The Human Learning Experience
What Artificial Intelligence Is Doing to Learning, and The Questions We Have as Educators
Dr. Analia Denmon
Learning Atelier Executive Director
Director of Learning Innovation & Development at Istanbul International Community School
Learning Atelier Executive Director
Director of Learning Innovation & Development at Istanbul International Community School
"AI has read everything but it has experienced nothing. It has never belonged to a community, felt the weight of an idea, been moved by another person’s thinking, or wondered who it was becoming. Education exists to produce all of those things. That is the entire argument."
Dr. Analia Denmon
Executive Summary
This paper makes a single argument. When students outsource their thinking to artificial intelligence, they are not simply submitting work they did not do. They are being deprived of the experiences through which human beings actually develop as thinkers, as social beings, and as people with a sense of who they are and what they value.
Drawing on ten converging bodies of research, from neuroscience to social psychology, from developmental identity theory to the philosophy of knowledge, and including landmark studies published between 2024 and 2025, this paper documents what is now empirically established. AI, when used as a substitute for cognitive effort, systematically intervenes in the biological, cognitive, social, embodied, and developmental conditions under which learning occurs.
The question education now faces is not whether to permit AI. AI is already integrated into students’ cognitive lives and will remain so. The real question is how to distinguish AI as a tool within a process the student owns from AI as a substitute for the student’s cognitive and developmental work, and how to redesign instruction, assessment, and curriculum so that the conditions of genuinely human learning are protected and amplified. Current assessment architectures cannot see this distinction. The profession’s urgent task is to build ones that can.
This paper makes a single argument. When students outsource their thinking to artificial intelligence, they are not simply submitting work they did not do. They are being deprived of the experiences through which human beings actually develop as thinkers, as social beings, and as people with a sense of who they are and what they value.
Drawing on ten converging bodies of research, from neuroscience to social psychology, from developmental identity theory to the philosophy of knowledge, and including landmark studies published between 2024 and 2025, this paper documents what is now empirically established. AI, when used as a substitute for cognitive effort, systematically intervenes in the biological, cognitive, social, embodied, and developmental conditions under which learning occurs.
The question education now faces is not whether to permit AI. AI is already integrated into students’ cognitive lives and will remain so. The real question is how to distinguish AI as a tool within a process the student owns from AI as a substitute for the student’s cognitive and developmental work, and how to redesign instruction, assessment, and curriculum so that the conditions of genuinely human learning are protected and amplified. Current assessment architectures cannot see this distinction. The profession’s urgent task is to build ones that can.
When the Proxy Breaks
Every significant educational technology has forced the same question. Given that machines can do this, what are humans for? The calculator made us ask what arithmetic was for. Google made us ask what memory was for. AI makes us ask what thinking is for.
There is a temptation to treat this as simply the next iteration of a familiar pattern. It is not. Previous technologies automated storage, retrieval, or calculation. They left thinking itself largely untouched. Generative artificial intelligence is the first technology in human history that automates the production of apparent thinking itself. It can argue, synthesise, explain, interpret, and produce reflective writing that reads as if a human had genuinely reflected. This is not a continuation of the pattern. It is a different order of disruption.
The essay, the research paper, the extended project, the primary instruments of assessment in IB, AP, A-Level, and most academic programmes worldwide, were always proxies. They were never the learning itself. They were evidence that the learning had occurred: that a student had wrestled with a problem, organised thought, constructed an argument, changed their mind, and produced something bearing the mark of genuine intellectual effort. That proxy has been broken. A student can now produce a sophisticated written argument without having thought sophisticatedly about anything. The artefact remains. The cognitive act it was meant to evidence has not taken place.
The evidence remains. The learning has not occurred. This is the situation every teacher currently walks into, under assessment systems that were not built to see the difference.
The question is not AI yes or AI no. That question is closed. What the profession must urgently understand is something harder: what, precisely, AI does to human learning when used as a substitute for thinking, and what conditions of human development it cannot provide. The research that follows is an attempt to answer that question or at least to open that conversation.
Every significant educational technology has forced the same question. Given that machines can do this, what are humans for? The calculator made us ask what arithmetic was for. Google made us ask what memory was for. AI makes us ask what thinking is for.
There is a temptation to treat this as simply the next iteration of a familiar pattern. It is not. Previous technologies automated storage, retrieval, or calculation. They left thinking itself largely untouched. Generative artificial intelligence is the first technology in human history that automates the production of apparent thinking itself. It can argue, synthesise, explain, interpret, and produce reflective writing that reads as if a human had genuinely reflected. This is not a continuation of the pattern. It is a different order of disruption.
The essay, the research paper, the extended project, the primary instruments of assessment in IB, AP, A-Level, and most academic programmes worldwide, were always proxies. They were never the learning itself. They were evidence that the learning had occurred: that a student had wrestled with a problem, organised thought, constructed an argument, changed their mind, and produced something bearing the mark of genuine intellectual effort. That proxy has been broken. A student can now produce a sophisticated written argument without having thought sophisticatedly about anything. The artefact remains. The cognitive act it was meant to evidence has not taken place.
The evidence remains. The learning has not occurred. This is the situation every teacher currently walks into, under assessment systems that were not built to see the difference.
The question is not AI yes or AI no. That question is closed. What the profession must urgently understand is something harder: what, precisely, AI does to human learning when used as a substitute for thinking, and what conditions of human development it cannot provide. The research that follows is an attempt to answer that question or at least to open that conversation.
What Is Actually Being Lost: Ten Layers of Evidence
Across ten distinct bodies of research, foundational theory spanning the twentieth century and confirmed by empirical studies published between 2023 and 2025, a coherent picture has emerged. Learning is not the acquisition of knowledge. It is a biological, cognitive, social, embodied, and experiential event that requires the learner to be present, to struggle, to belong, to feel, and to become. When these conditions are removed by cognitive outsourcing, the output may remain. The event does not occur.
1. Neurological: Learning is physical.
Hebb’s Law (1949) established that learning is the formation of neural pathways through effortful activity. Neurons that fire together wire together. In June 2025, MIT Media Lab researchers measured this directly. Using EEG technology with 54 participants writing essays across four sessions, they found that ChatGPT users showed up to 55% reduced brain connectivity compared to unassisted writers. Critically, the deficit persisted after AI use ended. 83% of ChatGPT users could not accurately quote or summarise their own essays shortly after writing them. The work had passed through them without becoming theirs.
2. Cognitive: The struggle is the learning.
Research on the Testing Effect (Roediger) and the Generation Effect (Jacoby, 1978) established that retrieval effort and generation attempts, even unsuccessful ones, are the primary mechanisms of deep encoding. A 2024 University of Bremen study analysing thousands of student essays found that AI users scored 6.71 points lower on subsequent exams than non-users. The effect was most damaging to students with the highest learning potential. The students who stood to learn most were harmed most by outsourcing.
3. Metacognitive: Knowing what you do not know. Metacognition, the capacity to monitor and regulate one’s own thinking, is the strongest known predictor of academic achievement (Flavell, 1979). It develops only through the experience of noticing one’s own cognition in action. AI is structurally incompatible with this development. When AI produces the output, the student has no thinking to observe. A 2025 paper in the British Journal of Educational Technology documented a pattern it termed “metacognitive laziness”: reduced self-monitoring, declining engagement, and weakened transfer performance in students who routinely use generative AI.
4. Motivational: The collapse of earned competence. Deci and Ryan’s Self-Determination Theory (1985) identified autonomy, competence, and relatedness as the foundations of intrinsic motivation. AI undermines all three. The MIT study found ownership of essays was lowest in the ChatGPT group. A Harvard study (2025) found AI makes users more productive but less motivated. Efficiency rises; meaning does not. A generation is being trained to produce without owning, perform without growing, and achieve without belonging.
5. Epistemological: The narrowing of what it is possible to think. Polanyi (1966) argued that tacit knowledge, the knowing built through accumulated lived experience, cannot be taught or generated. Doshi and Hauser’s 2024 study in Science Advances gave this argument empirical teeth. AI-assisted writers produced individually more creative stories, but the stories became significantly more similar to each other than unassisted stories. Individual outputs improve. Collective novelty collapses. Applied to classrooms, this means intellectual voices, still forming in adolescence, converge toward a centre they did not choose.
6. Social: Learning has always happened between people
Vygotsky (1978) established that cognition is social before it is individual. Every higher function develops first between people, then inside them. Lave and Wenger (1991) extended this: learning is participation in communities of practice, not transfer of information. Wegner’s (1987) research on transactive memory documented that groups generate knowledge no individual member possesses. AI cannot participate in any of this. It can simulate dialogue, but it cannot be genuinely disagreed with, moved, or surprised. The classroom as a collective cognitive instrument, producing thinking no participant could produce alone, is not available in solitary AI interaction.
7. Experiential and democratic: Dewey’s unfinished warning. Dewey (1916, 1938) argued that schools are miniature democratic communities, and that the capacity for democratic life, holding complexity, changing one’s mind in good faith, reasoning collectively, is built through sustained educational experience. He also introduced the category of the mis-educative experience: one that actively damages future learning capacity by producing habits that close off growth. An AI-assisted essay, repeated over years, may be precisely this. The student has produced work. They have formed the habit of prompting rather than thinking. The capacities democratic life requires have not developed. A 2025 APA health advisory on AI and adolescent wellbeing makes these developmental concerns explicit.
8. Embodied: The body is not incidental to the mind
Varela, Thompson, and Rosch’s The Embodied Mind (1991), along with subsequent neuroscientific research, established that cognition is grounded at every level in bodily experience. A 2024 Educational Psychology Review synthesis confirmed that learning is enhanced when brain, body, and environment mutually engage, through gesture, physical manipulation, handwriting, and social presence. AI interaction is structurally disembodied. It removes both the embodied dimension of the student’s own thinking and the multisensory social richness in which human cognition best develops.
9. Identity: School is where adolescents become someone
Erikson’s developmental theory established adolescence as the critical period for identity formation, a process that happens relationally, in sustained encounter with other minds that take the adolescent seriously. A 2024 review in Child Development Perspectives found that educational identity depends on genuine relationships, belonging, and sustained experience of being intellectually engaged by others. Immordino-Yang’s 2024 research shows that adolescent brain development is itself inherently social and emotional. AI cannot provide what Riolo et al. (2025) call the “co-regulatory, containing functions of human caregiving.” Adolescents may graduate with qualifications but without a formed intellectual self, never having been recognised by another mind as a particular and irreplaceable someone.
10. Affective: We feel, therefore we learn
Immordino-Yang and Damasio’s foundational 2007 paper established that emotion is not separate from cognition. It is the condition of cognition’s depth. We remember what we care about. We understand what moves us. AI is affectively flat. It cannot be curious, cannot feel the weight of an idea, and most importantly, cannot create the relational psychological safety that allows students to take genuine intellectual risks. Only another human being, present and invested, can provide that. Without it, information is processed but not formative. The student may acquire content. They will not be shaped by it.
Across ten distinct bodies of research, foundational theory spanning the twentieth century and confirmed by empirical studies published between 2023 and 2025, a coherent picture has emerged. Learning is not the acquisition of knowledge. It is a biological, cognitive, social, embodied, and experiential event that requires the learner to be present, to struggle, to belong, to feel, and to become. When these conditions are removed by cognitive outsourcing, the output may remain. The event does not occur.
1. Neurological: Learning is physical.
Hebb’s Law (1949) established that learning is the formation of neural pathways through effortful activity. Neurons that fire together wire together. In June 2025, MIT Media Lab researchers measured this directly. Using EEG technology with 54 participants writing essays across four sessions, they found that ChatGPT users showed up to 55% reduced brain connectivity compared to unassisted writers. Critically, the deficit persisted after AI use ended. 83% of ChatGPT users could not accurately quote or summarise their own essays shortly after writing them. The work had passed through them without becoming theirs.
2. Cognitive: The struggle is the learning.
Research on the Testing Effect (Roediger) and the Generation Effect (Jacoby, 1978) established that retrieval effort and generation attempts, even unsuccessful ones, are the primary mechanisms of deep encoding. A 2024 University of Bremen study analysing thousands of student essays found that AI users scored 6.71 points lower on subsequent exams than non-users. The effect was most damaging to students with the highest learning potential. The students who stood to learn most were harmed most by outsourcing.
3. Metacognitive: Knowing what you do not know. Metacognition, the capacity to monitor and regulate one’s own thinking, is the strongest known predictor of academic achievement (Flavell, 1979). It develops only through the experience of noticing one’s own cognition in action. AI is structurally incompatible with this development. When AI produces the output, the student has no thinking to observe. A 2025 paper in the British Journal of Educational Technology documented a pattern it termed “metacognitive laziness”: reduced self-monitoring, declining engagement, and weakened transfer performance in students who routinely use generative AI.
4. Motivational: The collapse of earned competence. Deci and Ryan’s Self-Determination Theory (1985) identified autonomy, competence, and relatedness as the foundations of intrinsic motivation. AI undermines all three. The MIT study found ownership of essays was lowest in the ChatGPT group. A Harvard study (2025) found AI makes users more productive but less motivated. Efficiency rises; meaning does not. A generation is being trained to produce without owning, perform without growing, and achieve without belonging.
5. Epistemological: The narrowing of what it is possible to think. Polanyi (1966) argued that tacit knowledge, the knowing built through accumulated lived experience, cannot be taught or generated. Doshi and Hauser’s 2024 study in Science Advances gave this argument empirical teeth. AI-assisted writers produced individually more creative stories, but the stories became significantly more similar to each other than unassisted stories. Individual outputs improve. Collective novelty collapses. Applied to classrooms, this means intellectual voices, still forming in adolescence, converge toward a centre they did not choose.
6. Social: Learning has always happened between people
Vygotsky (1978) established that cognition is social before it is individual. Every higher function develops first between people, then inside them. Lave and Wenger (1991) extended this: learning is participation in communities of practice, not transfer of information. Wegner’s (1987) research on transactive memory documented that groups generate knowledge no individual member possesses. AI cannot participate in any of this. It can simulate dialogue, but it cannot be genuinely disagreed with, moved, or surprised. The classroom as a collective cognitive instrument, producing thinking no participant could produce alone, is not available in solitary AI interaction.
7. Experiential and democratic: Dewey’s unfinished warning. Dewey (1916, 1938) argued that schools are miniature democratic communities, and that the capacity for democratic life, holding complexity, changing one’s mind in good faith, reasoning collectively, is built through sustained educational experience. He also introduced the category of the mis-educative experience: one that actively damages future learning capacity by producing habits that close off growth. An AI-assisted essay, repeated over years, may be precisely this. The student has produced work. They have formed the habit of prompting rather than thinking. The capacities democratic life requires have not developed. A 2025 APA health advisory on AI and adolescent wellbeing makes these developmental concerns explicit.
8. Embodied: The body is not incidental to the mind
Varela, Thompson, and Rosch’s The Embodied Mind (1991), along with subsequent neuroscientific research, established that cognition is grounded at every level in bodily experience. A 2024 Educational Psychology Review synthesis confirmed that learning is enhanced when brain, body, and environment mutually engage, through gesture, physical manipulation, handwriting, and social presence. AI interaction is structurally disembodied. It removes both the embodied dimension of the student’s own thinking and the multisensory social richness in which human cognition best develops.
9. Identity: School is where adolescents become someone
Erikson’s developmental theory established adolescence as the critical period for identity formation, a process that happens relationally, in sustained encounter with other minds that take the adolescent seriously. A 2024 review in Child Development Perspectives found that educational identity depends on genuine relationships, belonging, and sustained experience of being intellectually engaged by others. Immordino-Yang’s 2024 research shows that adolescent brain development is itself inherently social and emotional. AI cannot provide what Riolo et al. (2025) call the “co-regulatory, containing functions of human caregiving.” Adolescents may graduate with qualifications but without a formed intellectual self, never having been recognised by another mind as a particular and irreplaceable someone.
10. Affective: We feel, therefore we learn
Immordino-Yang and Damasio’s foundational 2007 paper established that emotion is not separate from cognition. It is the condition of cognition’s depth. We remember what we care about. We understand what moves us. AI is affectively flat. It cannot be curious, cannot feel the weight of an idea, and most importantly, cannot create the relational psychological safety that allows students to take genuine intellectual risks. Only another human being, present and invested, can provide that. Without it, information is processed but not formative. The student may acquire content. They will not be shaped by it.
The Evidence in Summary Read together, the ten layers reveal a single pattern. The losses are not localised. They are systemic, covering every dimension through which humans actually become thinkers.
The False Dichotomy
It would be easy, in light of the evidence above, to conclude that AI should be kept out of classrooms. This conclusion is as incorrect as it is tempting.
The debate over whether we are “for” or “against” AI in education is a false dichotomy, and time spent on it is time not spent on the questions that matter. AI is already integrated into the cognitive lives of students. Surveys published in 2024 reported that 86% of students globally use AI tools to support their studies. Any framework that begins from the premise of exclusion will fail practically and pedagogically.
What the evidence establishes is narrower and more useful. It establishes that AI used as a substitute for cognitive effort, as a means of producing outputs without engaging in the processes those outputs were meant to evidence, is not neutral. It is a systematic intervention in the conditions under which human learning occurs. But AI used as a tool within a process the learner still owns, as a sparring partner, a research assistant, a source of counterarguments to evaluate, is not the same thing. The MIT study found that students using search engines showed strong brain engagement comparable to the unassisted group. The problem is not technology. The problem is cognitive abdication.
It would be easy, in light of the evidence above, to conclude that AI should be kept out of classrooms. This conclusion is as incorrect as it is tempting.
The debate over whether we are “for” or “against” AI in education is a false dichotomy, and time spent on it is time not spent on the questions that matter. AI is already integrated into the cognitive lives of students. Surveys published in 2024 reported that 86% of students globally use AI tools to support their studies. Any framework that begins from the premise of exclusion will fail practically and pedagogically.
What the evidence establishes is narrower and more useful. It establishes that AI used as a substitute for cognitive effort, as a means of producing outputs without engaging in the processes those outputs were meant to evidence, is not neutral. It is a systematic intervention in the conditions under which human learning occurs. But AI used as a tool within a process the learner still owns, as a sparring partner, a research assistant, a source of counterarguments to evaluate, is not the same thing. The MIT study found that students using search engines showed strong brain engagement comparable to the unassisted group. The problem is not technology. The problem is cognitive abdication.
"The distinction that matters is not between AI use and AI abstention. It is between AI as a tool within a process the learner drives, and AI as a substitute for the learner’s cognitive and developmental work."
Dr. Analia Denmon
This distinction is currently invisible to most assessment systems. A student who has used AI as a rigorous cognitive companion and a student who has used AI to produce the work itself submit the same polished product. Both are graded against the same rubric. The education profession has no reliable means of telling them apart. Continuing to operate assessment systems that cannot see this difference is not a neutral choice. It is an ongoing failure to protect the learners in our care from the consequences documented above.
Open Questions for the Profession
The evidence presented here does not resolve itself into a single prescription. It opens a set of questions that educators, school leaders, curriculum designers, assessment boards, and policy makers must now address collectively. These questions do not have easy answers. The answers that will matter most will be developed through years of coordinated professional work, not pronounced in advance.
1. If the essay can no longer reliably evidence thinking, what can? What forms of assessment can see the difference between a student who has thought and a student who has prompted, and how do we scale those forms across large educational systems designed for an earlier age?
2. How do we design instruction and classroom practice so that the experiences the research identifies as irreplaceable, struggle, belonging, embodied engagement, affective safety, relational recognition, are actively cultivated rather than passively hoped for?
3. What does productive, learning-enhancing use of AI actually look like in practice, and how do we teach students to distinguish it from the cognitively abdicating use that the evidence warns against?
4. How do we protect the formation of intellectual identity during adolescence, the period when young people are becoming who they are as thinkers, in a technological environment designed to produce outputs faster than identities can form?
5. What new professional capacities must teachers develop to work well in this landscape, and how do we support them in developing these capacities without adding them to a workload already at breaking point?
6. How do we preserve and strengthen the classroom as a collective cognitive instrument, a community of practice that generates thinking no individual could generate alone, when the dominant pull on students is toward solitary, screen-mediated cognitive shortcuts?
7. What is the responsibility of assessment boards, qualification authorities, and governments whose current frameworks were designed for a world that no longer exists, and how quickly can they move?
The evidence presented here does not resolve itself into a single prescription. It opens a set of questions that educators, school leaders, curriculum designers, assessment boards, and policy makers must now address collectively. These questions do not have easy answers. The answers that will matter most will be developed through years of coordinated professional work, not pronounced in advance.
1. If the essay can no longer reliably evidence thinking, what can? What forms of assessment can see the difference between a student who has thought and a student who has prompted, and how do we scale those forms across large educational systems designed for an earlier age?
2. How do we design instruction and classroom practice so that the experiences the research identifies as irreplaceable, struggle, belonging, embodied engagement, affective safety, relational recognition, are actively cultivated rather than passively hoped for?
3. What does productive, learning-enhancing use of AI actually look like in practice, and how do we teach students to distinguish it from the cognitively abdicating use that the evidence warns against?
4. How do we protect the formation of intellectual identity during adolescence, the period when young people are becoming who they are as thinkers, in a technological environment designed to produce outputs faster than identities can form?
5. What new professional capacities must teachers develop to work well in this landscape, and how do we support them in developing these capacities without adding them to a workload already at breaking point?
6. How do we preserve and strengthen the classroom as a collective cognitive instrument, a community of practice that generates thinking no individual could generate alone, when the dominant pull on students is toward solitary, screen-mediated cognitive shortcuts?
7. What is the responsibility of assessment boards, qualification authorities, and governments whose current frameworks were designed for a world that no longer exists, and how quickly can they move?
The Human Experience
What the research documents is coherent and sobering. Education is not the production of outputs. It is the formation of human beings. The neural pathways built under struggle. The intellectual identity forged in the presence of other minds. The tacit knowledge that comes only from lived experience. The democratic sensibility that grows from years of real deliberation with others. The affective architecture through which anything can come to matter.
These are not peripheral to what schools are for. They are exactly what schools are for. And they are, each of them, placed at risk when cognitive effort is systematically outsourced.
The research also points toward what education must now consciously protect. The experiences that only humans can provide for other humans, the struggle witnessed by a teacher who adapts in real time, the belonging built through a community of peers who take one another seriously, the emotional safety that makes intellectual risk possible, the identity formed through being genuinely seen, these are not vulnerabilities of the educational project in the age of AI. They are its purpose. They have always been its purpose. AI simply makes that purpose visible by contrast.
What is needed now is neither resistance to a technology that cannot be resisted, nor uncritical embrace of a tool whose costs are being documented in real time. What is needed is renewed clarity about what education exists to do, and the collective professional work of building systems that protect and amplify the conditions under which human beings actually develop. The human experience of learning is the thing that cannot be generated. It can only be lived.
What the research documents is coherent and sobering. Education is not the production of outputs. It is the formation of human beings. The neural pathways built under struggle. The intellectual identity forged in the presence of other minds. The tacit knowledge that comes only from lived experience. The democratic sensibility that grows from years of real deliberation with others. The affective architecture through which anything can come to matter.
These are not peripheral to what schools are for. They are exactly what schools are for. And they are, each of them, placed at risk when cognitive effort is systematically outsourced.
The research also points toward what education must now consciously protect. The experiences that only humans can provide for other humans, the struggle witnessed by a teacher who adapts in real time, the belonging built through a community of peers who take one another seriously, the emotional safety that makes intellectual risk possible, the identity formed through being genuinely seen, these are not vulnerabilities of the educational project in the age of AI. They are its purpose. They have always been its purpose. AI simply makes that purpose visible by contrast.
What is needed now is neither resistance to a technology that cannot be resisted, nor uncritical embrace of a tool whose costs are being documented in real time. What is needed is renewed clarity about what education exists to do, and the collective professional work of building systems that protect and amplify the conditions under which human beings actually develop. The human experience of learning is the thing that cannot be generated. It can only be lived.
AI is not the question. The question is what kind of human beings we are educating, and under what conditions those human beings develop. The evidence tells us what those conditions are. Protecting and amplifying them is the work ahead.