

Education and training

The experience of dialogic space and its role in solving problems helps to clarify the traditional distinction between training and education. Training enables us to perform an established operation successfully. We might be trained to use a machine, follow a procedure, solve a familiar kind of equation or reproduce an accepted account of a historical event.

The difference between training and education lies in how learners relate to knowledge. Training can remain the acquisition of a predetermined performance for an external purpose. But it can also become part of education when it begins with a meaningful problem or challenge and draws learners into understanding, questioning and taking ownership of the practice. Education develops not only the ability to perform, but the judgment to know when and why a practice is appropriate, the capacity to adapt it to new contexts, and the enterprise to develop it further.

We can see the difference in almost any field. Someone might be trained to apply a scientific formula without understanding why it works. A medical student might memorise the standard treatment for a condition without yet being able to judge a complicated case. A history student might reproduce the lecturer’s account of the Reformation without comparing it with other interpretations or examining the evidence. In each case, successful performance may conceal the absence of education.

The educated self

Acquiring knowledge and skills, wonderful as this is, only becomes educational when this changes us: changing how we see, opening new possibilities and increasing our capacity to make responsible judgments and take appropriate action.

Opening, widening and deepening dialogic space means developing both answerability and authorship. To be answerable is to recognise that we cannot simply say whatever suits us. We must answer to other people, to evidence, to inherited knowledge and to the reality we are trying to understand. Cézanne was answerable to the mountain that he painted many times, he refused to allow the conventions of painting to see it on his behalf. He had to offer his own answer to the question that it posed. The necessary complement of answerability is what I call authorship. Educated people become authors of their own lives rather than merely living out roles given to them by others or by circumstance. Authorship is linked to authority, leadership and enterprise: it means being able to judge, respond and speak in a voice recognisably your own. This does not mean creating ourselves or the world out of nothing. It means responding to the call of life, to its challenges, questions and possibilities, in ways that take both ourselves and the world forward.

The educated person is therefore neither merely obedient nor arbitrarily self-assertive. Answerability without authorship becomes deference; authorship without answerability becomes irresponsibility. Education requires both the capacity to listen and respond to what is already there, and the growing capacity to participate in creating something new and possibly better.

What AI reveals and why this is an opportunity

Imagine two students using AI to consider a question, perhaps why the First World War began. The first asks for an essay, makes a few changes and submits it. The second begins with an interpretation of their own. They ask the AI to challenge it, identify contrary evidence and present the strongest competing explanation. They struggle to reconcile the different accounts and eventually arrive at a conclusion they can defend.

Both students have used AI. Both may produce competent essays. But it is mainly the second who has been engaged in education. For the first student, AI closed the dialogic space by supplying a finished answer. For the second, it helped to open and widen that space by introducing questions, objections and alternative possibilities. The student remained responsible for deepening the dialogue and reaching a judgment.

Education is needed to enable and empower students to use AI for their own purposes, rather than being used by AI systems for purposes they neither understand nor control.

But this challenge is not new. What AI reveals is the extent to which education has already been shaped by an earlier technology: print literacy and the institutions built through and around it. Much of our existing educational system confuses successful performance with education, rewarding students for producing expected answers, passing examinations and accumulating marks while paying less attention to whether their relationship to knowledge and so their capacity to judge and act has been transformed. It is often less concerned with education than with training students to compete for and find a place within a vast bureaucratic machine organised around the affordances of print.

By making this confusion between performance and education so clearly visible, AI offers us a great opportunity to rethink what we are doing in education. The answer is not to reject technology and return to a fantasy of the ‘human’ untrammelled by technology. It is to recover what is most educational about education: the expansion of dialogic space.

This expansion is not only a matter of entertaining more voices or possibilities. It changes who we are. Through education, what first confronts us as an external world of knowledge, practices and technologies becomes something we can understand from within, identify with and participate in shaping. This is an expansion of interiority, but not of the isolated self. The educated person becomes both answerable to the larger process and an author within it, both shaped and shaping, listening and responding. Education enables us to become more fully conscious participants in the dialogue through which both we and the world continue to be formed.

I keep reading that using AI damages students’ capacity to think and learn. The claim is often presented as though it were established fact. I find this odd because I also read a great deal of research suggesting that AI can support deeper conceptual understanding.

The studies cited to support the negative claim share the common feature of using AI within the existing curriculum using standard assessments. A good example is a randomized controlled trial by Bastani and colleagues involving nearly 1,000 secondary-school mathematics students (Bastani et al., 2025). Students given access to an unrestricted GPT-4-based assistant performed substantially better while using it. But when the AI was removed, they scored 17 per cent lower than students who had studied using conventional resources. The influential conclusion: generative AI can improve performance while undermining learning.

This was a good study and it also tested a more carefully constrained AI tutor, which performed as well as the no-AI control group. That third condition shows that how we design AI in education makes a difference. But the study illustrates a deeper problem with how we currently research educational technology.

“Learning” was defined as improved performance on an examination covering the existing mathematics curriculum, completed without AI. Given that definition, any new capability in the interaction between student and AI was not relevant and was not looked for.

Imagine a mathematics class where students learn to calculate using an abacus and use it in exams. An experimental group gets calculators during lessons, but must use an abacus in the final exam. If they perform worse, we might conclude that calculator use didn’t prepare them to operate an abacus. We could not conclude that calculators had damaged their capacity to learn mathematics. The Bastani study establishes that students given unrestricted GPT-4 became less able to solve problems with pen and paper when the new technology was withdrawn. It does not establish that they became less capable of mathematical activity in an environment where AI remained available, nor does it investigate what different mathematical capabilities education might require in such an environment.

This raises the larger question: how do we research the future?

Educational researchers cannot always simply take performance on existing assessments as a measure of ‘learning’. The research also has to ask what should we be learning. Many educationalists argue that it is not fair on students who will probably work and live in an AI-permeated world if schools do not teach them how to use AI.

Critics argue that educational research should follow the medical model: randomized controlled trials to discover whether interventions work. But the standardised tests needed for this model are normally those developed to measure learning within the existing system. Students are not patients in need of a cure for a clearly defined condition. Is ignorance a clearly defined condition? The aims of education are always contestable, and especially so now when we are introducing a radically new technology that many claim should change the nature of education itself.

Design-based research offers a different approach. Instead of asking whether a tool improves existing test scores, it asks: Does design X generate the anticipated process M? Did that process contribute to the intended outcome Y? Crucially: Did the experience of the design give us reason to reconsider the desirability of the outcome itself? This last question is important. Research should not merely refine the means while leaving the ends untouched. Our conception of what counts as learning may change when we encounter learners, teachers, and the realities of practice.

A recent Japanese study provides a concrete example. First-year university students worked in interdisciplinary groups to design a classroom for 2050 (Naganuma et al., 2026). They developed expertise in different areas, shared it through a jigsaw activity, and used generative AI to introduce additional possibilities. The researchers traced the development of ideas through classroom talk, interactions with AI, successive artefacts, and final designs.

The initial conjecture was that widening the range of ideas available would lead to more innovative designs. The intervention did produce “idea expansion”: students generated possibilities that went beyond the original materials. Yet relatively few groups turned this expanded range into coherent, genuinely innovative proposals. In a second iteration, more sophisticated prompting produced even more AI-generated ideas. However, this led to fewer contributions from students themselves. Generating more possibilities did not automatically produce better collective thinking. The next stage, not yet reported upon, will focus on teaching students how to select the most promising new ideas.

This result is valuable precisely because the original proposal did not simply “work”. It revealed a missing educational capacity: generating possibilities is different from judging which possibilities are promising, connecting them coherently, and deciding which deserve development. AI may be particularly good at widening possibilities while making human selection, judgment, and collective deliberation more visible and essential.

The study did more than test whether a tool improved performance. It helped produce a better account of what the educational aim should actually be.

This suggests a broader understanding of rigor. Good research requires careful observation, appropriate comparisons, and where useful, experimental testing. But research into educational futures also requires conceptual rigour: making explicit the desired ends, the processes expected to produce them, and the criteria used to recognize them, then exposing all three to criticism. It means actively searching for neglected evidence and rival explanations, asking what evidence would persuade us that our favoured idea was mistaken.

There is no conflict between this approach and randomized controlled trials. Once a design, its underlying theory, and its intended outcomes are sufficiently clear, an RCT may help establish whether it produces a difference and under what conditions. But moving too quickly to measurement can avoid the most intellectually demanding work: developing a worthwhile intervention and a plausible theory of how it actually operates in practice.

Research supports dialogue; it does not replace it

The fundamental questions of education differ from those of natural science. They are not “does this work?” but “what should we teach and how should we teach it?” Empirical research cannot answer that alone. It must contribute to genuine dialogue about what education should aim at. I do not mean here just a consultation where stakeholders comment on a predetermined design, but a real conversation where different perspectives encounter and challenge one another, with the possibility that everyone, including researchers, may change their minds.

Researchers contribute evidence about what is possible. What happens when AI enters a classroom? What do students actually do with it? What capacities do they develop, and what might they lose? Such evidence is valuable. But it cannot determine what should happen. That requires dialogue bringing together empirical evidence, practical wisdom, cultural values, concern for particular learners and communities, and competing visions of a better future.

To research the future, then, is to formulate possible futures clearly enough to build and examine them rigorously; to investigate what happens using appropriate methods; and to remain willing to revise both our designs and our understanding of educational purposes. We need research that tests ideas, but also research that helps us generate better ideas to test.

References

Naganuma, S., Shibukawa, S., Minematsu, T., Ohno, A., & Wakihama, Y. (2026). Beyond expert knowledge toward idea innovation: Potential and challenges of a generative AI-supported jigsaw method. International Journal of Educational Technology in Higher Education, 23(1), Article 24. https://doi.org/10.1186/s41239-026-00598-6

Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), Article e2422633122. https://doi.org/10.1073/pnas.2422633122

[The Siren Vase. British Museum]

Like many people at the moment, I have been thinking about The Odyssey, prompted by Christopher Nolan’s new film. My wife and I loved it so much that we went to see it a second time, this time on an IMAX screen, to make sure we had not missed anything.

Nolan has described The Odyssey as a foundational text. It is above all the great homecoming story. Countless later stories have been influenced by it. The Greek word for homecoming is nostos, one of the roots of our word nostalgia. Homer’s homecoming story is about the long detours, distractions and trials that lie between us and home, so that when we finally return, we return as someone else, changed by the journey.

I like this idea of returning to the beginnings of things. Everyday words in education, words such as knowledge, thinking and dialogue, are used so often and so easily that we forget where they came from. Reading the foundational texts in which these ideas first took shape can help us hear them freshly.

Those who read this Substack will know that thinking about AI and education often leads me back to Plato’s Phaedrus, perhaps the first critique of an educational technology. Plato has Socrates describe written words as being like painted figures. They stand before us as if alive, but if we ask them a question they remain solemnly silent. They go on saying the same thing forever. Once separated from the living voice that first spoke them, they cannot explain themselves or respond appropriately to the person before them.

As a contrast, Socrates describes another kind of word, one “written in the soul of the learner.” This word can answer when questioned. It knows when to speak and when to remain silent. It is not merely a record of something once meant. Phaedrus responds: “You mean the living word of knowledge which has a soul, and of which the written word is properly no more than an image?”

The difference Socrates is pointing to runs deeper than the technology of writing. It cannot simply be that speech is alive and writing is dead. A teacher can recite dead words from the front of a classroom, while a sentence written thousands of years ago can address me, disturb me and change my life.

Writing comes alive when it is taken up again in a real act of response. Reading is then not the transfer of information from page to memory, something that AI can do rather better than I can, but it is the taking up of what is written into my own concerns and projects, letting it question me and letting it matter. To read in this way is already to participate in the world.

Another foundational text, the Book of Genesis, offers a different image of what it means to know. In Genesis it is written that Adam “knew” Eve and she conceived. Here, knowing clearly means more than possessing information. It means intimate participation.

There is a beautiful folk etymology of the Hebrew word yada, “to know.” The earliest forms of Hebrew used pictorial signs. The two main characters in the word yada were, it is claimed, originally pictures of an eye and a tent door. On this reading, to know becomes an eye passing through a tent door.

Imagine standing outside a tent. You can locate it and describe it. You may produce a perfectly accurate representation. But another kind of knowing begins when you lift the flap, step inside, sit down and share tea and conversation with the people who live there.

Here is the distinction that follows from Socrates’ contrast between the living word and the dead word. Knowledge can be representation, but it can also be participation.

Modern education overwhelmingly favours remaining outside the tent. It is largely about teaching people to produce representations of knowledge and then checking whether those representations are correct.

Socrates’ foundational criticism of writing as an educational technology proved prophetic. Education can produce people who sound knowledgeable because they can reproduce what has been written, without necessarily knowing how to respond to another person, to a new situation or to the world itself.

There are enormous benefits to representation. Writing allows knowledge to travel across time and space. Books, diagrams, mathematics and scientific models have transformed civilisation. Representations allow us to stand back, compare, criticise and reflect.

The problem begins when representation becomes the only kind of knowing we recognise. We forget how to step inside again: to take what we have learned back into the living weave of our projects and relationships, until it becomes not merely an object we look at but a way of seeing and acting, and the ground for the next step forward.

My own experience of formal education led me to see the problem of focusing only on representational knowledge. I began life pretty happy most of the time, simply participating in the world. As a child, I did not stand outside life looking at it. I lived it.

Gradually, through education, I learned to see the world as an object to be represented rather than as a partner in dialogue. Worse, I came to see myself from the outside: as an image, a label, a personality, an object occupying a particular place in space and time. I began to see myself as I imagined other people saw me.

The centre of gravity shifted. Instead of living from the inside and using representations when I needed them, I became trapped in the representation. I suspect that many people experience something similar. They emerge from education knowledgeable and successful, but somehow estranged from their own lives. They know how they appear. They know how they have been classified. But they no longer know what it means simply to be alive and to answer the world from where they are.

If the soul is formed through living relationships, an education that consistently places representation above participation may do more than give us a narrow idea of knowledge. It may damage the soul.

It is interesting that it was important to Nolan to make The Odyssey without generative AI. He insisted on real places, physical sets and human performers, and has argued that this mattered to the kind of film he wanted to make.

AI represents a remarkable new stage in the history of writing: for the first time, the written word appears able to answer back. This is exactly what Socrates claimed that written words could not do. But this only makes Socrates’ challenge more urgent. What is now the essential difference between the living word and the dead word?

The Russian thinker Mikhail Bakhtin illuminates the issue. Dialogue is not simply an exchange of sentences. A living word is answerable. It comes from someone who has to stand behind it. A promise matters because someone must keep it. An apology matters because someone accepts responsibility. A declaration of love matters because a life is being placed at risk in the saying of it.

AI can produce answers, but it is not answerable. It has no life to answer for.

This is the crucial difference. A living word is not simply a better sentence. It is an event. Something happens. A possibility opens that was not there before.

And this answerability reaches further than the immediate exchange. We answer not only to a teacher, an employer or another individual, but to the larger horizon behind them: to other people, to the living world, to the future and to everything that has not yet been decided.

Perhaps this is how a soul comes into being: through learning to answer the call of the world. Education is the growing of that relationship.

Perhaps education now needs its own homecoming. I do not mean that we need to abandon writing, examinations, knowledge or technology. Odysseus could not return home without making the journey. A true homecoming includes everything learned along the way.

The question is: which serves which—or perhaps even, who serves who?

Representations should help us enter more fully and intelligently into life. They should not leave us standing outside our own experience, looking on and asking ChatGPT how we are supposed to feel or what we are supposed to do.

For centuries, education required students to store and reproduce representations. There was no other way for knowledge to be carried forward. But now AI can reproduce established knowledge faster and more fluently than any student. If producing the expected answer remains the main purpose of education, then the machine has already beaten us.

But the purpose of education is not merely to produce correct answers. It is to help people enter the conversation: to encounter other voices, make judgements, take responsibility, discover possibilities and learn how to answer for what they say and what they do.

Real education invites us to step inside: to discover how we are called by the world and how we might answer that call.

Like Odysseus, we come home changed. We do not return to childhood or to a world before writing. We return with everything the journey has taught us, yet able to recognise, as if for the first time, the value of what was there all along.

[Socrate insegna ai giovani la conoscenza di sè. Circa 1660. Pier Francesco Mola - Museo Civico di Belle Arti, Lugano]

Varsity, Cambridge’s oldest and largest student newspaper, is preparing a feature on students’ use of generative AI. Surveys suggest that nearly all students now use chatbots to help with their work. The big question is: what impact will this have? Emma Tenzler, Varsity’s Deputy Editor, sent me a series of questions. I share her questions—and my answers—here.

Question 1: You argue that AI can become a “cognitive poison” or support deeper thinking, depending on how it is used. Where exactly would you draw that line in the case of a student asking AI to summarise or explain a difficult text?

I used “cognitive poison” as an allusion to Plato’s Phaedrus, where writing is presented as a pharmakon. This word can mean both poison and remedy. Whether AI turns out to be cognitively enabling or cognitively damaging is not an intrinsic property of the technology but depends on how we use it.

Where exactly to draw that line is a question on which many people have opinions. At DEFI (deficambridge.org) we are interested in doing research on this question, and in bringing together the research of others. A recent study by Wong and Qiu compared students who used ChatGPT freely from the outset with students who first generated their own ideas and only then used AI to develop, challenge and evaluate them. The freely assisted students looked more impressive at first, but when the AI was taken away, their advantage disappeared. The “think first, ChatGPT later” students, by contrast, went on to outperform everyone when working independently. The study measured creative thinking rather than deep understanding of a text, but the lesson that sequencing matters probably transfers. AI is educationally productive when it follows an initial act of independent thought, and corrosive when it replaces one. A student who wrestles with a difficult text first, forms even a confused interpretation, and then asks AI to explain, challenge or clarify will benefit intellectually, whereas a student who asks for the summary first, to avoid the struggle, will probably find they are being drip-fed cognitive poison.

Question 2: Can a student engage in a meaningful dialogue about a text without first having read and struggled with the original text themselves?

When I began my PhD, I assumed that I was expected to read carefully everything I referenced. This troubled me because I found it impossible. I still remember the advice I was given: most papers do not need to be read line by line. You can usually grasp their main point from the abstract, the introduction and a quick skim. But there are a few texts that really do need to be read carefully: texts that do not simply reproduce familiar assumptions but offer something genuinely new. Reading such texts carefully can provide the kind of challenge to conventional thought that students need for intellectual growth.

The danger of turning to AI too early is that it removes precisely this interruption. Instead of first encountering the text in all its difficulty and resistance, the student is presented with an interpretation that reflects what the model predicts people are likely to think. AI can make difficult texts immediately accessible, but in doing so it can deprive students of the productive struggle through which those texts become transformative.

So my answer is a qualified one. Students can often have a useful conversation about a text they have encountered only second-hand. Indeed, much academic work depends on that. But for the relatively small number of texts that have the potential to change how you think, there is no substitute for wrestling with the original yourself.

Question 3: What would a genuinely dialogic use of AI by a humanities student look like?

Dialogic education is about learning from the tension between different perspectives, what Bakhtin called “mutual illumination”. Some people argue that this cannot really happen with AI because AI is not a conscious person. I do not think that follows. AI does not need to be conscious in order to introduce perspectives, arguments and voices that can challenge and illuminate a student’s thinking. However, the student may have to work deliberately to prevent the exchange from becoming merely agreeable or superficial.

I describe dialogic education in terms of opening, widening and deepening dialogic space. First, the student needs to open a space with a genuine question: something they are motivated to understand but do not yet understand. Without such a question, interaction with AI can easily become an exercise in producing acceptable answers.

AI can then help to widen the space. The student might ask what other interpretations exist, which voices are missing, how the issue has been understood in different historical traditions or from different theoretical perspectives.

Deepening the space requires more than accumulating perspectives. The student must place their own emerging view at risk. They might say: “This is how I currently understand the argument. What assumptions am I making? Challenge my interpretation. What would someone working within a different framework object to? Is there any evidence you can find that might make me revise my position?” Used in this way, AI becomes not a substitute for interpretation but a means of testing, complicating and developing it.

Question 4: When, broadly speaking, does students’ use of AI strengthen their capacity for independent thought, and when does it weaken it?

This last question takes me back to Plato. In the Phaedrus, the inventor of writing presents his invention to the king as a remedy for memory: a pharmakon. The king disagrees. Writing, he says, will lead to forgetfulness, because people will rely on external marks instead of remembering for themselves. Worse, learners will absorb many things without instruction and will seem to know much while understanding little, acquiring the appearance of wisdom rather than the reality.

I think that Plato’s suspicion about writing is worth taking seriously. Much of the current education system rewards the superficial appearance of wisdom more than the reality. Is being able to write a polished Oxbridge essay really the pinnacle of human achievement to which everyone should aspire? Maybe the advent of AI will help us turn education away from its fixation on reading and writing, and back to what is of real value: posing problems together, solving them together, and cultivating individual judgement and collective wisdom.

The real question here is not technological but educational. I suspect that what Plato wrote about writing over 2000 years ago is also true for AI. Some students will use it to produce superficially impressive work, gaining the appearance of wisdom without the reality, exactly as the king predicted. Others will use it as a partner in thinking: testing ideas, exposing assumptions, refining their understanding.

AI can support independent thought, but it cannot create the motivation for independent thought on the student’s behalf. If a student has a genuine question, a real desire to understand and a willingness to have their assumptions challenged, AI can be a powerful tool. It can introduce new perspectives, expose weaknesses in an argument and help the student pursue an inquiry further than they could alone. If, however, the student’s main aim is simply to pass an examination, complete an assignment or impress someone else, AI use can be damaging because it makes it easier than ever to simulate understanding without developing it.

The catch is that to use AI as an independent thinker, one must already exercise some independence of thought: choosing the question, judging the response, resisting easy answers and remaining responsible for the final position. AI cannot give us a sense of purpose. Nor can it cultivate the communicative virtues needed to learn with and from others: virtues like open mindedness, humility, courage and integrity. Cultivating both a sense of purpose and these communicative virtues may be the central challenge facing education now. Perhaps it always should have been and the challenge of AI is helping us to see this more clearly.

At the recent Oracy Conference here in my college at Cambridge, Hughes Hall, I found myself talking with Kate Paradine, CEO of Voice 21, about the new report, chaired by Estelle Morris, on why white working-class children are being left behind in education. Kate emphasised one detail from the coverage. Mothers in Grimsby interviewed by the BBC were saying that what mattered to them was not “education”, but work. They wanted their children prepared for the local world of work and what the school offered seemed irrelevant to this, a kind of detour from their real lives.

Kate and I agreed that they had a good point. Those of us lucky enough to have had Saturday jobs and summer jobs learned a great deal from them: how to deal with customers, how to turn up on time, how to work with people we had not chosen, how to cope when things went wrong. That was educational, and often more motivating than school, because it was real.

So the question is not: how do we persuade working-class families to value education more? The better question is: why does education so often fail to feel connected with the forms of life and work that families value?

This is particularly striking from where I sit. Cambridge is often imagined as the archetypal ivory tower: ancient courts, echoing libraries, clever people talking to each other in a language no one else understands. But the image is misleading. I work in a Faculty of Education where we train teachers. In my college I meet people working in medicine, law, architecture, engineering, business and public policy, most of them deeply engaged with hospitals, industry, government and communities.

So there is a paradox. If even Cambridge, supposedly the peak of academia, is full of people connecting knowledge with practice, why does school still feel to so many families as if it is a detour from reality? I will come back to where I think that gap came from. First, let me describe something that suggests how it might be closed.

At a recent conference at Cambridge Judge Business School, colleagues described how they are using AI with executive MBA students. The traditional Harvard-style case method gives students a written account of a real business situation to read, analyse and argue over. The Cambridge team have gone further. Instead of a static brief, the AI-supported case is populated with voices. Students can interview the CEO, question the finance director, hear from workers, customers and other stakeholders, each with their own perspective, information and blind spots. The situation answers back, from several directions at once. Students have to weigh conflicting testimony, probe for what is not being said, and then commit to a judgement.

The feedback has been very positive, and no wonder. It is one thing to read about a problem. It is another to be placed inside it and asked to think, question, judge and act.

Why should this be reserved for adults already inside elite professional worlds? Imagine a mathematics lesson in which students run a market stall — pricing stock, working out wastage, and deciding whether to extend credit to a regular customer, while a rival stallholder undercuts them and the council warden threatens a fine. Imagine chemistry through a smokehouse where a batch has come out wrong, and the owner, the environmental health officer and an angry customer all tell a different story. Imagine physics and engineering through storm damage on the seafront, negotiating between the landlord, the tenant and the builder. Imagine biology and business through a community centre facing funding cuts, where the manager, the parents and the inspector all want different things and the students must find a way through. Imagine history and politics through the Cod Wars, where students must listen to voices carrying different narratives, grievances and hopes — trawlermen, dock owners, Icelandic fishermen, and a government that made promises it didn’t keep.

This is not about turning school into job training. It is about showing that every subject is already part of a living conversation about how we understand and act in the world. Nor is the idea to replace the teaching of knowledge with case-based and problem-based learning. Cognitive science warns, rightly, that novices can flounder when thrown into complex problems without the knowledge to make sense of them: if students did not know anything, they would have trouble understanding and responding well to the cases. The idea is to augment the existing knowledge-focused curriculum — perhaps accelerated with AI but with teachers guiding and framing throughout — and to show how that knowledge can be applied to understand real world situations and solve real world problems.

Education, after all, is much bigger than print. Most of what human beings learn, we do not learn from books or formal lessons. We learn by watching, joining in, copying, being corrected, trying again and gradually taking on more responsibility. This is true historically of education before print literacy took over but it remains true still in many situations outside of school, in the home, in the field, in the workshop, in the kitchen, in the factory, in the office and in the hospital. The separation of education from life is not the normal condition of human learning. It is the anomaly.

John Sweller offers one explanation for how the anomaly arose: formal schooling became necessary, he claims, because some knowledge, above all literacy, cannot easily be picked up just by joining in. Writing is a powerful technology, but it pulls learning away from immediate participation. That is a good defence of formal schooling — and it is also, I think, the answer to the paradox I raised earlier. But it suggests a possibility too: if one technology helped create the gap between school and life, another technology might help us mediate it.

My own work on dialogic education has been inspired partly by Michael Oakeshott, who taught history for many years at Gonville and Caius College, Cambridge, and described education as joining “the conversation of mankind”. That phrase still resonates for me. But we need to widen Oakeshott’s image beyond history, philosophy and classics. There is a global conversation about bread-making. There is a conversation about how to design a better boiler, how to organise a healthcare system, how to build houses, how to respond to climate change. Every trade and every practice is sustained by intelligent conversations, local and global, technical and ethical. The point of education is to induct students into these conversations so that they can participate in them.

This is where my three-loop model of dialogic education might be useful.

The first loop is learning to think through dialogue: to listen, question, give reasons, take the perspective of others, live with uncertainty and sometimes change your mind. We learn to think for ourselves by learning to think with others. Some children learn this around the family dinner table. Some do not. Schools rarely teach it systematically. This is why the new focus on oracy in the curriculum is so important for children who might otherwise feel unable to connect with formal education.

The second loop is induction into culture. No one thinks from nowhere. We inherit long conversations: mathematics, science, history, literature, technology, craft, politics, religion and work. These are not just bodies of information. They are living traditions of argument, evidence, judgement and practice. To learn history is to enter a conversation about evidence, causation, power and memory. To learn science is to join a disciplined conversation about how the world works. To learn a trade or craft is to enter a community of standards, distinctions and accumulated expertise.

The third loop is a kind of dialogue with the real world around us. Thinking and knowledge need to be brought to bear on solving real world challenges and also posing questions, not simply accepting the world but learning how to change it.

Schooling has concentrated mostly on a narrow version of the second loop, and has often delivered even that in a rather abstract, cut-off sort of way: knowledge presented as content to be memorised for examinations, detached from the conversations that produced it and the purposes it serves. The first loop, learning to think through dialogue, is largely left to chance. The third loop is deferred indefinitely: learn this now, you may need it later.

AI-supported, multi-voiced cases could bring all three loops together. In the first loop, AI can act as a Socratic partner, asking students, in groups, to clarify, distinguish, justify and reconsider. In the second loop, it can bring the voices of a discipline or practice into the classroom: the scientist, the historian, the engineer, the patient, the customer, the citizen. In the third loop, it can help students apply their thinking and knowledge to real problems where evidence is incomplete, perspectives conflict and judgement is required.

Students would still need knowledge, teachers and dialogue with one another. But they would apply that knowledge to real world situations, so that they learn through experience why knowledge matters. They would not simply learn about mathematics, science, history or work. They would begin to experience what it means to join the powerful knowledge rich dialogues through which those fields understand and change the world.

The mothers in those interviews were not wrong to value work. The mistake we have made as a culture is to make it possible to experience the world of work as distant from the world of education, as if one were real and the other somehow removed. Used badly, AI could make education even more artificial. Used well, and designed to support dialogue, it could help every student experience school as an invitation to join the conversation.

Adapted from Gustave Doré’s illustration of Dante’s Paradiso, Canto 31 (public domain).

Democracy is under strain. Trust in institutions is falling, political debate is hardening into hostility, and many citizens suspect that the challenges we face exceed the capacity of existing political systems to respond effectively. The rise of artificial intelligence has intensified these concerns. Will AI further undermine democratic self-government, or could it help develop new forms of collective intelligence better able to respond to the challenges we face?

Most discussions of AI and democracy focus on risks. I want to take seriously the possibility that AI could help renew democracy. Not because I am optimistic, but because I think we need a design stance. Hope here is not a feeling but a necessity. If we want to build a better future, we have to act as though improvement is possible. Criticism is necessary, but it is not enough. The challenge is to identify practical ways forward and begin building them.

The crisis is usually described as a crisis of institutions. But institutions depend on citizens. Democracy only works when citizens possess the capacities needed for collective self-government. If democracy is faltering, perhaps we are not educating people for it.

What democracy increasingly requires is dialogic intelligence: the ability of individuals and groups to think for themselves, learn through encounter with others, and participate responsibly in collective inquiry, judgement and decision-making. This means epistemic agency, perspective-taking, and participation in collective intelligence.

AI could support each of these capacities, but only against the grain of what the technology, left to its commercial defaults, wants to become. Most incentives push AI towards becoming an answer machine. Products are rewarded for frictionless answers, engagement is rewarded for confirmation rather than challenge, and convenience usually wins over effort. The uses described below are not what we will get by default. They are what we would have to deliberately design, fund and teach for.

1. Epistemic Agency

The first challenge is not disagreement but disinformation: conflicting claims, manipulated images and algorithmically amplified outrage.

One response is to provide more information. But information alone has never been the issue. Democracy requires citizens with epistemic agency: the capacity and willingness to interrogate sources, weigh evidence, recognise manipulation and exercise independent judgement.

This implies a particular relationship to AI. Rather than accept its outputs, students learn to question them, comparing responses, checking sources, surfacing hidden assumptions and testing claims against evidence. AI becomes an instrument for practising inquiry rather than a substitute for it.

However, questioning is not the whole story. Modern education, especially at university level, often rewards criticism more than judgement. Students learn to expose assumptions and identify weaknesses in arguments. These are valuable skills, but democracy also depends on the ability to make decisions and accept responsibility for them under conditions of uncertainty.

Here AI could support a more practical form of citizenship education at scale. The executive MBA case-study method offers a useful model. Learners are placed inside a complex situation, given incomplete information, asked to make a decision, and then required to defend and revise that decision as consequences unfold. AI could generalise this approach, generating realistic civic dilemmas, representing competing interests, introducing new evidence over time, and helping students reflect on the consequences of different choices.

The aim is not to replace judgement with AI. It is to give more students opportunities to practise judgement: moving from criticism to decision, from analysis to responsibility, and from passive commentary to democratic agency.

2. Perspective-Taking

The second challenge is polarisation.

Democracy does not require agreement. It requires citizens who understand why others think differently.

AI can be used as a scaffold for more constructive encounters between people. It can surface unnoticed areas of agreement and steelman the strongest arguments against one’s own position. When disagreement arises, it can provide historical, cultural and social context, encouraging participants to understand where different perspectives come from rather than dismissing them. It can prompt reflection on hidden assumptions and help sustain difficult conversations long enough for genuine understanding to emerge.

The goal is not tolerance. Tolerance is often little more than putting up with people one disagrees with. The goal is a dialogic democratic identity: a sense of self that combines commitment to one’s own voice with commitment to the larger space in which many voices meet, challenge one another and learn together.

Democracy depends not only on having a voice. It depends on learning how to listen. Not listening in order to reply, refute or defeat, but listening in order to understand.

3. Collective Intelligence

The third challenge is complexity. Climate change, migration, public health, technological transformation and economic inequality are not problems that individuals can fully understand, still less solve, on their own. They require collective intelligence.

But collective intelligence does not mean that everyone knows everything. In complex societies, much of what we know depends upon trust: trust in science, institutions, professions and systems of public reasoning that are larger than ourselves. The democratic challenge is not how every citizen can become an expert, but how citizens can locate themselves intelligently within wider fields of knowledge and judge when trust has been earned and when it has not.

This is where AI may have democratic value. Used carefully, it can help people see the structure of a debate: the main positions, the evidence supporting them, the fears and hopes motivating them, and the possible bridges between them. It can support both direct collective intelligence, where people reason together, and indirect collective intelligence, where people learn when and why to trust wider systems of expertise.

The platform Pol.is illustrates the first possibility. By revealing patterns of agreement and disagreement, it helps participants see where common ground exists. But such mapping is only the beginning. The real democratic work starts when people use these maps to understand opposing arguments, test their own assumptions and search for new proposals capable of attracting broader support.

The second task is perhaps harder. Most citizens cannot personally verify the work of epidemiologists, courts or statistical agencies. What they can examine is whether institutions admit mistakes, whether claims have been subjected to critical scrutiny, and who benefits if those claims are accepted. These are judgements that non-experts can reasonably make.

AI should not replace public reasoning. But, designed well, it could help convene, structure and sustain it at a scale that would otherwise be difficult to achieve.

The educational implication is clear. Students should not only learn to express opinions or master knowledge. They should learn how to participate in shared inquiry and how a non-expert can reasonably judge expertise without becoming an expert.

From Artificial Intelligence to Dialogic Intelligence

An AI built primarily to deliver answers encourages passivity. An AI built to support inquiry, dialogue and collaboration could cultivate precisely the capacities democracy depends upon.

Education alone will not solve the problem. Polarisation and fragmentation are also shaped by business models, platform design and political incentives. The capacities democracy requires are not necessarily the capacities markets reward. Outrage can be profitable. Simplification can be profitable. Manipulation can be profitable.

A democratic use of AI therefore requires more than technical innovation. It requires institutions, incentives and public policies aligned with the cultivation of dialogic intelligence.

Still, education remains a necessary starting point. We have some sense of where we need to get to, but it is far less clear how we get there from here. The present generation has largely been shaped by institutions and technologies designed for a different world. The wager of educational design is that the next generation can be shaped differently, cultivating the capacities needed to flourish in the world now emerging.

AI is already transforming education. The open question is what kind of citizens that transformation will create.

AI will not save democracy by giving us better answers.

But it may yet help revive democracy, if it helps us learn to think together better.

I was reading a post by recently and it made me uneasy. His claim was that too much educational research is not evidence-based but eminence-based: instead of testing ideas rigorously, we defer to persuasive theories and well-known names.

Some of the thinkers he criticised are people I respect. But unfortunately, I had to admit that he is not entirely wrong.

Education has often been vulnerable to ideas that feel right without being either conceptually grounded or properly tested. That raises an uncomfortable question: when people respond positively to concepts that I put forward like “dialogic education” or “expanding dialogic space,” are they responding because these ideas are good and useful, or simply because they sound nice and fit with their worldview?

What kind of discipline is education?

Kirschner contrasts education unfavourably with medicine, where controlled trials, replication, and systematic review have imposed a certain discipline. There is something to learn from that. Some educational questions really are like medical ones. If we want to know whether retrieval practice improves recall, we can test it directly and accumulate robust evidence.

But taking that as the model for all educational research implies a misunderstanding of the discipline.

Education is not only about what works best to achieve given goals. It is also about what those goals should be or, more fundamentally, what we should teach our children.

Disciplines are defined by their central questions. In medicine, the guiding question is about understanding, preserving, and restoring health: how to keep people well and make sick people better. Education asks what should we teach, and only then, as a secondary question, how should we teach it?

Where there is wide agreement about educational goals — the value of teaching young children basic literacy and numeracy, for example — then research of the medical kind is valuable. But if education were reduced to only that kind of research, it would become both conservative and narrow. Research on effective teaching inevitably focuses on how best to achieve given outcomes, rather than on questioning those outcomes themselves.

A more appropriate comparison might be architecture. Some aspects are technical and measurable. Architects need to know if their bricks or iron girders will support the weight, or if the plumbing will work. But these easily measurable and replicable factors do not tell us whether a building is any good. That depends on what kind of life it enables, what relationships it supports, and how it shapes those who inhabit it. Educational research works in the same way. It is a design science: not only asking how things are, as the natural sciences do, or how to fix what is broken, as medicine does, but how things ought to be. And in education, that means asking what is worth cultivating — asking what kind of future we would like for our grandchildren.

So what is education for?

When I trained as a teacher in Bristol in 1990, I became aware of something odd. Reading the official documents and talking with others on the course, it was not clear that anyone really knew what education was for. At one level, there were answers. A government report at the time, led by a banker rather than a Professor of Education, suggested that education existed to support the British economy — to help us compete internationally. Alongside that were familiar aims: socialisation, preparing people for roles, public health messages, and so on.

But none of these quite captured what seemed most essential.

As I began to look into this more deeply, and later during my PhD, it struck me that the answer might not be complicated. There is a widespread intuition about what education is for, even if it is rarely made explicit.

I saw this most clearly working with small groups of students in classrooms in Milton Keynes. A student would be struggling with a problem, locked into a particular way of seeing it. Then something would shift. There would be a pause. They would stop assuming they understood. Then they would begin to hold more than one perspective at once. They would question the frame that had constrained them. And then — often visibly — understanding would dawn.

This moment is familiar to teachers: the light going on.

What is happening here is not simply the acquisition of a correct answer. It is an opening of a space of reflection. That space is then widened — as multiple perspectives are held in relation — and deepened through questioning and integration.

This is what I mean by education as expanding dialogic space. It can sound abstract, but in practice it is simple and observable: it is what happens when understanding grows through the encounter of perspectives.

This process does not only occur at the level of individual learners. It also scales. As we develop knowledge, we bring more perspectives into relation and deepen the frameworks through which we understand the world. Despite setbacks, there has been a broad expansion of shared knowledge over time: not simply as an accumulation of facts but as an increasing capacity to relate perspectives in order to understand our situation and solve problems.

After many years of enquiry, my working hypothesis is this: education is for expanding dialogic space. What should be taught is dialogic intelligence — the capacity to think with others, including machines, by opening, widening, and deepening dialogue. How this should be taught is through structured activities that induct learners into effective participation in dialogic space.

The challenge of AI

Advocating for the importance of dialogue and expanding dialogic space in education is relevant precisely because it is counterfactual. Contemporary education systems are not organised explicitly around this aim. While moments of opening, widening, and deepening do occur in classrooms, they are not what the system is primarily designed to produce. The dominant structures of schooling — shaped by print literacy, standardised curricula, and assessment regimes — tend to privilege convergence, recall, and individual performance. The whole system seems to be built around high-stakes exams which do not have any obvious intrinsic educational function but are useful for certification, ranking, and the allocation of scarce resources.

The emergence of generative AI makes this misalignment between real education and our education system more visible. AI does not simply add a new tool to existing practices. It challenges the assumptions of a system built around the transmission and testing of fixed knowledge. If machines can now perform many of the tasks that print-based education has traditionally prioritised, then the question of what education is for can no longer be deferred.

This returns us to Kirschner’s concern. If educational research is to avoid becoming merely eminence-based — driven by persuasive ideas rather than cumulative evidence — then claims about aims must be grounded. The hypothesis proposed here is one perspective in the ongoing debate about what education is for. Research does not simply settle educational questions by declaring propositions true or false. It feeds a disciplined cultural debate about what is worth knowing, what is worth becoming, and what forms of life education should sustain. This is how knowledge develops in any serious discipline: not by the mere accumulation of facts, but through argument, interpretation, evidence, and challenge. In that process, identities shift, consciousness expands, and shared culture develops. Research matters because it keeps this debate answerable to experience rather than to preference, fashion, or authority.

What recent AI research shows

A body of recent empirical work on AI in education is beginning to reveal when and how AI expands dialogic space — and when it contracts it. Across four studies, the lesson is fairly consistent: AI can help to open and support dialogic space when it is designed as a moderator, role-holder, reflector, or provocateur, not merely as an answer machine.

Risk & promise

Feng — GPT-4 in collaborative problem-solving

Six small groups used a GPT-4 chatbot for collaborative problem-solving. The key warning: AI can become the centre of interaction. Students may talk more to the chatbot than to each other. That expands access to ideas, but can shrink peer-to-peer dialogic space if the framing is wrong. Design matters as much as the technology.

Social dynamics

Doherty et al. — LLM jigsaw-partner agents

LLM agents used in jigsaw-partner roles reduced social loafing and produced more respectful collaboration language. But the researchers stress an important caveat: LLMs can also reinforce poor social dynamics unless carefully designed. The quality of the dialogue depends on the quality of the design, not on AI capability alone.

Building community

Breideband et al— CoBi, dialogue as object of inquiry

Students set community norms for their discussion; CoBi listened for examples of those norms; the class then reflected on the AI’s output together. This made dialogue itself an object of inquiry, rather than using AI simply to complete a task. An interesting experiment in using AI to help build community and collaboration.

Widening perspectives

Nguyen et al. — role-based environmental science chatbots

Multiple chatbot personas — fisherman, student, influencer, researcher, engineer — gave students access to situated perspectives on climate change and local ecosystems. This is a clear design for widening dialogic space by multiplying voices. Students were not receiving a single authoritative account but navigating a range of positioned ones.

The pattern across these studies points toward a design principle: AI can widen the space of voices and hold open the space of inquiry. This requires intentional design. The default behaviour of most AI systems — efficient, helpful, convergent — works directly against it.

What we still need: measures

The problem is not a lack of evidence. It is a lack of measures that would allow this evidence to accumulate coherently — measures of dialogic space itself, not just of proxies like test scores or talk-time. If “expanding dialogic space” is to anchor a research programme, it needs to be observable. As a starting point, the following provisional definitions are proposed:

Six provisional dimensions of dialogic space

Questioning: Do contributions open new directions, surface assumptions, or reframe the problem?

Widening: Is a broader range of perspectives brought into play and genuinely engaged with?

Deepening: Are underlying assumptions examined, leading to more integrated understanding?

Integration: Do new responses emerge that incorporate previously incompatible positions?

Judgement: Are decisions made that draw on multiple perspectives and navigate uncertainty?

Sustaining: Is engagement maintained in the presence of disagreement without collapse?

The challenge is to turn these verbal pointers into indicators that can be observed and refined without reducing them to tick-boxes that capture the surface of dialogue while missing its substance.

A research strategy

Design-based research is appropriate here because the aim is not simply to test predefined variables, but to iteratively develop and refine both concepts and practices. Small-scale studies can explore how different AI-supported designs affect dialogic processes. Over time, through comparison and refinement, more stable constructs and design principles can emerge.

Central to this is the development of a research community. Teachers, researchers, and designers need to share cases, compare findings, and jointly refine both theory and method. The community is not secondary to the research. It is the primary site in which knowledge is constructed.

Only once constructs are sufficiently clear does it make sense to move towards larger-scale validation.

An invitation

The broader aim is a reorientation of education around the expansion of dialogic space — understood as the development of collective wisdom: the capacity to engage with complexity, difference, and uncertainty in ways that support human flourishing through shared processes of judgement and decision.

This is an open research programme. It will only develop through shared critique, comparison, and collaboration.

The arrival of generative AI challenges the global education system so deeply that it gives us an opportunity to ask again a fundamental question: what is education for?

In an earlier blog I suggested that AI could support a search for meaning by connecting learners to real-world problems and larger systems. On reflection, however, this answer feels a bit thin. Connecting to bigger networks, the global dialogue of scientific inquiry, for example, might work for some people, but it will not work for everyone. It presupposes what first needs to be formed: people open to finding meaning through participation.

Meaning runs deep

The deeper challenge for education is that meaning is not usually formed consciously or individually. Human beings are shaped pre-reflectively through embodiment, relationships, language, ritual, and culture long before they are able to articulate beliefs about themselves or the world. It is through induction into forms of life, within families, communities, friendships, and institutions, that we first learn who we are, what matters, whether the world is trustworthy, and whether life is worth living.

The wellbeing data here is striking. Children raised within religious communities consistently show lower rates of depression, anxiety, addiction, and suicide, alongside a stronger sense of meaning and belonging. This finding comes not from religious apologetics but from mainstream social science, including the work of Jonathan Haidt, a self-described atheist who has written candidly about where the evidence appears to lead. State education systems must serve children of all faiths and none. But that should not prevent us asking what religious traditions still understand about formation, ritual, belonging, and moral purpose that secular institutions may be able to learn from.

The hidden curriculum

If we do not shape culture consciously, culture will shape people unconsciously anyway. Much contemporary education rests on assumptions it rarely makes explicit. Science, as commonly taught, can imply that the universe is mechanical and indifferent. Literature often encourages suspicion toward the idea that meaning genuinely resides in the world. History, as taught in schools, often dissolves larger narratives into contingency and struggles for power. Official aims speak of teaching for awe, creativity, and human flourishing, yet the hidden curriculum transmitted through assessment, credentialism, and competition points elsewhere.

Into this vacuum students are offered only one clear path to meaning: personal success. Whether in school exam results or comparisons of lifestyles on TikTok, it becomes clear that being ordinary is not enough. What counts is individual achievement, status, or success. The results are visible. Not wisdom or belonging, but anxiety, isolation, and the sense that meaning must be manufactured alone against a universe that offers none of its own.

Schools once existed within thicker webs of formation where families, neighbourhoods, religious communities, and civic cultures all shared responsibility for shaping identity and purpose. Today social media and AI are increasingly absorbing that role without democratic mandate or accountability.

One of the strongest critiques of social media and AI is that they shape us unconsciously. Haidt and others are right to worry. Yet the standard response often misunderstands the problem in a way that actually serves further to contribute to it. We are told that we need more transparency, more individual control over data, or simply banning phones for young people until they are old enough to give full consent. Beneath these responses lies a questionable assumption: that there exists a fully autonomous self standing outside culture, capable of giving consent and rationally choosing its own formation once sufficiently informed.

But we humans are always already formed through language, education, institutions, media, and relationships. This is not a failure of freedom. It is the condition through which meaning and freedom become possible. The self is not prior to formation but emerges through it.

The question, then, is not whether technology shapes us. It always will. The question is whether we can consciously and collectively take responsibility for shaping the culture that shapes us.

Education and faith

Education presupposes faith in the possibility of transformation: the possibility of learning, growth, and deepened understanding. Without this faith the very idea of education becomes self-contradictory.

Progress through education and inquiry is not just about science and technology. In earlier blogs I quoted Plato. We still read Plato more than two thousand years after he wrote. Philosophy has been described as ‘footnotes to Plato’ but progress in philosophy is real. It takes the form of expanded dialogue. Where once a few carefully preserved texts were read with reverence in medieval universities, today almost anyone can read Plato in dialogue with Confucius, Buddhism, feminism, ecology, postcolonial thought, and neuroscience and learn from participation in this expanded dialogue.

Progress then is not a line from A to B, from error to truth, but from A to A-plus-B: through widening relationships, expanding perspectives, and deepening reflexive awareness. Technology has made this possible. Meaning, for those with sufficient faith in the larger tradition of education, grows through participation in expanding processes of inquiry, connection, and shared understanding.

Modern science itself did not emerge from a rejection of religious faith, as many now claim, but largely from developments within Christianity grounded in the conviction that a rational Creator had made a world that rational creatures could investigate. Medieval Islamic scholarship made indispensable contributions in mathematics, optics, and medicine. Rational inquiry into the conditions that support human psychological flourishing is central to strands of Hinduism, Buddhism, and other philosophical and religious traditions.

Science, in this broader sense, is disciplined collective inquiry, not only into the apparently outer world, but also into the inner world of experience, ethics, and spirituality. This collective inquiry is not the opposite of faith but one historical continuation of humanity’s search for truth through dialogue, humility before evidence, and openness to revision. Inquiry generates its own ethical demands: honesty, patience, courage, trust, and willingness to learn from others. A genuine culture of inquiry is already a culture of formation.

The body and ritual

Religious traditions remind us of something modern education consistently underestimates. Human beings are not shaped by ideas alone but through shared practices and embodied participation: through singing, mourning, celebrating, gathering, spending time in nature, and participating in rituals and stories that connect individual lives to larger patterns of meaning. Meaning is often lived collectively before it is articulated intellectually.

Modern education still contains rituals, but many are organised primarily around testing, ranking, performance, and competition rather than around shared inquiry and meaningful participation in collective life.

Used badly, AI will intensify the worst tendencies of the existing system: passive consumption, cognitive outsourcing, manipulation, and the shallow optimisation of measurable outcomes at the expense of everything that matters.

Used well, however, AI could help education recover what modern institutions have partly lost. It could connect learners across cultures around shared human questions, stage dialogues between traditions separated by language and history, and enable forms of collective inquiry at scales previously impossible. It could support sustained reflection, collaborative creation, and practices of dialogue that reconnect individuals to something larger than themselves: not through indoctrination, but through participation in humanity’s ongoing conversation about truth, meaning, and the future.

The important point is that social media and AI systems are never neutral. They already shape habits of attention, norms of communication, assumptions about success, and even emotional tone through the forms of feedback they provide. Current social media often reward speed, outrage, self-promotion, and compulsive engagement because these increase measurable activity. But different systems could be designed around different formative goals.

AI tutors, moderators, and social platforms could instead be designed to reward curiosity, listening, intellectual humility, courage in questioning assumptions, collaborative problem-solving, and sustained inquiry. Rather than maximising addictive engagement, systems could encourage periods of reflection, dialogue in diverse groups, participation in shared projects, or even practices that reconnect people with embodiment, community, and the natural world. AI could help organise collective rituals of inquiry and reflection at local and global scales in ways that social media platforms currently do not.

The issue, then, is not whether AI will shape human beings. It already does. The issue is whether we consciously shape these systems around educational and civilisational values supportive of human flourishing.

Conclusion

Education always inducts students into a worldview and a way of being in the world, consciously or not. The real question is what kind of worldview and way of being in the world we now wish to cultivate. Not dogmatic certainty. Not technological utopianism. Not atomised individualism. But a renewed faith in the practice of collective inquiry itself — faith in the possibility that through dialogue, imagination, science broadly understood, and shared practices of creative sharing and learning, human beings can continue expanding their world of meaning together.

Science in this broader sense is not merely a collection of disciplines. It is a civilisation-scale practice of collective learning through which humanity reflects on itself, tests ideas against reality, revises assumptions, and consciously participates in shaping its own development. Democracy in this deeper sense is not merely a mechanism for aggregating preferences but the collective practice of self-determination: the ongoing attempt of a society to reflect upon, question, and consciously reshape the conditions of shared life. We can be proud of what we have achieved so far and what we might yet achieve in the future and not ashamed to induct our young into that shared pride and that shared project.

The educational challenge of the age of AI is therefore not simply how to use new tools efficiently. It is whether education can help cultivate forms of collective intelligence and collective wisdom capable not only of sustaining but also of calling into being a self-reflective, self-organising, and consciously self-evolving planetary culture.

That, at its deepest level, is what education has always been for.

Talking with young people, and listening to the news, I keep arriving at the same uncomfortable conclusion: loss of meaning has become a genuine crisis. Health data points to declining healthy life years, linked to inactivity, poor diet, and drug use. Surveys of teenagers show rising anxiety, deepening pessimism, and a collapse of trust in institutions.

Education is implicated in this. Across cultures and throughout history, one of its central functions has been to induct the young into a sense of meaning and purpose. Among the explanations most often offered for the current malaise, technology figures prominently. The argument has a familiar form: as machines take on more roles, something essentially human is diminished; attention is captured, habits become passive, agency erodes.

There is some truth in this. But it rests on a misunderstanding of both meaning and technology.

Where meaning actually comes from

Meaning is often treated as a private feeling that individuals either have or lack. Framed in this way, it becomes natural to contrast meaningful human life with supposedly meaningless machines. But if we look more closely at when meaning actually appears, a different picture emerges.

Meaning arises through participation in processes that extend beyond the individual: when a problem is worked through rather than bypassed; when a contribution is taken up and transformed by others; when local activity connects to something larger.

Consider what happens when someone engages seriously with a difficult problem. At the outset there is genuine indeterminacy: the problem resists, the path is unclear. Resolution, when it comes, involves a reorganisation of understanding. The person who emerges is not quite the same as the one who began.

Work becomes meaningful not because of its content alone, but because of the relation the worker has to the process: whether they can recognise their contribution within a larger movement, and whether they participate in shaping it.

Technology as a medium for participation

Seen in this light, technology is not the enemy of meaning. In some form technology is integral to the achievement of meaning.

Writing, institutions, disciplinary traditions and communication networks are all technologies that connect local action to wider processes. AI is the latest and most powerful instance of this.

The question, then, is not whether to use it, but how its use structures participation. Used in ways that bypass tension and struggle, AI undermines the conditions under which meaning arises.

But the same technology can reduce alienation. It can make complex domains more navigable, connect individuals to conversations they would otherwise be excluded from, and expand the scope within which genuine participation is possible. The technology does not determine the outcome. The structure of engagement does.

Why dialogue clarifies the issue

In classroom work on the Thinking Together approach, it was often striking how absorbed children became in problems that would in other circumstances disengage them. The bell would go, and instead of leaving, they wanted to continue in order to finish the argument, test a counter-argument, see where the shared thinking would lead. Meaning had emerged in the interaction, and it was tangible.

Meaning is neither something we possess privately nor something imposed from outside. It emerges in the act of responding to a question, to another person, to a challenge, to a world that makes demands on us.

Some accounts of meaning begin with the isolated individual, who imposes meaning on a passive world. Others dissolve the individual into larger systems such as society, history, or nature. Dialogue offers a different, more dynamic way of thinking about meaning. It situates meaning in the space between: an unfinished process that both connects and transforms those who participate in it.

To enter dialogue is to find a voice — but only by responding to something beyond oneself.

How education should respond

The issue with AI is not that it is inhuman. It is that it can be used in ways that are alienating for example by doing the work for the learner, or for the worker, bypassing judgement and effort.

But equally, it can extend agency: enabling more demanding and consequential forms of engagement with problems that matter.

The task of education is therefore to form identities oriented towards contribution — learning how to take a position within a field of dialogue, how to respond, and how to shape what is being done.

The world is not short of problems. What is lacking is the conviction that engaging with them makes a difference.

That loss of conviction is perhaps the most serious educational failure.

There are already signs of what is possible. In science, design, medicine, and social enterprise, individuals working with AI are beginning to achieve exciting things.

If we are to respond to the crisis of meaning among young people, education must centre on meaningful participation: engaging with real problems, using powerful new tools, and recognising how that engagement reshapes both the world and the self.

Without this, AI will be experienced as diminishing. With it, the same technology can become a medium through which purpose is found.

I recently spent a day with a group of academic leaders in the arts and humanities talking about higher education after AI. What struck me was how uneasy many of them were with the idea of progress. When I said I felt optimistic about the future with AI, one distinguished professor replied that he could not understand such optimism. The people of the future, he said, will be so different from us that we cannot know their world or judge whether it will be better.

I could see the point. New technologies do not automatically improve life. It is not obvious that a modern urban life, dependent on complex systems, is always better than older, smaller-scale, or more traditional ways of living.

And yet I felt morally challenged by this scepticism. For those who now survive diseases that once killed, for those who have clean water where before there was none, for parents whose children live where once they would have died, progress is not an illusion. It is something real. Scientific and technological advances have reduced suffering and expanded human possibilities.

But in fact, I suspect many of those same humanities scholars would agree that education itself depends upon a real idea of progress. We educate because we believe people can become more aware, more reflective, and more capable of understanding other points of view. In my terms, that is progress through the expansion of dialogic space: the widening and deepening of the space in which different perspectives can meet, challenge one another, and generate new understanding.

The problem, perhaps, is that many people in the humanities still think of technology as something external to the human. On that view, machines may transform the world around us, and technological development may show a clear cumulative progress across history, but human understanding is seen as something different. It does not simply accumulate. It has to be renewed in each generation, and in each individual student. I think that is mistaken. Human beings have always been shaped through technology. Language is a technology. Writing is a technology. Print is a technology. And now digital networks and AI are technologies that shape how we think, remember, imagine, and communicate.

Once we see this, the contrast between human progress and technological progress begins to break down. Technologies do not just give us new tools. They change the conditions of thought itself. A modern mathematician, for example, does not think alone, but through a vast inherited structure of symbols, proofs, and concepts built up over centuries. In the same way, AI offers a new layer of support for thinking. It can help us search, compare, test, and reframe ideas. But it can also make us lazy, dependent, and passive.

That is why the question of progress is now an educational question. The issue is not whether technology is simply good or bad. The issue is what kinds of human-technology relations help us think better, and what kinds weaken judgement.

This is where I think that AI literacy becomes needed. AI literacy is not just knowing which tools exist or how to write a prompt. It is knowing how to use AI in ways that expand thought rather than replace it.

Take two students writing an essay on climate justice. One asks AI to write the essay and submits a lightly edited version. The other uses AI to test an argument, find the strongest objection, compare perspectives, and discover what they still need to understand. The technology is the same, but the educational result is very different. In the first case, thought is outsourced. In the second, thought is extended.

This is why I think we still need the idea of progress, but we need to understand it differently. Progress is not a simple story of history automatically getting better. It is not inevitable, and it is not uniform. But it is real enough to guide judgement. Progress happens when education and technology work together to expand what I call dialogic space: the space in which different perspectives can meet, challenge one another, and generate new understanding.

Technologies can help expand that space, or they can close it down. Social media can connect people across the globe, but it can also trap them in polarised loops. AI can support curiosity, reflection, and creativity, but it can also encourage imitation, dependence, and the illusion of understanding. Nothing is guaranteed.

This is why education matters more than ever. The central task is not to protect a pure humanity from technology. It is to teach people how to live and think well with technology. In the age of AI, that means learning to recognise the difference between a use of technology that expands thought and a use that replaces it. That, to my mind, is the heart of AI literacy.

At a recent online meeting about AI and education, a leading academic objected strongly to the phrase “hybrid intelligence”. AI, this person argued, is like a pair of glasses. Glasses may shape what someone can do, and may have a significant effect on a life, but they remain a tool. To ascribe agency to glasses is a misuse of language. To describe a person wearing glasses as a cyborg or hybrid might be technically true in some trivial sense, but it is not useful for understanding what it means to be human.

To dramatise the point, this person mimed stamping their glasses underfoot — quite difficult to do on a Zoom call — and said that, even without them, they remained pretty much the same person.

Behind the objection was a serious political concern: if we describe AI as agentic, intelligent, or as a genuine partner in thought, we may end up helping big tech companies present it as a replacement for human beings.

The academic looking in the mirror saw a human animal wearing glasses and did not identify with the glasses. Presumably they identified more with the organic flesh and bone. But this already assumes that who we are is best understood from the outside, as an object in the world, rather than from the inside, as a living process of perception and meaning-making.

From the inside, the experience is different. Once worn, glasses are no longer simply external. They withdraw from attention and become part of the structure of seeing itself. The glasses, like the retina and the optic nerve, are part of the condition under which the world appears to you.

The same is true, more profoundly, of language, literacy, and education. These are not external instruments that leave thought untouched. They become part of the way thought happens. The very concepts with which the academic criticised hybridity, concepts such as “agency”, “tool”, and “misuse of language”, are themselves products of a history of education, writing, and print literacy. Technology in this form shapes thought from within.

Human intelligence has always been formed through language, symbols, writing, institutions, and tools. We are always already hybrid. The real question is whether working with AI might open up new and perhaps better forms of hybrid thinking.

Why agency matters

Agency is not a simple term. It admits of degrees, and the interesting question is always which degree we are talking about.

A river has causal power. A thermostat regulates temperature. Neither has moral responsibility. An AI system that filters your email, proposes actions, books meetings, summarises documents, and prompts decisions clearly does more than a pair of glasses or a hammer does. It does not merely focus light or extend force. It selects, prioritises, and initiates within a defined field. That is a real kind of agency, even if it is not yet moral agency.

So perhaps AI does not “have agency” in the full human sense. But it has some kind of agency. We might say that AI can exhibit operational agency without possessing moral agency. Put that way, the claim becomes less alarming and more precise.

In fact, even the human claim to agency is less secure than we often assume. We like to imagine ourselves as conscious authors of our decisions, clearly knowing why we do what we do. But psychology offers a more modest picture. People often confabulate reasons after the event. Actions emerge from habits, contexts, impulses, and social cues long before they are neatly gathered under the story of an autonomous self acting for good reasons.

This does not mean human agency is unreal. It means it is situated, mediated, and always distributed across communities that include both people and technologies. Our most effective agency often lies not in sovereign moments of choice but in the slower capacity to shape the environments, habits, institutions, and relationships within which action becomes possible. We can design our environments and our future selves in ways that increase our capacity for moral agency. I think that this is what education is for.

What then is AI literacy?

AI is not only another tool to be used efficiently to achieve pre-given ends. It is a new medium of thought. Print literacy changed cognition by making certain habits of mind easier: linear argument, explicit analysis, separation of variables, cumulative reasoning across extended texts. These habits came to seem natural, but they were historically produced.

AI literacy may be similarly transformative. It can make it easier to externalise half-formed thoughts and develop them through dialogue. It can make it easier to generate alternatives, test assumptions, connect domains, simulate consequences, and revise ideas iteratively. It invites a style of thinking that is less about producing a finished statement in one go and more about reframing, probing, comparing, and redesigning.

This does not mean AI makes people more intelligent by default. It can just as easily support laziness, offloading, and conformity. Everything depends on how it is framed and used. A lot depends on the model built by AI providers, but a lot also depends on the pedagogy, on how we teach AI literacy. If approached instrumentally, as a shortcut to outputs, it may well diminish thought. But if approached dialogically, as a way of opening, widening, and deepening reflection, it can support new forms of individual and collective intelligence.

AI does not yet inhabit a world of meaning in the way human beings do. It does not desire, care, answer, or assume responsibility in the full sense. But it can participate in human processes of meaning-making in ways that reshape how thinking happens. That is already enough to make AI literacy far more than a technical competence.

In the end, the question is not whether AI is more like a pair of glasses or more like a colleague. It is whether we learn to use it in ways that make us more reflective, more creative, and more capable of acting well together. That is what AI literacy is for.

What is education technology actually for?

A few years ago, during a visit to South Africa, my thinking about technology in education shifted. In a museum in Cape Town, I came across David Lewis-Williams’ The Mind in the Cave, which argues that San rock paintings, some of which date back many thousands of years, can be read as an early form of education technology. They supported initiation practices through which a community inducted newcomers into a shared cultural world.

In this account, the animals on the walls are not there to teach hunting technique. They are cultural presences, voices through which the tribe understands itself and sustains itself across generations. For the San, the eland was not merely an animal to hunt but a being saturated with spiritual power. It embodied a concentrated life-force that shamans sought to enter during trance dances and initiation rituals. Unlike the Mantis which deceived with clever tricks, the Eland brought deeper insight and personal transformation. To be inducted is not simply to be shown images. It is to be drawn into practices in which these voices can be heard and, in a sense, spoken from within. Trance dance, guided interpretation, story, touch, and ritual enable these presences to enter and transform new members of the community. Possession, in this sense, is participation. It is how a biological individual becomes a bearer of a culture, able to feel, perceive, and act from within an ongoing collective life.

This reframes what “education technology” means. Education has always been mediated by technology, from cave painting to writing, from print to networked digital media. Yet it is never mediated by technology alone. It is mediated by technology together with pedagogy, with the routines, rituals, institutions, and assessment practices that shape not just how a technology is used, but what it is used for. Different technologies, coupled with different pedagogies, make possible different forms of thinking, and also different forms of belonging.

What a technology makes possible

If we look at cave painting this way, we can see both its power and its limits. It can carry a world through time. It can hold an identity together and bring people into that identity with emotional force. But it is local. It speaks most fully to those who are near the site, near the stories, near the guides who can interpret and perform the meanings. It can bind a group together, yet it does not easily bind many groups together across distance.

Writing changed that, especially when joined to lightweight portable media like papyrus. Text can travel across territory, be copied, stored, and applied in many contexts. That helped support larger-scale coordination such as empires and religions, legal systems and schools. It also stabilised new styles of thinking. Images tend to support holistic and poetic modes of feeling and thinking. Writing can support those too, but it also makes possible sustained analysis, fine-grained distinction, abstraction, and the accumulation of technical knowledge.

None of this happens through the medium alone. It happens through the coupling of medium and educational practice.

That brings us to the urgent question of the present. What are the new possibilities and the new risks opened up by AI-enhanced digital media? What new kinds of induction are becoming possible, and what new ways of thinking and belonging might follow? To start to answer that question, it helps to look at what is happening in China, where universities and policy actors are consciously trying to anticipate and shape the future through educational innovation. If we want to see early, large-scale experiments in the future of higher education, developments such as MAIC at Tsinghua University, and the MashOn and TruFar ecosystem at BUPT, are worth paying close attention to.

MAIC at Tsinghua: from MOOCs to something that can actually teach

For years the most visible attempt to scale higher education online has been the MOOC — the Massive Open Online Course. MOOCs scale content. They give many people access to lectures, readings, quizzes, and forums. Yet they rarely scale what students most need when learning becomes difficult: responsive teaching, sustained guidance, and a sense of learning alongside others. Completion rates remain stubbornly low, and it is not hard to see why. Being left alone with a fixed sequence of resources is not the same as being taught.

MAIC — Massive AI-empowered Courses — is designed to correct for this. The underlying idea is to scale something closer to a teaching relationship and a classroom ecology. A course is treated not only as a package of materials but as a pattern of interaction over time: questions, explanation, feedback, practice, and the management of attention and effort. Students interact with an AI layer that can take on different roles such as an instructor voice, a teaching assistant, sometimes also a peer in this way providing dialogue, rapid clarification, and structured practice so that learners are not left alone with content. They refer to this as a ‘1 student + N AI agents’ model because it is not just a single tutor, but a structured ensemble of dialogic partners. The question, however, is not how many agents they can deploy, but how their orchestration expands the space of reasoning.

A second emphasis is on what one might call learning traces. When learning interactions are mediated digitally, they become visible to students, to teachers, and to course designers. In the best case, that creates feedback at multiple levels simultaneously: the student learns where they are going wrong; the teacher sees patterns across a whole cohort; and the course itself becomes a living design that can be refined over time rather than a fixed artefact delivered once and left unchanged.

The initial large-scale use of MAIC at Tsinghua has been to teach a course on AI literacy. This is revealing. Perhaps one emerging role of the university is not to resist AI, but to induct students into disciplined, critical, and dialogic engagement with it

Making things together at BUPT

MashOn is a guided tutoring layer built to avoid simply handing over answers. It is designed to push students to reason rather than retrieve. TruFar is the larger platform around it: it is an environment in which teaching and learning tools, or “agents,” can be built, shared, and reused across an institution. This matters because it shifts the emphasis from consumption to creation.

When I visited BUPT recently, I was shown many projects using the system to build educational resources, especially games. One student team, for example, had built a game for learning to write in English. Players were presented with short stories which were flawed, funny and sometimes deliberately absurd and they had to spot errors, suggest improvements, and make editorial judgements about the texts. Get it right and animations fired: characters celebrated, scores ticked upward and new levels were unlocked. It was unambiguously a game, and an engaging one. But underneath the animations was a structured encounter with grammar, narrative logic, and the feel of good prose. The team had not just learned about writing by building it. They had made something that could teach others, and that others would actually want to use.

This is the pedagogy that has grown around the TruFar infrastructure, and it changes what learning looks like. The unit of learning becomes a team producing an artefact which could be a game like this one, a simulator, a guided inquiry sequence, a role-play environment, or a study companion for a course. Importantly, not all of this requires conventional programming expertise. Because the system supports what is now called “vibe coding” or specifying what you want a piece of software to do in ordinary language, testing it quickly, refining it, and only dropping into code where genuinely necessary, the approach extends beyond computer science. A group of history or literature students can build something that embodies their disciplinary understanding and makes it teachable to others.

Assessment shifts accordingly. It is no longer primarily a measure of what an individual can reproduce, but of what a group can design and sustain together: what they build, what it enables others to learn, and what they themselves come to understand through iteration and refinement. Collaboration, negotiation, empathy with users, and reflective judgement are not peripheral skills. They become the medium of learning itself. And alongside this, students learn something equally significant: how to work with AI in ways that generate added value.

From the eland to AI agents

The eland matters here because it points to something modern education often forgets. Education is never only about information and skills. It is also about identity and emotion, about being drawn into a world that feels meaningful, and about becoming the kind of person who can speak and act with others in ways that change the world. The history of education technology is therefore not simply a history of tools. It is a history of ways of carrying a shared world, sustaining collective intelligence, and enlarging the self through participation.

Generative AI adds a new layer to that story. It makes possible interactive synthesis over networked knowledge, conversational externalisation of thought, multi-voice simulation, and scalable forms of guidance. The promise behind MAIC and behind TruFar and MashOn is that these affordances can be shaped into forms of higher education that are not merely efficient, but genuinely developmental.

Yet neither system replaces the university, and it would be a mistake to think they could. The drivers of sustained learning still depend on social structure: recognition, belonging, shared rhythm, and credible assessment. If a student has no reason to persist, and no community in which their learning matters, even the best AI tutor will fail. These new systems make most sense when they are integrated into institutional teaching and connected to assessments that reward reasoning, judgement, and growth rather than mere output.

Universities won’t vanish. But they may become less about delivery and more about convening: seminars, tutorials, supervision, and projects that anchor a wider, technologically mediated learning ecology, extending beyond the campus and into lifelong learning. The question then becomes: what is the university for, in the age of AI? One answer may be that it is where students learn disciplined ways of working with AI, ways that deepen inquiry and enable them to create new forms of value that serve the public good.

Induction into dialogue, not into conclusions

The outcome, though, is not guaranteed. A great deal depends on pedagogical intentions, assessment design, the governance of data, and the role teachers play. Paulo Freire named the underlying tension decades ago. Education can become a way of imposing a stable order on minds, controlling and limiting them. Or it can become a practice of growth, in which people learn to name the world, question it, and participate in dialogue with it. The difference between these two futures is not always obvious in advance. We can make choices that narrow possibilities almost without noticing we are doing so.

Yet education must always induct newcomers into some order, some language, some culture, some way of seeing. That is unavoidable. The task, I think, is to induct people not into conclusions reached for them in advance, but into participation in the dialogues through which new futures are continuously being made.

The collective intelligence that education supports should not be only a defensive reaction, designed to avoid threats and maintain cohesion. It should also be an induction into deliberation about what possible futures look like, about how we judge which futures are more desirable, and about how we might work together to reach them. That is perhaps already what the eland was doing on the cave wall, in its own way.

My PhD student Bo Yu, a member of the Cambridge AI&ED community, is organising an event to share lessons from MAICs and from TruFar and MashOn more widely, while also fostering meaningful dialogue with leading scholars in AI in higher education from UCL and LSE. It is open to all on 17th March — details here: AI in Higher Education: A UK–China Dialogue – Registration – Fill in form

I was struck by a word used in a review of my last book: contamination. My book argues that education is most fundamentally about expanding dialogue, and the reviewer responded with a reference to Anna Lowenhaupt Tsing’s The Mushroom at the End of the World, where contamination is often presented positively as the way encounters with the strange and unfamiliar change us. In Tsing’s account, contamination is not corruption but creativity. New forms of life emerge through entanglement. Her emblem is the matsutake mushroom, which grows in devastated landscapes through complex collaborations between fungi, bacteria, trees, soil, and people—and which is also delicious. Contamination, here, is how life goes on.

But it was the negative resonance of the word contamination that made me pause. At a recent education conference in Hawaiʻi, I heard a story that challenged my earlier emphases. I learned about local history from an Indigenous Hawaiian keynote speaker. When James Cook first arrived in 1778, he encountered a people who, by many accounts, appeared extraordinarily curious, playful, open, and generous. There was hospitality and exchange, laughter and intimacy.

Within decades, the consequences were catastrophic. At first, the catastrophe came not through violence but through disease. Contamination arrived through touch, breath, intimacy, and trust. A succession of epidemics tore through the islands, with devastating effects on population and social life. And within a century, biological collapse was followed by political takeover. In 1893, there was a coup against Queen Liliʻuokalani backed by U.S. Marines. What followed was not only the loss of sovereignty, but the remaking of culture with compulsory schooling conducted in English, and periods in which the Hawaiian language and cultural practices were marginalized or punished. It seems that in this case a stance of “openness to the other” did not work out so well.

This story has forced me to rethink my enthusiasm for openness as the defining characteristic of dialogue. I had been shaped by an educational concern with closed minds to think of learning as expanding horizons and unsettling certainties. That concern remains valid. But I now see more clearly that dialogue has two distinct failure modes: excessive closure (rigidity, defensiveness, refusal to engage with difference) and excessive openness (collapse of boundaries, loss of form, destructive contamination).

The story of Hawaiʻi raises another troubling issue: the impact of differences in power. The ideal of constructive dialogue tends to presuppose reciprocity, which requires at least enough symmetry that each party can risk being changed without being erased. History is full of encounters that lacked such symmetry with differences not only of power but of immunity, weapons, wealth, legal authority, and the ability to define what counts as “reasonable” or “civilized.” In situations of imbalanced power, “openness” can easily become exposure to exploitation and abuse.

Yet contamination is not only negative. Even deeply unequal encounters can give rise to new hybrid cultural forms such as those visible today in the flourishing of contemporary Hawaiian–American art, music, dance, and environmental sensibilities, and in their wider influence on American culture. This does not negate the injustice, but it shows why the question is not whether to risk contamination, but how to manage it. The challenge is maintaining enough symmetry that encounter can be creative rather than catastrophic such that parties can risk transformation without facing erasure.

None of this means that dialogue cannot work in educational contexts. Classrooms and online learning environments can be made safe enough for learners to take risks and be transformed. But it does suggest that we need to be careful about what we mean by dialogue, and about the conditions it requires.

This issue matters urgently today. When social media first arrived, I was enthusiastic about its potential to support dialogue across distance. Unfortunately, not everything has gone well. Many of those now calling for bans on social media in schools aren’t against the technology itself. They are saying: this technology is not working properly yet; take it away and fix it; and only then will we welcome it back into our lives.

With the advent of AI, there is a new temptation. An alien intelligence has essentially arrived on our shores. It seems friendly, so we want to welcome it wholesale into our cultures, classrooms, and childhoods. But perhaps the experience of Hawaii should make us pause and ask: would that be wise?

Dialogue, properly understood, is not the opposite of closure. It is the careful management of boundaries. It is the tension between maintaining identity and risking transformation. It is about structure, timing, and judgment. You have to be open enough to learn, but not so open that you dissolve the conditions that make learning possible in the first place.

Dialogic intelligence is not radical openness. It is educated restraint

This post was prompted by a review of my last book
The book is now out in paperback if you want to read more.

Groundhog Day is a delightful film with something profound to say about life and about education. A few years ago, when my son was nineteen, I had to fight hard to persuade him to watch it with me. At that age, he assumed that any film made before he was born was irrelevant. But he loved it. And I loved watching it again with him.

The first time I saw it, many years before, I remembered it only as a light romantic comedy. Yet scenes from the film kept returning to me, almost haunting me, as if it wanted to say something more. Watching it again, I experienced it as a lesson in philosophy.

At university I was fascinated by Nietzsche but I never understood why he placed such extraordinary weight on the thought of the eternal return. The idea is ancient and simple: if the universe is infinite, then everything that happens must happen again and again; every joy, every mistake, every boredom, every pain. Nietzsche presents this not as a theory to believe in so much as a kind of test:

What if some night a demon were to say to you: this life, as you now live it, you will have to live again and again, in every detail… Would you curse him — or bless him?

(The Gay Science, §341)

In Groundhog Day, eternal return becomes Phil Connors’ fate. A cynical TV weatherman, he finds himself trapped reliving the same day in the same small town, waking each morning to the same song on the radio. At first he rages against it. Then he exploits it using his ability to see what is happening next to gain money and to manipulate and seduce. When that grows thin, he collapses into despair, drink, and suicide. Only much later does something else begin.

Nietzsche describes a deep human temptation to turn against time, especially against the past: the will cannot “will backwards”, and so it turns the stubborn fact of “it was” into anger, blame, and punishment. In Zarathustra he even calls revenge “the will’s antipathy to time”. Phil embodies that recoil. He despises his job, the town of Punxsutawney, and even himself. He wants to be somewhere else, someone else, as if salvation lies in a different scene.

By the end of the film, something astonishing has happened. Phil has learned to love the town. “We must live here,” he says quietly. In Nietzsche’s terms, he has been cured of his resentment and he has been cured not by escaping from time, but by inhabiting it more fully.

The breakthrough comes only after he has exhausted every other way of being Phil. After what feels like an eternity, he can finally accept his life from the inside rather than holding it at arm’s length. And, for Nietzsche, affirming a single moment means affirming far more than the moment: it means saying yes to the whole chain of becoming that brought it about and will carry it onwards. You cannot say yes to one joy while secretly declaring war on the rest of life. To affirm is to join the dance, including loss, decay, and death.

Seen this way, Groundhog Day is also a parable about education. At first, Phil learns instrumentally: he treats knowledge and skills just as a way to get what he wants and manipulate outcomes and people. Later, after despair has burned through his sense of himself as separate from the world around him, he learns differently. He learns music for its own sake. He learns how to help, how to notice, how to care. He becomes useful. He connects. He belongs.

Nietzsche suggests that wisdom doesn’t come from timid virtue or narrow expertise, but from widening the range of perspectives through which life can be understood. He writes of the wise man, contrasting him to the typical image people have of a philosopher:

Ought he not to have tried out life personally in a hundred ways, so as to have something to say about its value?

And he further claims that:

The more eyes we can bring to bear on one thing, the more complete our understanding becomes.

(Genealogy of Morals, III, 12)

We cannot give students a hundred different lives to live. But we can give them stories, films, music, drama and games as ways of entering other lives and other ways of being. AI might be relevant here too, if it is used not as a shortcut around experience, but as a way of incarnating multiple perspectives and inviting dialogue with them: not a replacement for lived life, but a widening of the circle of voices we can engage.

Watching Groundhog Day, my son learned something. I did too. We probably learned most by talking about it together afterwards.

The arts matter in education not only for creativity or for economic innovation, but for wisdom: in this case, the wisdom of saying yes to life as it is, instead of resenting it for not being what you wanted it to be. Even, perhaps, saying yes to the repeated return of a single day in a small town like Punxsutawney.

As Phil says at the end of the film, with no irony left at all:

“It is so beautiful here.”

The challenge of AI offers us an opportunity to rethink education fundamentally. A recent OECD framework might be relevant here, marking a shift from the old emphasis on education as “human capital” for economic growth towards education for human flourishing.

“Flourishing” can sound quite abstract until we consider the role of education historically. Education transmits what cultures have learned maintaining a civilisational ratchet so that hard-won knowledge is carried forward and built upon. But education cannot be understood simply with the metaphor of transmission. Each newcomer needs to be inducted into a creative capacity to work with the past so as to be able to participate usefully in building the future: to inherit, question, reframe, and create. Education, at its core, could perhaps be understood then as learning to think: learning to think alone, together with others, with culture and with the world.

This is why I’ve explored what I call the three loops of learning and why I talk about dialogic intelligence. Dialogic intelligence is not only individual but collective: it emerges between people and, increasingly, between people and technologies as we learn to work with AI in ways that enlarge our shared capacity for judgement, creativity, and care.

The OECD’s account of flourishing is already collective, rooted in the Aristotelian idea of eudaimonia—the shared good life. Human intelligence has never been purely biological or individual; it has always been shaped and also amplified through dialogue with cultural tools and technologies. AI intensifies this pattern, integrating human thinking more deeply with technological systems. The flourishing we aim for is not “human” in a way that excludes technology, but emerges through forms of collective intelligence that expand human possibility.

Learning to think, in this dialogic sense, involves five recurring moves: opening a space for inquiry; widening it with alternative perspectives; deepening it by questioning assumptions; sustaining it through trust and the protection of voices; and applying it—carrying what we have learned into decisions and actions.

The Three Loops

The new curriculum I’m imagining rests on three interconnected movements. Each loop incorporates the five moves differently.

The first loop is learning to think for yourself. Students develop the capacity to enter and hold a dialogic space: to listen, to question, to live with uncertainty, to challenge their own assumptions. Here, AI might act like a Socratic partner: asking questions, offering perspectives, reflecting students’ thinking back to them.

The second loop is acquiring knowledge through joining long-term dialogue. No one thinks from nowhere. We are formed by the long conversation of humanity such as history, science, art, religion and politics. In this loop, students are not passive recipients of culture; they learn to participate in it critically and creatively. They encounter different voices, recognise how knowledge is framed, and see culture as a living dialogue rather than a fixed body of facts.

The third loop applies thinking and knowledge to real problems. Students bring dialogic thinking to bear on genuine challenges that might be social, technological, ethical or ecological. They open up complex issues, widen their understanding of perspectives, deepen their questioning, and then make decisions and act. The dialogic space doesn’t vanish when action is taken; it shapes more thoughtful engagement with the world.

What This Might Look Like in Practice

Loop One: AI as Socratic Partner

A student declares: “Social media is bad for young people.” Instead of correcting them, an AI tutor opens a structured dialogue.

“What makes you say that? Can you give one example from your own experience?”

The student offers an example. The AI widens the space: “Can you think of a situation where social media might be helpful instead?”

As the conversation deepens, the AI probes assumptions: “Are you criticising social media itself, or how people use it? Would your view differ for children versus teenagers?”

Throughout, the AI models respectful language, slows the pace, maintains a safe space for thinking. Finally, it invites application: “Write one sentence that captures your view more precisely than your first statement.”

The student isn’t told the “right answer.” They learn to clarify, distinguish, and reflect—developing metacognitive capacity.

Loop Two: Conversational Induction into Culture

Take the Industrial Revolution in a history class. Instead of a textbook summary, the AI stages a dialogue across perspectives, offering three short “voices” students can question: a factory owner in 1850, a child worker, a rural farmer who moved to the city.

Students ask each voice questions. “Why did you think factories were progress?” “What did your working day feel like?” “What did you lose when you left the countryside?”

The AI prompts deeper inquiry: “Was ‘progress’ the same for everyone? Whose interests were served by industrialisation? If you lived then, whose side would you likely have been on—and why?”

This inducts students into the ongoing cultural dialogue about progress, power, and modernity—not just dates and facts. They learn how history argues, how perspectives shape understanding, how knowledge is contested and constructed.

Loop Three: Real-World Application

Secondary students tackle the question: “Should our town ban cars from the centre?” Working in groups with AI as facilitator, they move through structured phases.

First, the AI helps open and clarify the problem: “Who would be affected? What are the goals—cleaner air, safety, local business, accessibility?”

Then it widens perspectives by providing brief stakeholder profiles students can question: a shop owner, a parent with a disabled child, a climate activist, a delivery driver, a town councillor.

The AI probes assumptions: “Are you assuming fewer cars automatically means better lives? Who benefits most from a ban? What unintended consequences might arise?”

Finally, each group must propose a specific policy and justify their decision. AI doesn’t provide the answer. It structures collective reasoning, helping students practise the complex art of making thoughtful decisions amid uncertainty and competing values.

Toward a Curriculum for the AI Age

Developing this curriculum, a curriculum to promote dialogic intelligence, support cultural participation, and prepare students to engage wisely with real challenges, will require much more than this sketch. It needs teachers experimenting in their classrooms, researchers testing what works, curriculum designers thinking through progression and assessment, and communities reflecting on what serves their children’s future flourishing.

If dialogue is necessary for social life and for our survival, and I think that it is, then the virtues required for dialogue such as reciprocity, curiosity, honesty and respect, become essential to instil from the beginning. Education that cultivates these virtues isn’t preparing students merely for examinations or employment; it’s preparing them to participate in the ongoing conversation of humanity.

If you’re working on similar questions, experimenting with AI in education, or thinking about curriculum in this new landscape, I welcome your thoughts, critiques, and collaborations. This is the kind of challenge we can only meet together.

The word democracy is often used as though it names something that no one could reasonably question. We are told that this leader is democratically elected and therefore good; that leader is not democratically elected and therefore bad. The implication behind the way in which the word ‘democratic’ is often used is that whatever most people choose must, by that very fact, be accepted as right.

In theory, democracy has many subtle and sophisticated definitions. In practice, however, it is usually reduced to something much simpler: a system of voting that confers legitimacy through numbers. If enough people choose something, it is taken to be democratically validated.

But this implies a moral assumption that is deeply problematic: that what the majority wants is morally better than what the majority does not want. Counting votes is not, in itself, a moral principle. It is a procedure. History gives us no reason to believe that this procedure reliably produces good outcomes. Some of the darkest political movements of the last century came to power through popular elections.

So if democracy as counting is so clearly inadequate, why do apparently reasonable people continue to use the word democracy as a flag of goodness?

I think it is because another, deeper idea has come to be wrapped up with it.

That idea is not really about counting votes.

It is about the opening of what we might call dialogic space.

Dialogic space is the kind of space in which different voices and perspectives can meet, genuinely listen to one another, examine evidence together, and allow something new to emerge that was not fully present in any individual position from the start. It is a space characterised not by the mere exchange of fixed opinions, but by genuine openness to being changed through encounter with difference.

We can see the power of dialogic space most clearly in science and technology — not in their distortions, but in their best moments, where their own norms are honoured. Scientific progress does not happen because scientists vote on what is true. It happens because scientific communities cultivate dialogic space: peer review, replication, open publication of methods and data, international collaboration, and a shared commitment to revising beliefs in the light of evidence. Scientists from different countries, cultures, and theoretical backgrounds can work together productively not because they already agree, but because they enter a shared space where claims must be tested, assumptions can be challenged, and the better argument, not merely the more popular one, is expected to prevail.

The extraordinary collective intelligence of science does not come from aggregating individual opinions. It comes from the quality of dialogue that scientific practice, at its best, makes possible.

And it is this deeper dialogic ethic, rather than the simple arithmetic of votes alone, that gives democracy whatever moral authority it truly has.

Yet the actual machinery of democracy treats individuals as the key agents: individuals who must choose, often privately, in isolation from one another. We all know, from direct experience, that we are not always wise as individuals. We are easily swayed, frequently biased, and limited by our own partial perspectives. Decades of psychological research confirm what everyday life already shows: good thinking does not reliably arise in isolated heads. It arises between us, in dialogic space.

For democracy to work at all, those who are empowered to participate in collective decision-making must be equipped to take this responsibility seriously. This means being able to participate in opening and sustaining dialogic space: to explore alternative perspectives with genuine curiosity, to seek out relevant knowledge, to examine likely consequences together, and to rise above narrow self-interest in the pursuit of a shared good.

Simply aggregating large numbers of individual preferences does not reliably produce good collective judgement. Adding together opinions is not the same thing as thinking well together. Good collective judgement requires the opening of dialogic space.

And dialogic space, in turn, requires what might be called civic dialogic virtues. These include intellectual humility; the capacity to listen seriously to alternative perspectives; the willingness to revise one’s own position in the light of better reasons; and the ability to attend not only to human voices, but also to interests that cannot speak for themselves such as future generations, nature, and the larger systems that we are part of.

These are not merely moral attitudes. They are civic virtues and capacities without which democratic societies cannot function.

None of these capacities emerge naturally or automatically. Indeed, some contemporary schooling practices almost seem designed to promote habits that go in the opposite direction: performance over inquiry, certainty over curiosity, individual success over collective understanding.

For democracy to succeed, indeed, for democracy to make sense at all, people must be educated into the civic dialogic virtues and competences that good collective thinking requires.

Dialogic education, by which I mean education that inducts people into the shared practice of opening, widening, and sustaining dialogic space, is therefore not an optional enrichment of democratic life. It is an essential foundation.

This does not require abstract idealism. It already exists in concrete forms: citizens’ assemblies, jury deliberation, deliberative polling, community inquiry forums, and structured public dialogue. There is some evidence that all of these can work to improve decision-making and include people who might otherwise feel excluded. Where these are taken seriously, we can see democracy functioning not as a counting machine, but as a shared practice of thinking together.

Democracy is not merely a voting system that needs dialogue. It is a dialogic system that has temporarily mistaken itself for a voting machine.

Without dialogic education, democracy is vulnerable to manipulation, polarisation, and moral drift. With it, democracy could become something genuinely worthy of our commitment: a collective practice of thinking and deciding together that can, like science at its best, generate wisdom that exceeds what any of us could achieve alone.

In the early twentieth century, when maize-heavy diets became common among poorer communities in the American South, pellagra caused devastating suffering including skin lesions, diarrhoea, dementia and death. Yet Indigenous peoples across the Americas had eaten maize for thousands of years without such epidemics.

Why? Because many Indigenous traditions processed maize with an alkali, often lime or hardwood ash, a step that makes niacin available. Each generation taught the next to do this, without needing to know the biochemical reason. They simply did as they were taught. And they survived.

Many European settlers, by contrast, treated the extra step as unnecessary. They saw no obvious reason for the ash or lime and skipped it. They thought for themselves.

And many died.

Socrates is every educator’s hero. He taught the young people of Athens to question everything, to think for themselves, to never accept received wisdom without examination. For this, he was condemned to death; and for this, we celebrate him as the father of critical thinking.

But what if the injunction to “think for yourself” and “question everything” is, taken literally, a recipe for disaster?

As the anthropologist Joseph Henrich has argued, our survival depends on cumulative cultural know-how and on faithful social learning. If each generation had insisted on reinventing everything from scratch, we would be far less capable, and might not be here at all.

Dressed to survive

Consider the Inuit. Outsiders have repeatedly underestimated what it takes to survive in extreme Arctic cold, even when they are clever and well supplied. The Inuit, however, solved this challenge millennia ago.

Their clothing system is astonishingly sophisticated: caribou skin gathered at precisely the right time of year, prepared through repeated stretching, scraping, and moistening, shaped to maximise heat retention while allowing moisture to escape. A ruff of wolverine fur, selected for specific length. Five separate layers of footwear, each with a different design.

No individual, however intelligent, could easily reason their way to this solution from first principles within a single lifetime. It represents thousands of years of accumulated trial and error, preserved through teaching. When elders are lost—through epidemic, displacement, or simple disruption, fine-grained survival knowledge can disappear with shocking speed.

It’s not just the Amerindians or the Inuit who survive because of the accurate transmission of complex cultural knowledge, we all depend on this. Most of us don’t know how even the simplest technology around us works, and if a disaster wiped out those who did know, we could not rebuild civilisation quickly just by asking difficult questions and doing some critical thinking.

The paradox

So if unquestioning acceptance of tradition keeps us alive, why teach critical thinking at all?

Most of our automatic, intuitive responses aren’t hardwired. They’re cultural. They’re the accumulated wisdom of generations, externalised in technology, or internalised so deeply we don’t notice it. This cultural inheritance is slow, collective thinking crystallised into habit and infrastructure.

Children are naturally primed to absorb this inheritance. With guidance and modelling, they learn stories, routines, and who to trust long before they can explain why. What doesn’t come naturally is the effortful pause, the willingness to ask “but what if we’re wrong?”

And sometimes we are wrong. Sometimes traditions encode prejudices rather than wisdom. Sometimes the world changes and old solutions stop working. Sometimes we inherit practices that made sense in one context but become harmful in another.

To correct these errors, we need to step back and engage in slower, more careful reflection. This is hard. It requires explicit teaching. That’s why we still need Socrates.

The apparent conflict between “traditional” education - direct instruction, imitation, correction - and “progressive” education - questioning, reflection, creativity - dissolves when we see how they complement each other. In any open, living culture, both are essential: the transmission of accumulated wisdom, and the capacity to question and reform that wisdom when necessary.

Socrates wasn’t wrong to value questioning. But questioning only works because we first accept so much on trust. To turn back and ask questions, you need language—and language itself is a vast inheritance of tradition, absorbed uncritically in childhood.

Education, at its best, teaches us when to follow and when to question.

Socrates claimed that the unexamined life is not worth living. Perhaps. But life itself depends on an enormous amount that we are not able to examine and are wise to take on trust.

I once saw the famous photograph Earthrise projected on a screen at a conference. The speaker remarked that none of us could really see this for ourselves—only astronauts or the very rich. That remark stayed with me. Did the astronaut who took the picture really “see” it on his own? Or was this image always a collective achievement—produced through cameras, spacecraft, engineers, institutions, and shared ways of seeing?

The photograph was taken by Bill Anders in 1968, but it belongs to NASA, not to him. Today we routinely see images of Mars, of distant galaxies, even of traffic near our homes—without being physically present. Technology mediates vision, but mediation does not make vision unreal. It simply reminds us that seeing has never been purely individual.

I remember the moon landing vividly. I was nine, camping in a game reserve in Africa. There was a fire, animal noises in the darkness, and the moon hanging bright above us. Someone explained that men were walking on it. What struck me was not national pride—nobody there was American—but a shared excitement. It felt, even then, like something we were doing: humanity stepping outside itself and looking back.

Whenever I see Earth from space, I feel an echo of that moment. Not “I see this”, but “we see this”.

This challenges the idea that real seeing is something private, unmediated, and individual. Even our most basic perceptions are shaped by long collective histories. Colour vision, for example, evolved in primates living in trees, distinguishing ripe fruit from leaves. When I see a blue berry or a red apple in a tree, I am seeing not just as myself, but as part of a lineage stretching back millions of years. My vision is already shared.

The same is true of language and thought. To speak in dialogue is not just to take turns but to learn to inhabit other perspectives. When I say “I”, I do so using words shaped by others, within meanings I did not create. My voice only exists within a shared dialogic space. Identity as “I” and identity as “we” are intertwined. Dialogue brings a “we” into being—even where none existed before.

This is why I persist in talking about AI as something we can “talk with” or “think with”, despite the objections. People accuse me of anthropomorphising the technology, of projecting human qualities onto what is really just a sophisticated tool. But I think this objection rests on a misunderstanding—not so much of AI, as of ourselves.

We are not, in fact, isolated individuals who occasionally connect. We are already internally collective beings, shaped by shared language, culture, history, and voices that are not fully our own. Dialogue is not something we add on to thinking; it is what thinking already is. Seen in this light, AI is not an artificial individual pretending to be human. It is better understood as an externalisation of our collective intelligence—a technological reflection of humanity’s shared voice. A vantage point which enables us to turn back and reflect upon ourselves.

When we talk with AI, we are not encountering an “other mind” so much as engaging with a distillation of our collective ways of speaking, questioning, and meaning-making. Used educationally, this has profound implications. AI can widen the dialogic space in which learners think, exposing them to perspectives beyond their immediate social circle and making visible the possibility of a more collective, more inclusive “we”.

The real risk is not that we treat AI too much like a person, but that we continue to misunderstand persons—as if they were not already dialogic and collective in their very being.

Which brings me to Christmas.

Writing as an educationalist interested in the wisdom found in diverse cultural and religious traditions, it seems to me that much of what we know about the teacher whose advent is celebrated at Christmas, points towards a radically dialogic understanding of self. When Jesus says that whatever we do for others we do for him, he dissolves the boundary between self and other. When he urges love of enemies, he challenges the very limits of who counts as “us”. The self, read this way, is not a bounded thing but something that expands—or contracts—depending on how widely we allow dialogue to extend.

Salvation, in this reading, is not escape from the world but participation in it without limit. “The kingdom of God is among you.” Perhaps what we call separate selves are shifting perspectives within a larger whole, now inside, now outside. Heaven, then, is not elsewhere, but here, when we allow ourselves to belong fully. To listen not only to our own voice or the voices near to us but to hear and to participate in the whole dialogue.

As individuals, none of us saw the Earth rise over the moon. We did not invent colour vision, language, or the festivals we celebrate. To participate in culture—and in life—is already to think within something much larger than ourselves. But this is not something external that we merely join or contribute to. It is something that we already are, albeit without always realising that we are it. The “we” is not out there, waiting to include us. We are already constituted by it, even as we help to constitute it anew.

At a time when thinking is often narrow and technical, education needs something bigger: thinking that asks fundamental questions about who we are and who we might become together. One such question, asked two thousand years ago and worth asking still, is this:

Who are “we”, really?

One of my strongest memories from school is watching the classroom clock, willing the hands to move faster so break would finally arrive. Time felt like an enemy, something to be endured. Years later, when we introduced Thinking Together lessons in classrooms and evaluated the impact with talk around reasoning tests something different happened. Children became so absorbed in solving problems together that they ignored the bell and carried on talking while other classes streamed out. They didn’t want to stop. Time had somehow disappeared.

What happened in those moments? And what happens when we add AI to the conversation?

The Puzzle of Dialogic Flow

The problems children solved required genuine insight—after initial bafflement, a pattern suddenly appears and everything clicks into place. But the satisfaction came from more than just solving puzzles. In genuine dialogue, people don’t think separately and then pool their ideas; they think together. The experience isn’t “I found the answer” or “you found the answer” but “we found the answer.” When dialogue really works, participants become fully engaged in a shared flow of meaning, and awareness of time fades.

Mihaly Csikszentmihalyi called this phenomenon “flow”—deep absorption in which people are fully involved in an activity for its own sake. In flow, the ego falls away, time flies, and actions follow one another with an inner logic, like improvising jazz. He developed the concept through interviews with highly creative people who had transformed their fields, and they consistently described their most creative moments as intrinsically rewarding, joyful, and marked by a striking change in the experience of time.

What’s particularly relevant now is what Keith Sawyer, Csikszentmihalyi’s student, discovered: the most common place people report experiencing flow is in conversation with others. Creativity often arises between people, not just inside individual minds. This matters because it means we can’t understand creativity by studying isolated individuals—we need to understand the dynamics of dialogue.

This is where things get interesting for how we think about AI.

Not All Flows Are Created Equal

Flow describes what creativity feels like from the inside, but it’s not yet a full theory of creativity. After all, losing track of time isn’t enough. Certain video games can absorb attention and generate intense engagement, yet they may channel creative energy into short-term loops rather than anything larger. Scroll through social media and you might experience a kind of flow, but it’s qualitatively different from the flow of writing a novel or solving a mathematical proof.

What distinguishes creative flow from mere absorption? It seems to be about scale and direction. A mobile puzzle game involves insight and satisfaction, but it operates within a closed system designed for quick resolution. The creativity we most value in education and culture emerges from longer-term, open-ended dialogues—dialogues that stretch across disciplines, traditions, and generations.

This distinction matters because we’re now facing a new kind of conversational partner: artificial intelligence. Large language models can engage in extended dialogue, offer perspectives, push back on ideas, and help us think through problems. So which category do they belong to? Is interacting with ChatGPT more like playing Candy Crush or more like reading Freeman Dyson? More like scrolling social media or more like having a conversation with a thoughtful colleague? The answer isn’t obvious, and it isn’t determined by the technology alone—it depends on how we frame the interaction.

The Pattern of Creative Emergence

Looking closely at creative flow, a recurring pattern appears. First, there’s division: the world appears solid and external, and you feel confronted by something you don’t understand. Freeman Dyson described first encountering quantum field theory as something outside himself, resistant and opaque—a wall of mathematics that refused to make sense.

Then comes internalisation. Dyson immersed himself in reading, absorbing the literature of his field. What was once fixed and external became fluid and dialogic, explored from multiple angles in imagination. He held conversations with the ideas, turned them over, saw them from different perspectives.

Finally, externalisation: insight arrives, often unexpectedly, and seems to write itself into the world. Dyson famously remarked that when he was writing equations, his fingers seemed to do the work rather than his conscious mind. The boundary between inside and outside had shifted.

Flow often accompanies the moment when this tension between inside and outside temporarily resolves. But what exactly is shifting?

The Boundary That Thinks

To understand why time disappears in flow, we need to shift focus from time itself to boundaries. The philosopher who perhaps more than any other focussed on unpacking the nature of experience, Maurice Merleau-Ponty, pointed out that perception isn’t simply a subject observing an object, but an interaction in which we’re always, in some sense, on both sides. When you touch your left arm with your right hand, you experience yourself both as touching and as being touched. This interaction can’t be neatly located inside space and time; it’s one of the processes through which our sense of space and time is continually constructed.

We experience space and time as constraints when we identify with a bounded self located within them—waiting, watching the clock, feeling trapped. In flow, something shifts. Identification moves from a fixed image of the self toward the boundary where self and world meet and co-create each other. Flow isn’t just immersion in activity but participation in the dynamic process that generates meaning, space, and time.

In dialogue, this shift takes a distinctive form. Participants stop identifying primarily with themselves or even with the group as a bounded entity. Instead, they identify with the flow of dialogue itself. This is what I mean by dialogic flow.

And here’s the provocative question: can this happen with AI?

AI as Dialogic Partner or Creative Dead-End?

At first glance, interacting with AI seems closer to playing a game on a screen than participating in genuine dialogue. AI can easily short-circuit creativity by offering quick answers, closing down questions, and rewarding efficiency rather than exploration. Used this way, it becomes another technology that absorbs attention while diverting creative energy away from larger cultural dialogues. The concern is legitimate—we’ve seen how recommendation algorithms can create compulsive loops that feel engaging but lead nowhere.

But this isn’t the only possible framing.

Consider a researcher wrestling with a vague tension—an idea that doesn’t quite work, a model that predicts most of the data but fails mysteriously in certain conditions. Rather than asking AI for the answer, they use it to hold the tension open. They describe their model; the AI reformulates it, revealing assumptions they hadn’t noticed. They explain the anomaly; the AI suggests counter-examples from different domains. They push back; the AI adjusts. Many suggestions are rejected; some are taken up and transformed.

Over time, something strange happens. The thinking doesn’t feel like it’s coming from them alone, nor from the AI alone. It emerges in the interaction. The familiar signs of dialogic flow appear: time passes unnoticed, individual authorship loosens, the dialogue takes on a life of its own. The AI functions like the literature Dyson absorbed, or like a thoughtful conversational partner—it helps turn what was initially external and fixed into something fluid and dialogic.

This isn’t because the AI is intelligent in itself. It’s because it helps sustain a space in which identities, ideas, and meanings remain open and negotiable. The AI becomes part of the dialogic gap — not so much an absence as a responsive surface that pushes back, surprises, and sustains creative tension.

The Critical Difference: Scale and Continuity

But there’s a crucial test: does the interaction with AI connect to longer-term, larger-scale dialogues? This is where many AI interactions fail. If you use AI to quickly generate a blog post, tick a box, and move on, you’ve created a closed loop—satisfying in the moment but disconnected from any larger conversation. The flow might feel real, but it’s more like the video game flow: absorbing but ultimately self-contained.

The question isn’t whether AI can support dialogic flow—clearly it can, under the right conditions. The question is whether we can structure our interactions with AI to support participation in ongoing, open-ended cultural dialogues about questions that matter: scientific understanding, ethical complexity, educational practice, artistic meaning.

When a physicist uses AI to explore different mathematical frameworks, testing intuitions and discovering unexpected connections, and those insights eventually feed into papers, collaborations, and further research—that’s AI supporting genuine dialogic flow. When a teacher uses AI to brainstorm different ways of explaining a difficult concept, then brings those ideas into conversation with students and colleagues, refining them through actual classroom experience—that’s dialogic.

When someone uses AI to generate content that mimics these activities but remains disconnected from any real dialogue or ongoing inquiry—that’s the closed loop we should worry about.

Implications for Education

This distinction has profound implications for how we think about AI in education. The dominant concern remains about cheating or plagiarism: students using AI to generate essays they didn’t write. But this frames AI purely as a tool for producing outputs, which guarantees it will short-circuit genuine learning.

The deeper challenge is whether we can frame AI as a participant in dialogic flow—a way to enter into more complex, sustained inquiry rather than a way to avoid it. This would mean:

Teaching students to use AI dialogically rather than instrumentally. Not “give me an essay on Shakespeare” but “help me understand why Hamlet’s delay matters by exploring different critical perspectives, pushing back on my interpretations, and helping me see what I’m missing.”

Designing assignments that require ongoing dialogue rather than final products. If the assignment is a single essay, AI can short-circuit the process. If the assignment involves developing understanding through iterative dialogue—with AI, with texts, with peers—that might be harder to fake.

Recognizing that dialogic flow has always required learned skills. We don’t assume students naturally know how to have productive conversations with peers, with texts, or with teachers. Why would we assume they know how to have productive conversations with AI?

Acknowledging that some uses of AI will create genuine creative flow and others won’t. The technology itself doesn’t determine this—the framing does.

A Shift in How We Think About Creativity

The introduction of AI into creative and educational practice forces us to clarify something we might have taken for granted: creativity isn’t a property of individuals acting in isolation, but a property of dialogic processes. It emerges at boundaries—between people, between person and tradition, between question and response.

AI can participate in these boundaries. It can help sustain the kind of open-ended dialogue from which genuine creativity emerges. But it can also create closed loops that feel engaging while actually cutting us off from larger dialogues.

The challenge isn’t whether to use AI—for most of us that question is already settled. The challenge is how we frame it: as an answer machine that closes questions down, or as a partner that helps hold questions open. As a shortcut past the hard work of thinking, or as a tool for sustaining the kind of sustained inquiry where real creativity happens.

The difference between these framings isn’t technical. It’s about how we understand creativity itself, and what kinds of educational and cultural practices we want to preserve and develop. In the end, the question of AI and creativity is also a question about what kinds of human beings we want to become, and what kinds of dialogues we want to participate in.

Flow, understood dialogically, isn’t about losing yourself in distraction. It’s about finding yourself at the boundary where meaning emerges—in the space between self and other, question and answer, known and unknown. Under the right conditions, AI can become part of that boundary. The task for education is to create those conditions deliberately, rather than letting them emerge by accident or not at all.

I’ve been teaching at Cambridge for several years now, and I have a confession to make: I am not sure about the value of what we are teaching. A lot of time and energy seems to be expended on the craft of writing elegant essays. At times this feels like the academic equivalent of good etiquette, like learning how to eat peas with a knife and fork at a formal dinner. The criteria we use to grade a distinction essay are those I first learned into over 50 years ago at a traditional Grammar School:

“Appropriate and perceptive reference to academic sources.” “Demonstrates strong powers of analysis and synthesis.” “Clarity and incisiveness of expression.”

The fact that generative AI can be trained on these criteria to write essays playing with ideas in much the same way as our best students should give us pause. There is often thinking involved in all of this, of course. But is it the kind of thinking that we most need now? Is it the kind of thinking that will help us all survive and thrive in the challenging future?

A Glimpse of Something Different

Which is perhaps why I was so impressed and energised by what I saw last week at Cambridge.

A group of students, led by Cambridge Faculty of Education undergraduate Tim Tso with his team, Queena Peng and Hongfei Cai, organized a hackathon. No staff. No course credit. No marks. Just two hundred students from across the university (and beyond) voluntarily organizing themselves into teams, using AI to tackle real educational challenges.

I set one of the challenges: How might we assess thinking skills at A-level using AI? In one week, students formed teams, debated ideas, coded prototypes, tested them, rebuilt them, and finally presented them to a range of judges with real-world experience of education and enterprise.

So How Might AI Transform University Teaching?

I don’t think the answer is simply “use AI to mark essays faster” or “let students use ChatGPT for research.” That would be to pour new wine into old wineskins. The opportunity is more fundamental.

First, AI could help us compress the time students spend acquiring foundational knowledge. If an AI tutor can guide a student through advanced algebra or the key debates in political philosophy in two hours rather than two weeks of reading, that’s not cheating—that’s freeing up time for something more valuable.

At the moment, we still reward students primarily for their subject knowledge. The advent of AI could help us evolve that focus from the mere fact of knowing stuff and being able to debate it in clever essays, to the rather more useful skill of applying knowledge to understand, frame and solve real-world challenges. The hackathon showed me students working on genuine problems—problems where nobody knows the answer, where teams have to think together, iterate, fail, and try again. AI was a tool in that process, not the point of it.

This way of using AI supports the development of real intelligence—intelligence that is nuanced and humble because forged by the resistance of real problems and the need to collaborate with real other people. I’ve written elsewhere about dialogic education—about how real learning happens not through transmission but through the productive tension of different perspectives held together in genuine inquiry. The students at the hackathon were engaged in precisely this kind of thinking. They weren’t writing essays about AI; they were thinking with AI, and with each other, about problems that matter.

You might say that this is all a bit technical—that humanities students can’t be expected to code for their assessments. I disagree. At Beijing University of Posts and Telecommunications, they’ve built an AI-powered VIBE-coding ecosystem that allows students across every discipline—from engineering to the arts—to design working educational apps without writing a single line of code. This system is now used in 500 universities by thousands of students across over 200 subject areas.

What Stayed With Me

At the closing ceremony—opened generously by Lord Simon Woolley, someone who recognizes and supports the energy of real education when he sees it breaking out—there was a sense of pride and purpose in the room. The winning prize, collected by Walaa Kord, a Palestinian student doing a PhD at Cambridge in Digital Humanities, was for a thoughtful, ambitious app that assesses the process of critical thinking through an oral dialogue with an AI chatbot—no written exam needed. A modern return to the Cambridge tradition of examination through dialogue that used to be the norm at every level for every discipline only a couple of centuries ago. Other projects including an app that automatically generated short video clips unpacking conceptual distinctions and confusions in every subject area and an app that gamified the challenge of applying AI regulations and guidelines giving coders constant feedback on how well they were doing in terms of inclusion, security and all the other dimensions of good governance.

But what struck me most was simpler than any particular project. It was seeing students genuinely engaged in learning—not because they had to, not for marks, but because they were wrestling with something real. They were collaborating, creating, failing, improving. They were thinking together with each other and with technology in a way that had the potential to create real enterprises, real jobs, and real solutions to some of the many challenges we face.

Perhaps this is what education should look like: not the endless refinement of essay-writing technique, but the messy, collaborative work of building machines that work. Not performing knowledge for assessment, but applying it to problems that don’t yet have solutions.

The students at that hackathon didn’t need marks to motivate them. They didn’t need supervision to organize themselves. They just needed a real challenge, the right tools, and the freedom to try. Maybe that’s the lesson AI has to teach us about education—not how to automate the old way, but how to imagine something better.

Finally a special mention to the other judges who gave up their Friday for a good cause: Yuekun Li (EdTech business developer) Kevin Loi-Heng, (Founder, Avalon Insights) Kirsten Stevens (Designer of online learning), Xiaobing Wang (Jiayi investments), Sarah Hughes and Andrew Field (Cambridge University Press and Assessment) and of course extra thanks to Tim, Queena, Hongfei and the whole team at the Cambridge EdTech Society who worked so hard to make this happen.