Abstract
As generative artificial intelligence (AI) increasingly provides knowledge and outperforms students on conventional assessments, education faces a fundamental challenge. This paper argues that the appropriate response is to rethink what we mean by intelligence and reposition dialogic intelligence as a central educational aim. Dominant conceptions of intelligence as an individual trait, measured through rapid problem-solving, are historically tied to print literacy’s cognitive affordances and obscure the relational conditions under which meaning emerges. The rise of large language models creates an opportunity to recover an older conception of intelligence as the capacity to sustain responsive relationships through which truth and value emerge. Dialogic intelligence arises in the relational space between people, between people and nature, and between people and technology. It involves dwelling in open inquiry, allowing new insights to emerge, and critically testing and refining them. While AI can support knowledge generation and verification, dialogic intelligence depends on uniquely human capacities for care, ethical responsibility, and sustained attention. The paper outlines pedagogical approaches and assessment frameworks foregrounding relational processes rather than individual performance alone.
The rapid development of generative AI has unsettled long-standing assumptions about the purposes and practices of education. Generative AI systems based upon Large Language Models (LLMs) now equal or outperform students on many conventional assessments, including essays, problem-solving tasks, and even some professional qualifications (Liu et al., 2024; Newton & Xiromeriti, 2024). As AI improves, reports of different systems performing better than most students on high-stakes examinations are increasing. Educators and policymakers have responded with tactical measures: banning AI tools, redesigning assessments, or devising “cheat-proof” examinations (An et al., 2025; Curtis, 2025). Yet such measures miss the deeper challenge. If AI can perform the tasks our education systems are designed to measure, then training students to replicate these tasks risks leaving them without meaningful roles in the future.
This paper focuses specifically on Generative AI based on the use of Large Language Models (LLMs), which is the technology underlying conversational AI systems now widely used in education such as ChatGPT, Gemini, DeepSeek, Grok, and Claude. Research on the cognitive impact of this form of AI falls into three broad streams. The first warns that AI use inhibits thinking, producing “metacognitive laziness” as users offload cognitive effort (Gerlich, 2025; Lee et al., 2025). The second, hybrid intelligence, proposes that AI can amplify human capabilities when the two are appropriately integrated (Cukurova et al., 2019; Järvelä et al., 2023). The third suggests that human–AI interaction may generate genuinely novel cognitive capacities that neither human nor machine possesses alone. The present proposal aligns primarily with this third stream while incorporating elements of the second: dialogic intelligence can emerge through the relation between human and AI, but this emergence depends on humans cultivating distinctively human capacities such as care, questioning, and ethical responsibility.
Empirical findings support this differentiated view. Some studies find that students who rely on AI show declines in critical thinking and independent problem-solving (Gerlich, 2025; Lee et al., 2025). Others demonstrate that AI can enhance creativity, reasoning, and metacognitive engagement when framed as a partner in inquiry (Habib, 2024; Iqbal et al., 2025; Rana et al., 2025; Singh et al., 2025; Wang et al., 2025; Yang & Xia, 2023). What matters is not the use of AI as such but how AI is being used. When AI is deployed as a shortcut to answers, it undermines students’ intellectual growth. When it is engaged with more dialogically and used as a prompt, challenger, or co-inquirer, it can deepen reflection and stimulate new forms of understanding (Beale, 2025; Buckingham Shum, 2023).
These findings suggest a need to shift the focus of education away from getting students to perform fixed tasks with predetermined correct answers towards teaching students how to work together with AI to explore genuine questions. In this paper, I argue that this shift requires not merely new practices but also new theory. We need to change the way we think about education from a monologic paradigm to a dialogic paradigm.
Monologic is the idea that there is one true correct way of understanding everything: it assumes a unified voice or perspective and can be summed up by the initial claim of classical logic that “a thing is what it is and not another thing”. While this principle seems useful and intuitive in many local contexts, it fails to capture the bigger picture of how meaning arises. Dialogic, from the Greek for “through” or “across” logos, implies that meaning emerges only in the play of multiple perspectives. I focus on intelligence because understanding what thinking is, where it happens, and how it happens is one area where the significance of this paradigm shift can be seen most clearly.
Building on dialogic theory (Bakhtin, 1986; Buber, 1970; Freire, 1996; Voloshinov 1986; the later Heidegger, 1968; and the later Merleau-Ponty, 1968), this paper proposes a conception of dialogic intelligence as the capacity to open, widen, deepen, integrate, and maintain shared meaning-spaces through contingent responsiveness, generative questioning, and sensitivity to the perspectives of others—including non-human others and technological agents.
The paper proceeds as follows. First, I trace the history of intelligence as a cultural-technological construct, showing how print literacy shaped the dominant monologic conception and why LLMs now challenge it. Second, I develop the concept of dialogic space, drawing on Buber, Bakhtin, and Mead to articulate how thinking emerges through responsive relationship. Third, I consider how AI, specifically Generative AI based on Large Language Models, can participate in dialogic intelligence when engaged appropriately, while also attending to the risks of uncritical engagement. Finally, I outline approaches to teaching and assessing dialogic intelligence in educational settings.
A Brief History of Intelligence
Intelligence as Discernment and Judgement
The word intelligence derives from the Latin intelligentia, rooted in intelligere (“to understand, discern”), itself combining inter (“between”) and legere (“to choose, pick, read”). Originally, the term did not refer to a fixed cognitive trait but to a relational capacity: the ability to discern meaning, make judgements, and act appropriately in context. For Cicero, intelligentia was closely tied to sapientia (wisdom), denoting the discernment of moral duty and the prudence of knowing when to speak and when to remain silent (Cicero, 44BCE, 1913).
The Indo-European root word for intelligence, leg-, also connects to the ancient Greek word logos, which for Plato and Aristotle encompassed reason, speech, and the patterned order of reality. Socrates, in the Phaedrus, contrasted the “living word” of dialogue, which is responsive, generative, and open to questioning, with the “dead word” of writing, which is fixed and unable to answer back (Plato, ca. 370 BCE, 1997, Phaedrus, 275d). Intelligence, in this foundational view, is dialogic: the capacity to participate in the ongoing conversation through which truth emerges.
However, even within Plato’s writings, we can trace the beginnings of what would become the monologic reduction. Dmitri Nikulin’s analysis shows how Plato’s formalization of the dialectical method in works such as the Republic (Plato, ca. 380 BCE, 1992, Republic, Bks. VI–VII) began to treat dialogue as a procedure for testing predetermined logical relations rather than as the creative space from which genuine insights emerge. Nikulin (2010) emphasizes that real dialogues are uncertain and full of possibilities, precisely because of the live play of different voices. Plato’s redescription of dialogue as dialectical method was made possible by the technology of writing, which allowed conversations to be transcribed and then retrospectively analysed as a single logical journey towards a final conclusion.
The Narrowing of Reason and Psychometric Enclosure
Despite the limitations of Plato’s formal account of dialectic, the association of reason with dialogue persisted in ancient Greece and Rome, continued through the curricula of medieval universities, and was carried into the Renaissance (Pedersen, 2009). It was not until the seventeenth century that this dialogic-rhetorical conception of reason gave way to what Toulmin (1990), drawing on Spinoza, calls the “geometrical method”. Euclid’s Elements, among the first mathematical works printed and subsequently appearing in over a thousand editions, became the paradigmatic model for certain knowledge.
This transformation was not merely intellectual but technological. Print culture systematically privileged particular cognitive habits such as linear reasoning, abstract categorization, and decontextualized analysis, over the contextual, responsive, and collaborative forms of intelligence that oral cultures had cultivated (Goody, 1977; Olson, 1994; Ong, 1982). The printing press did not simply spread existing knowledge; it reconstructed what counted as legitimate knowing. Meaning, truth, and insight, which had previously been understood as emerging through relationship, came to be located within individual minds processing internal representations of an external world. The relational space between perspectives, where what I term dialogic intelligence occurs, was obscured and forgotten behind ways of thinking about thinking that emphasized what happens inside individual subjects.
The decisive enclosure of intelligence within individual minds occurred with the development of psychometric testing. Francis Galton’s pioneering statistical approaches to mental measurement reflected the broader scientific ambition to quantify human phenomena according to natural science methodologies. Alfred Binet and Théodore Simon’s (1973) approach, designed to identify children needing educational support, established a crucial precedent: intelligence became whatever formal schooling required and whatever standardized tests could measure. The tasks Binet and Simon selected, including vocabulary, reading comprehension, arithmetic, and abstract analogies, reflected particular cultural practices rather than neutral indicators of cognitive ability. These tasks aligned closely with the skills cultivated by print literacy culture.
Spearman’s (1904) statistical identification of a “general factor” (g) underlying diverse cognitive tasks provided apparent scientific legitimacy for this construction. The mathematical elegance of factor analysis obscured its ontological assumptions: that intelligence consists of measurable individual traits rather than relational capacities that emerge through interaction with particular cultural-technological environments. The strong correlation between intelligence quotient (IQ) scores and academic achievement (around .83) reflects not the measurement of pure cognitive ability but the assessment of minds shaped by similar institutional processes designed around print literacy’s cognitive demands (Kaufman et al., 2012). This historical shift helps explain why LLMs feel so disruptive: they demonstrate that the cognitive operations privileged by print culture can be performed by non-biological systems, forcing us to ask what, if anything, remains distinctively human about intelligence.
Expansions and Their Limits
Cross-cultural research revealed the historical contingency of measured intelligence. Vygotsky and Luria’s research in Uzbekistan found that individuals from oral cultures often struggled with the abstract classification tasks that formal education privileges, not due to cognitive deficiency but because their intelligence had developed through different cultural-technological practices emphasizing contextual reasoning and practical problem-solving (Cole, 1996; Cole & Scribner, 1974; Flynn, 2007). The Flynn effect, a term used to refer to the substantial IQ gains across populations throughout the twentieth century, provides strong evidence that measured intelligence is profoundly shaped by cultural transformation.
Gardner’s (2011) theory of multiple intelligences asserted at least eight different kinds of intelligence, from linguistic and logical-mathematical to bodily-kinaesthetic and musical. Sternberg’s (2021) triarchic theory distinguished analytical, creative, and practical intelligences. The influential concept of emotional intelligence (Goleman, 1995; Mayer et al., 2000) highlighted socio-emotional competencies. These models enriched public and educational discourse, broadening what counts as intelligence. However, from a dialogic perspective, these frameworks continue to treat intelligence as an individual trait.
The proposal to foreground dialogic intelligence in education does not claim to replace well-established psychometric traditions, which have demonstrated predictive validity for specific educational and occupational outcomes. Rather, it offers a complementary framework that foregrounds aspects of intelligent behaviour, such as emergence, relationality, and responsiveness, that individual-focused measures necessarily bracket. Dialogic intelligence does not replace these frameworks but operates at a different level of analysis: where multiple intelligences expand what counts as intelligence while retaining the individual as the unit of analysis, dialogic intelligence shifts the unit of analysis itself from the individual mind to the relational space between minds.
Towards Collective and Dialogic Recognition
Lauren Resnick’s research pointed beyond approaches to intelligence in education that focus only on the individual. Her characterization of “higher-order thinking” as non-algorithmic, complex, open-ended, judgement-based, multi-criterial, uncertain, and meaning-making (Resnick, 1987) described processes found in group thinking as well as individual thinking. Her later development of “Accountable Talk” explicitly positioned effective reasoning as emerging through structured dialogue rather than individual cogitation (Resnick et al., 2015).
Research on collective intelligence has provided additional evidence for intelligence as an emergent social phenomenon. Woolley and colleagues’ identification of a “collective intelligence factor” (c) that predicts group performance beyond individual member capabilities suggests that groups can exhibit genuine intelligence irreducible to individual cognitive properties (Woolley et al., 2010). The UK “Thinking Together” programme demonstrated that when students learn explicit norms for collaborative reasoning, both group and individual performance on standardized reasoning tests improve significantly (Mercer & Littleton, 2007; Wegerif et al., 1999).
Initially, the Thinking Together project framed productive classroom talk mainly in terms of explicit reasoning: making thinking visible through justifications, challenges, and co-construction of shared solutions. Over time, however, evidence from video analysis suggested that this emphasis on explicit verbal reasoning did not fully capture how breakthroughs actually emerged. Long pauses, shared silences, playful exchanges, and moments of joint attention often proved more significant than explicit logical argument. The teaching of what was called Exploratory Talk, a type of talk characterized by asking questions, giving reasons, and seeking agreement, correlated with better conceptual understanding and with improved reasoning test performance, but the process of thinking itself was not always verbal. Many times, the event of a student solving a problem was observed physiologically before it was expressed verbally: a student’s eyes would light up, their head would nod forward, they would say something like “Ah” or “Got it”, and then they would try to explain to the others. At this point, once a solution had been seen, explicit verbal expression and reasoning became important, but they were not always relevant to solving the problem itself.
This led to a shift in interpretation. The role of talk was clearly important because when exploratory ways of talking were taught, groups solved more problems. But when it came to directly solving problems with insight, the role of talk was indirect. Asking a question or offering a challenge did not directly contribute to the solution of a puzzle, but it did contribute indirectly by opening up what came to be called a dialogic space. It was the dialogic space itself that led to the emergence of new meaning (Wegerif, 2008).
Expanding Dialogic Space
Buber points out that dialogue is not simply defined by an exchange of signs but is more fundamentally defined by the way in which we attend to an “other” with whom we are in dialogue. His influential book I and Thou (1958/1970) distinguishes two orientations: the “I-It”, in which all other beings are located and ordered within the framework of the world, and the “I-Thou”, in which the encounter discloses an inexhaustible dimension that resists objectification. This shift of orientation extends beyond human interlocutors to include the way we can stand before a tree, an animal, or indeed any presence that addresses us. In the I-Thou relation, the finite encounter opens onto an infinite horizon, a depth of relationality that cannot be contained within the categories of representation but gestures to a more originary ground of meaning. Buber refers to this relational field as das Zwischen, which translates as “the Between”, a space that is reducible to neither subject nor object but arises in their encounter; this notion resonates closely with what, following research on how groups solved problems in classrooms, became referred to as dialogic space, the opening in which meaning emerges through the interplay of voices.
To illustrate: consider a classroom where students are discussing a moral dilemma. In an I-It mode, each student might simply state their position and defend it. In an I-Thou mode, students genuinely attend to each other’s perspectives, allowing themselves to be affected. The space between them becomes generative not because anyone has the “right answer” but because their mutual attention creates conditions for new understanding to emerge. This same dynamic can occur between a student and an AI system, provided the student approaches the AI not merely as an information source but as a voice capable of offering perspectives worth genuine consideration.
Dialogic space is not merely a metaphor for collaborative talk but a hypothesis about the nature of thought, or, more specifically, about the nature of what has been called “higher-order thinking” and is here called dialogic intelligence. When learners enter dialogic space, ideas cease to belong to isolated egos or bounded groups and instead resonate within a shared field of possibility. This space cannot be adequately understood within the framework of classical substance ontology, which treats things as self-identical entities defined by fixed properties. In other words, it is not simply a space “between people”; it is not located physically in any simple sense because it is an opening within the monologic idea of an external fixed world of measurable space and time. Normally we create a coherent surface of reality and locate ourselves within it, but the tension of a real dialogue disrupts that smooth surface and opens up a gap which connects us to the underlying space of possibilities. This space of possibilities, referred to by Merleau-Ponty (1968) as the pre-thematic or “raw being”, is ontologically prior to the world of space and time that we generate and imagine we experience as we go about pursuing everyday ends.
From this dialogic ontological perspective, thinking does not occur inside bounded entities but emerges in the relational spaces between them. Meaning arises not through representation but through responsive encounter. Intelligence manifests not as symbol manipulation but as the capacity to allow dialogic spaces to form and so to participate in the ongoing dialogue through which reality continually discloses itself.
Dialogic Space and Thinking
Heidegger reminded us in What is Called Thinking? (1968) that the following of algorithmic or scientific procedures does not in itself amount to thinking. Thinking, Heidegger claims, is always a response to a call: the call of that which is most thought-provoking. It is about being called to ask questions and then dwelling in the uncertainty and openness of what he called the clearing, a concept which could be appropriately paraphrased as the “space of the question”. Dialogic space opens whenever there is a challenge, the tension between different “incommensurate” perspectives, meaning differences that cannot be reduced to a shared single frame, and is about the capacity to dwell hopefully in the uncertainty of that space if and until a new insight emerges. From this source I derive the insight that dialogic space is a relational clearing in which questioning, responsibility, and the possibility of thought are sustained.
Dialogic Doubleness and the Emergence of Insight
For Bakhtin, double-voicedness is when language is inhabited by two voices, one’s own and another’s, creating tension, resonance, or irony. It is a key example of his broader claim that discourse is always dialogic rather than monologic because there is always more than one perspective in play even in a single word or utterance. Entering into dialogue always involves doubleness of this sort. On the one hand, when we speak, we identify with our own point of view; we put forward an argument or claim from our perspective on the world. But even as we do so, our words are informed by awareness of the other’s point of view, or at least what we imagine this point of view might be. It is this becoming double that is the source of thought understood as a spark of insight across difference. For that spark to happen, you have to invest in understanding the other point of view sufficiently to find yourself, at one and the same time, on both sides of the issue in a state of tension that seeks resolution, and if you are able to stay with the tension, that resolution often takes a creative new form.
George Herbert Mead pointed out that reflective thought requires internalizing the “generalized other” (the community’s attitudes or “how everyone thinks”) so the self can conduct an inner dialogue with this more generalized voice. But this is not meant to be a takeover of the self by the generalized other. According to Mead, it is about becoming double such that the socially shaped “me” is continually in dialogue with the spontaneous “I”, which injects novelty and creativity (Joas, 1997; Mead, 1967). For Mead there can be more than one generalized other, and they can also be in dialogue together. This explains why the classroom approach to teaching thinking championed by Resnick and others was called “accountable talk”. To think well in most educational contexts, you need to be accountable to rules of good thinking in a community, for example, the community of scientists or the community of historians.
Dialogic doubleness gives rise to sparks of insight. This insight is possible not only between specific people in dialogue together but also between individuals or groups and fields of enquiry. In his book on creativity as flow, Csikszentmihalyi (1996) presents the story of quantum physicist Freeman Dyson, who struggled for days with a complex field of theory that he did not feel he really understood, only to find, after taking a break, that the field itself seemed to take on a voice and speak to him through his own fingers, such that he claimed his fingers on the keyboard seemed to be typing the answers before he himself was aware of them. We see exactly the same effect in small groups of students solving reasoning problems. At first the problem is opaque, but after a struggle with it, after talking together to explore all angles, sometimes the problem itself seems to speak back and offers a new configuration or way of seeing that was not visible before. Suddenly things fall into place and the answer seems obvious. This only happens if you allow yourself to engage with the problem and give it the space to breathe and reveal its secret.
Dialogic Thirdness: The Infinite Other
In practice, dialogue is never just a relationship between two separate selves or egos. In imagining the other’s point of view, I am imagining an outside gaze that locates me. Implicit in that imagining is, as Bakhtin points out, a kind of “superaddressee”, or an outside voice that understands what I am saying even if the specific other person that I am talking to does not. Bakhtin had great respect for Buber, and it is notable that a related claim is made by Buber. Implicit in talking to any “thou” is a kind of thouness in general, or what Buber calls the “eternal thou”. This is a claim based on experience that you can try out for yourself. In any real dialogue with another person about a topic that matters to you both, you will find that you are in dialogue not only with each other but also with a kind of horizon of truth and/or goodness that emerges and that takes the form of a voice or perspective that you end up referring to, either explicitly or implicitly.
Some dialogues have a fairly fixed idea of the “superaddressee” or “generalized other”. Dialogue about subtraction in a mathematics classroom is probably implicitly or explicitly invoking the community of mathematicians. An AI trained only on vetted material in mathematics could easily be created to personify the thinking of the community of mathematicians in this way. Similarly, a dialogue about theology in a Catholic seminary might invoke a fairly culturally specific and unchanging view of God and what God thinks as the relevant superaddressee. But according to the superaddressee logic of Bakhtin, based on his claim that there is always a “third voice” or “witness” in any dialogue, it follows that if we engage in dialogue with the voice of mathematicians or with any specific image of “God”, this dialogue will also generate a new perspective or third voice, and so on ad infinitum. This leads to my proposal that we are not only in dialogue with this or that superaddressee voice but always also in dialogue, if only often implicitly, with what I call the Infinite Other. The Infinite Other is not a static voice but rather a process of going beyond any static voice or any settled conception that we might have (Levinas, 1969). In any dialogue there is always the possibility of questioning assumptions and going deeper or further; there is always an outside voice or perspective that one could try to learn to listen to (Wegerif, 2013, 2025).
Dialogic Intelligence and AI
AI in conversational chatbot form can now participate in dialogues as a voice, though not one grounded in embodied subjective experience. It has no feelings or awareness; it is a simulated voice. Yet as a simulated voice it can potentially stand in for and represent whole fields of knowledge, drawing upon the sedimented archive of what has been said and thought on a topic. Thinking dialogically with such a voice requires temporarily inhabiting its perspective in order to hear what has been established and what counts as common sense or consensus within a domain. But if one remains only there, there is the danger of being trapped in cliché, in the collective point of view that George Herbert Mead called the generalized other. To think creatively with AI, students must learn to use this perspective as a platform from which to move further and to generate dialogic space by leaping beyond doubleness into a “thirdness” that invokes not just what has been said but what might yet be said. This horizon of possible meaning, what I call dialogue with the perspective of the Infinite Other, is what calls thought towards questioning, expansion, and transformation.
The conception of intelligence proposed here resonates with posthumanist frameworks that question boundaries between human and non-human agents (Braidotti, 2013; Hayles, 1999). However, the present proposal maintains that certain capacities, including care, ethical responsibility, and the motivation to question, remain distinctively human contributions to dialogic intelligence. AI can participate in dialogue, but humans bear responsibility for the dialogue’s direction and ethical implications.
From this perspective, dialogic intelligence is not another cognitive skill to be added to existing curricula but a fundamental ontological reorientation changing our understanding of the nature of thinking. It is the cultivated capacity to participate consciously in the dialogic processes through which meaning and reality emerge: the opening, widening, and deepening of dialogic space.
Critical Awareness of AI Limitations
LLMs are not neutral interlocutors. They are trained on datasets that encode particular cultural perspectives, often reflecting biases present in internet text and the choices of their corporate developers (Bender et al., 2021). Their responses are further shaped by reinforcement learning from human feedback that optimizes for user satisfaction rather than truth or educational value. Students engaging dialogically with AI must therefore develop critical awareness of these constraints. Dialogic intelligence includes the capacity to recognize when AI responses reflect systematic bias, to probe assumptions underlying AI-generated content, and to maintain independent judgement even when AI appears authoritative.
The chief risk of bringing AI into education is conformity to the generalized other: students ending up as subjects being “spoken by” the discourse of prevailing norms embodied in the AI. Not only is this discourse likely to reflect many prevailing biases, it might also be subject to politically or commercially driven manipulation. The best response to this challenge is a dialogic education that forms people able not only to inhabit discourse but also to reshape it: to ask better questions, hold tensions productively, and take responsibility for the direction of inquiry.
Distinguishing Dialogic From Hybrid Intelligence
The concept of dialogic intelligence shares some features with hybrid intelligence as developed by Cukurova and colleagues and extended by Järvelä and colleagues (Cukurova et al., 2019; Järvelä et al., 2023). Hybrid intelligence, as these researchers define it, integrates human and AI capabilities so that the two can co-evolve and complement each other, with each compensating for the other’s limits. This is a valuable notion. There remains an element of hybridity in dialogic intelligence, with machines taking on the checking and universalizing aspect of intelligence that they can do best.
However, hybrid intelligence accounts do not sufficiently challenge and transform the traditional concept of human intelligence which was critiqued earlier in this paper. Dialogic intelligence is not really about maintaining separate intelligences and then combining them according to their separate affordances so much as about rethinking intelligence itself as relational and arising in the gap between self and other: human and nature; human and machine. Where hybrid intelligence treats human and AI as separate systems to be optimally combined, dialogic intelligence locates thinking in the relational space between them.
Teaching Dialogic Intelligence With AI
How AI is framed makes all the difference. Used instrumentally as a shortcut to answers, it narrows effort and encourages over-reliance on technology. Framed dialogically, AI becomes a partner that complicates assumptions, surfaces alternative perspectives, and helps sustain inquiry. Machines now surpass us at all operations that can be described as algorithms, including deduction, literature review, triangulation, and error-checking, but the really significant difference from humans is that they do not really care. Human care is what ultimately motivates enquiries and projects and what provides the drive that keeps a question open long enough for an unexpected leap of insight to occur. Teaching dialogic intelligence between humans and AI is about cultivating the doubleness required to listen to AI as a concentrated cultural voice (a “generalized other” or specific “superaddressee”) while also stepping beyond it, because we care enough to wait for the spark of insight that AI alone cannot supply.
Although intelligence always starts with intuition and insight, intelligence is more than just having an insight. Insights always come for a reason, but sometimes they steer us the wrong way in context. They require careful checking against the available evidence, literature review, triangulation, and possibly also logical critiques from other tried and tested theories. AI can be trained to perform this checking and critique function effectively, making the iterative dialogue between human and AI more effective and perspicacious.
A pedagogy coherent with dialogic intelligence treats learning less as the accumulation of correct representations and more as induction into the practice of opening, widening, deepening, sustaining, and applying dialogic space. In practical terms, this implies classroom designs in which inquiry is the default stance and the teacher’s role is to curate conditions under which different perspectives can be held together long enough for new meanings to emerge. One way to operationalize this is to structure learning around recurring “moves”: opening a genuine question (rather than a disguised recall task), widening the space through alternative viewpoints (including those generated or staged by AI), deepening through explicit questioning of assumptions and frames, protecting the space through norms of respect, reciprocity, and careful listening, and then applying what has been learned in decisions or actions that matter beyond the classroom. Within such designs, AI is most educationally valuable not as an answer engine but as a dialogic partner that can help stabilize the inquiry by offering counter-positions, requesting clarification, modelling careful paraphrase, and prompting students to make their reasons visible while leaving responsibility for judgement and direction with learners and teachers.
Used monologically, AI can weaken thinking; used dialogically, it extends and expands thinking into human–AI dialogic intelligence contributing to the larger collective intelligence of the planet as a whole (Wegerif & Major, 2024). Re-orienting education around dialogic intelligence is about enabling people to learn and create with machines and with each other, in the service of problems that matter. Used in this way, AI is neither a threat nor a mere tool but part of teaching and learning for collective intelligence (Wegerif, 2025).
Assessing Progress in Dialogic Intelligence
Unlike IQ, which gained traction through a single score given to individuals, dialogic intelligence is an emergent relational process. We need to be able to specify dialogic intelligence sufficiently to be useful as a guide for pedagogy and formative assessment without slipping into the danger of reifying it as if it were a fixed individual trait. If intelligence emerges through dialogue rather than residing within individual minds, then measuring it requires attending to relational processes and collective outcomes rather than individual traits and standardized performances. Practically, it often takes the form of a repertoire of moves and dispositions: attentive listening that seeks meaning as well as information; questioning that probes assumptions and evidence; contingent responsiveness that allows one’s stance to shift; creative reframing that introduces new metaphors or problem formulations; tolerance of ambiguity that resists premature closure; integrative synthesis that crystallizes shared understanding; and an ethical orientation that treats dialogue as answerable to truth, others, and the common good.
Dialogic intelligence manifests through observable patterns of engagement that can be witnessed and described, even if not precisely measured in conventional terms. Table 1 summarizes a few of these patterns, their theoretical foundations, and observable indicators.
Observable Patterns of Dialogic Intelligence.
Note. *Ontological listening: attending not merely to informational content but allowing oneself to be genuinely affected and expanded by the encounter with another’s perspective (English, 2016)
From a dialogic perspective, assessment itself becomes a collaborative inquiry rather than individual measurement. Instead of external evaluators judging individual performance against predetermined criteria, assessment involves participants, including the AI interlocutor, in ongoing reflection on the quality of their collaborative thinking. Students, teachers, and AI systems can become partners in recognizing and developing dialogic intelligence. This might involve structured reflection on moments when dialogue opened new possibilities, collective analysis of conversations that led to significant insights, and shared consideration of how collaborative processes might be improved.
Rather than focusing only on final products, assessment attends to the quality of collaborative processes through which outcomes emerge. This could include analysis of dialogue transcripts that reveal patterns of questioning and responding, documentation of how perspectives evolved through conversation, and reflection on moments when groups successfully navigated disagreement or uncertainty.
This initial framework for assessment is not about grades but about living, formative feedback. AI can provide a growing profile and portfolio for each person. In the early stages of teaching dialogic intelligence with AI at Cambridge, we are experimenting with frameworks and ways of giving feedback. However, it is possible to imagine, with enough data, that a statistically robust measure of how individuals collaborate with AI and with others to participate in and facilitate better collective intelligence might be possible in the future. This would not be a universal measure of cognitive ability in the abstract but a more situated account of how individuals participate in thinking with AI and with others, which can be used as a basis for further educational development.
Further Research
This theory of dialogic intelligence as the new aim of education suggests a practical research agenda to test it and develop it further. A first line of work concerns measurement: whether the assessment pointers sketched above can be operationalized in ways that are rigorous and useful for supporting practice, without reducing dialogic intelligence to a decontextualized trait. The challenge is to develop situated indicators that travel across contexts while respecting that dialogic intelligence is emergent, relational, and normatively answerable. That points towards mixed-method designs combining (i) qualitative micro-analysis of dialogic episodes (including silence, joint attention, and reframing moves), (ii) process measures derived from dialogue traces (human–human and human–AI), and (iii) domain-appropriate outcomes (e.g., conceptual change, decision quality, creativity, ethical sensitivity, and the capacity to sustain inquiry).
A second line of work concerns design: building and evaluating pedagogies and interfaces that scaffold dialogic intelligence across the curriculum. Some of this is already underway in Cambridge work on AI roles such as a moderator (supporting equitable participation and productive disagreement) and a dialogue coach (prompting paraphrase, challenge, and reflective pauses). The next step is to develop a third role: disciplinary induction agents that induct learners into curriculum areas such as mathematics, history, science, or literature, as living, contestable traditions of inquiry, through dynamic and responsive conversational encounter. Alongside this, a critic/triangulator role can systematically stress-test emerging claims against evidence, alternative interpretations, and potential bias. Design-based research can iterate between classroom trials and theory refinement, comparing interface and prompt structures, tracking norm internalization over time, and testing whether students transfer dialogic dispositions to situations where the AI is absent. If dialogic intelligence is an ontological reorientation rather than a technique, then future research must ask not only “Does it work?” but “What kinds of persons and communities does it help to form, and at what ethical cost or risk?”
Conclusion
The advent of generative AI into our lives as the new dominant cognitive technology offers us a chance to rethink what we are doing in education. The way in which AI can already achieve so many cherished educational goals suggests that we have been seeing both thought and education through the lens of print literacy in ways that are limiting. Print literacy–based formal schooling has promoted a monologic paradigm for understanding knowledge and also intelligence as if both only exist trapped inside little space-time boxes. The success of AI reveals to us that the reality of mind is quite different and much more expansive.
The dialogic alternative I propose is that we recognize that thinking is not a product of individual brains but of the dialogic space that opens up between perspectives when they are held together in tension. Far from being trapped in space and time, this dialogic space opens up awareness of an infinite or unbounded horizon within which different perspectives clash, play, and sometimes come together in the creation of something new. Dialogic intelligence involves caring for the world and questioning it in a way that sustains this space of openness until insights come and creative ways forward emerge. A key mechanism of dialogic intelligence involves becoming double so as to be able to see a topic from at least two sides: one’s own side and that of an interlocutor. When the interlocutor is an AI agent, that becoming double can be particularly expansive, enabling learners to explore and interrogate large areas of knowledge. There is a potential here for a kind of leap-frogging progress through which students expand their identities and their possibilities for action through dialogue with machines.
The distinctive contribution of dialogic intelligence lies in its ontological reframing rather than its pedagogical techniques alone. Where hybrid intelligence models treat human and AI as separate systems to be optimally combined, dialogic intelligence locates thinking in the relational space between them. This shift has practical implications: rather than asking how AI can augment individual human cognition, we ask how humans and AI together can create and sustain spaces in which new understanding emerges. The goal is not to produce more efficient individual thinkers but to cultivate people capable of participating in, and taking responsibility for, the collective thinking through which humanity navigates its challenges.
Just as print literacy once supported a massive expansion of schooling which transformed human thinking, AI now offers us the chance to amplify and extend knowledge and intelligence further and also faster than ever before. Education’s task is therefore not to defend against the threat to existing systems posed by AI but to embrace it as a participant in the ongoing evolution of our collective intelligence. If we succeed in cultivating dialogic intelligence at the centre of pedagogy, AI may become not a disruption but a partner in fulfilling the deepest vocation of education: enabling humanity, with its machines, to learn, create, and flourish together in an ever-expanding dialogue with the universe.
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
