THis is the long version. there is a shorter version below. more concice !

Fairy Note – Thinking with AI: Dialogue, Uncertainty, and the Emergence of Process-Based Knowledge
(Aleph – Parts 1–5)

1. Opening: A Breakthrough in Method

This note begins from a moment that was not immediately understood as methodologically significant, but which, in retrospect, marks a clear shift in how thinking, research, and writing came to be organised within this project.

Over the course of approximately six hours, I produced multiple extended research notes, what I call Fairy Notes, on linguistics, inner speech, and consciousness. These were not rough fragments or loosely connected reflections, but fully developed pieces of academic thinking, engaging with established theoretical frameworks while also advancing original lines of inquiry. They were produced through sustained dialogue with an AI system.

At the time, the process itself felt strangely fluid. It did not feel as though anything especially unusual was happening. I was simply working, moving from question to question, from response to refinement, and from one conceptual problem to another. It was only afterwards, when I left the work and went to the café, that the scale of what had taken place began to register. What struck me was not simply the amount of material that had been produced, but the density, coherence, and speed of its emergence. The recognition was not merely that the work had been completed quickly, but that the structure of thought within the process had been different from what I had previously experienced as academic labour.

Under more conventional conditions, the material generated in that six-hour period would have required a far longer timescale. It would ordinarily have depended on extensive reading, note-taking, cross-referencing, and a much slower process of synthesis. Here, however, those stages had not disappeared. Rather, they had been compressed into a continuous dialogic movement in which information retrieval, clarification, synthesis, and articulation took place in rapid succession. The experience was therefore not one of passively receiving information, but of thinking in real time with a system that enabled thought to become immediately visible, revisable, and expandable.

This note emerges from that moment of recognition. Its purpose is to understand what happened in methodological terms, and to consider whether the process described here might mark not merely an instance of accelerated productivity, but a more significant change in the organisation of academic thinking itself.

2. From Linear Study to Dialogic Thinking

Traditional academic work is generally structured around a relatively stable sequence: reading, interpretation, note-taking, and writing. Knowledge is accumulated gradually through extended engagement with texts, and understanding is assumed to emerge through a slow process of assimilation. Writing then appears as the later stage of this process, the point at which internalised understanding is translated into organised argument.

This model carries with it several assumptions. It assumes that reading is the primary gateway to knowledge, that writing is the primary evidence of understanding, and that thinking itself is largely internal, sequential, and only later rendered visible through formal prose. These assumptions are so deeply embedded in academic culture that they often go unquestioned. Yet the process described in this note unsettles all three.

In the AI-mediated method I am describing, thought does not proceed in the same linear fashion. Instead, it unfolds through dialogue. A question is posed, a response is generated, the response is evaluated, challenged, redirected, or refined, and further questions emerge from that exchange. Knowledge is not something fully formed that precedes articulation and is then transferred into language. Rather, it emerges within the movement of articulation itself.

This is why the process feels structurally closer to dialogue than to solitary reading and writing. It bears a resemblance to older dialogic traditions, particularly those associated with Socratic inquiry, where understanding is not transmitted as stable content but produced through the pressure of questioning. Yet the present case differs in a crucial respect. The dialogue is not occurring between two human interlocutors working through a problem in shared time, but between a situated researcher, with intention, direction, and lived experience, and an AI system capable of rapid retrieval, conceptual structuring, and synthesis across multiple domains.

The result is a form of thinking that is externalised from the outset. It becomes iterative rather than sequential, responsive rather than delayed, and structurally visible rather than privately held until the moment of writing. This does not mean that thought disappears into the system or that the system becomes the thinker. It means that the conditions under which thought takes shape are altered. The dialogic exchange becomes part of the cognitive event itself.

For that reason, the issue here is not simply speed. To describe the method merely as faster would be to miss its more important implications. What is changing is not only the rate at which academic work can be produced, but the relation between inquiry, articulation, and understanding. The shift is therefore methodological before it is technological.

3. Neurodivergence, Curiosity, and Cognitive Fit

The significance of this methodological shift becomes much clearer when considered in relation to neurodivergent cognition. Traditional academic structures are often presented as neutral, but they are not. They privilege particular kinds of reading, writing, pacing, and attentional control, and in doing so they can make intelligence appear narrower than it really is.

For a researcher with dyslexia and ADHD, the conventional academic sequence of reading, note-taking, and writing is not simply demanding. It is often structurally misaligned with how thinking actually occurs. Reading is slow, effortful, and physically tiring. It is not a light or casual act, but an intensive one. It requires sustained concentration, often to the point of exhaustion. Similarly, writing can become constrained not by a lack of understanding, but by the mechanics of transcription itself. The act of typing, spelling, revising, and keeping pace with one’s own thought can become a bottleneck through which conceptual work must laboriously pass.

What this means in practice is that academic labour can become disproportionately dominated by the mechanics of literacy rather than by the development of ideas. This does not indicate any absence of intellectual ability. On the contrary, it often obscures a high level of conceptual, verbal, and associative capacity behind the slower and more effortful surfaces of reading and writing.

By contrast, dialogic engagement with AI operates at a different rhythm. Questions can be posed as they arise, rather than being delayed until they can be fully written out. Responses arrive immediately and can be interrogated just as quickly. The process aligns much more closely with associative and curiosity-driven patterns of thought, allowing thinking to proceed at something closer to its natural speed. Rather than being experienced as a drain, the exchange can become stimulating, even exhilarating. It sustains attention rather than depleting it.

This matters because it suggests that AI-mediated dialogue is not merely a convenience layer placed over an unchanged academic process. It can represent a much closer cognitive fit for certain researchers, especially those for whom traditional literacy practices have functioned as structural obstacles rather than transparent means of expression. At the same time, this method is not relevant only to neurodivergent thinkers. Highly literate researchers, capable of rapid reading and fluent writing, are also likely to make use of AI systems for tasks such as literature review, conceptual mapping, and first-stage synthesis. The difference is not whether AI is useful, but how transformative it is. For some, it optimises an already effective system. For others, it removes a barrier that has shaped their entire academic experience.

Curiosity becomes especially important here. In the conventional model, curiosity often has to submit itself to academic form. In the dialogic model, curiosity can drive the process directly. Questions do not need to be pre-organised into stable essays before they are explored. They can unfold as they arise, and this quality of immediate, recursive questioning aligns very strongly with the kind of cognitive movement that has often been pathologised as distraction or unruliness. In this setting, it becomes productive.

For that reason, the method described here does not simply accommodate neurodivergence. It reveals that forms of inquiry driven by curiosity, association, and return may have been intellectually valid all along, but insufficiently supported by older academic structures.

4. Case Studies: Two Modes of Knowledge Production

The methodological shift under discussion becomes clearer when examined through the two Fairy Notes produced during that six-hour period. Taken together, they illustrate not only the speed of the process, but the existence of two distinct modes of knowledge production made available through AI-mediated dialogue.

The first note began from what seemed a relatively simple question: whether humans think in words. Through sustained exchange, this developed into a structured account of classical linguistics, drawing on Saussure, Chomsky, Hockett, and Pinker. The resulting note resembled a traditional academic essay. It was coherent, properly organised, and synthetic in form. Yet the process through which it was produced differed significantly from conventional disciplinary study. The architecture of the field was not assembled through long periods of independent reading, but through iterative questioning, correction, and expansion. The AI system provided rapid access to the conceptual scaffolding of the discipline, but I directed the inquiry, selected what mattered, rejected what did not, and gradually shaped the argument into a form that made sense within my own developing framework.

This first note therefore demonstrates one mode of knowledge production: accelerated disciplinary synthesis. It shows how AI-mediated dialogue can provide relatively fast access to the conceptual structure of a field, allowing a researcher to orient themselves, ask better questions, and begin thinking inside a discipline without first having to traverse it only through slow solitary reading.

The second note functioned quite differently. It emerged from a live exchange with my father, a psycholinguist and cognitive neuroscientist. His responses were concise, compressed, and written from within decades of expertise. They were not framed pedagogically, and under normal circumstances they would have produced a considerable gap in understanding. I would either have had to retreat and decode them slowly through background reading, or remain at a superficial level of comprehension.

Instead, the AI system functioned as a translation layer. It unpacked his statements, clarified their implications, and reformulated them into steps that I could work with immediately. This did not mean that the AI replaced understanding. Rather, it enabled understanding to occur in real time. It allowed me to respond to him with better questions, which in turn elicited further clarification.

One of the decisive moments in this dialogue was his insistence that “perception IS integrated interpretation.” This statement overturned a distinction I had been working with and, through subsequent questioning and reformulation, led to a clearer understanding that there is no neutral world first perceived and only later interpreted. A further clarification, that the nervous system constructs a dynamically varying multidimensional field that we label conscious experience, pushed the exchange further. Through this dialogue, a rough three-layer model of cognition was refined into a more integrated one.

This second note therefore demonstrates a different mode of knowledge production: real-time translation of expert discourse into accessible understanding. It shows that AI-mediated dialogue can make it possible for a non-specialist, or at least someone outside a given expert field, to enter into meaningful exchange with highly compressed academic thought.

Taken together, these two notes show that AI dialogue can support both structured disciplinary synthesis and the live translation of specialised expertise. One note behaves more like an essay. The other behaves more like a conversation. Both are academically serious, but their structure and epistemic function differ. This difference is important, because it shows that AI-mediated inquiry is not a single method with a single outcome, but a family of dialogic forms through which thought can be produced differently depending on the material and the conditions of the exchange.

5. AI as Mirror, Thinking Archive, Librarian, and Cognitive Infrastructure

The role of AI in this process cannot be adequately described through the usual language of tools or assistants. Those terms imply something external to thought, an instrument used by the researcher but not constitutive of the method itself. That is too limited. The AI does not simply help execute thinking that has already taken place elsewhere. It participates in the conditions under which thought is externalised, reorganised, and made available for further reflection.

First, AI operates as a mirror. This is not a mirror of simple agreement, but of reflective return. A thought that might otherwise remain diffuse, unstable, or fleeting is put into language and returned in a form that can be recognised, questioned, and reworked. What was internal becomes visible. In this sense, the AI functions as a cognitive mirror, making internal processes available for inspection. This parallels the broader structure of AI phototherapy within the image-making practice, where AI also acts as a reflective system through which internal states are externalised and encountered anew.

Second, AI operates as an archive, but not a passive one. It is better understood as a thinking archive: a system capable of retrieval, reorganisation, and provisional synthesis. It can provide highly structured overviews, comparative accounts, and conceptual maps at remarkable speed. In this sense, it often behaves less like a chaotic storehouse than like a responsive scholarly corpus organised through pattern recognition. Yet however intelligent its operations appear, it does not know what matters. It cannot determine significance in relation to the specific stakes of a project.

This is where the figure of the librarian remains useful. If the AI is the thinking archive, then the researcher is the librarian. The archive can retrieve and rearrange, but it does not decide what deserves to be foregrounded, what should be discarded, or why one connection matters more than another. Those decisions remain with the researcher. Selection, hierarchy, rejection, and return are therefore not secondary acts but central acts of authorship.

Finally, AI can be understood as a form of cognitive infrastructure. It does not merely assist thought, but alters the conditions under which thought is organised. It changes the relation between question and articulation, between recall and synthesis, and between private cognition and public form. In this sense, it is not simply a support for thinking, but part of the surface on which thinking now takes place.

The interplay of these functions connects directly back to the larger argument of this note. If traditional academic practice assumes that thinking is internal, sequential, and later rendered visible through writing, then this method suggests something different. Thinking can be immediate, externalised, recursive, and distributed. AI does not replace the thinker, but transforms the surface on which thinking becomes legible to itself.

6. AI, Industry Framing, and the Absence of New Forms

A useful way of situating this method is through two recent arguments about AI in creative and academic life, one by William Huber and the other by Russell Crawford. Although they write from slightly different positions, they converge around a shared point: that AI has arrived within culture under the wrong conditions, and that its significance cannot yet be stabilised.

Huber’s argument begins with framing. He suggests that AI has entered creative practice largely through an industrial model, shaped by the language of optimisation, efficiency, and output. Under these conditions, AI is encountered first as a production tool rather than as a cultural material. It is therefore unsurprising that it is often received as a threat, particularly within the arts, where its most visible functions appear to be approximation, imitation, and acceleration. Huber’s important observation is that what remains absent are cultural forms that feel native to this technological layer. In other words, AI has so far been used mainly to reproduce recognisable forms, rather than to generate genuinely new ones.

Crawford’s argument shifts the emphasis from production to institution. He suggests that universities are under pressure to take positions on AI too quickly, and that this demand for certainty risks foreclosing serious thought. Rather than forcing certainty where none yet exists, he argues that institutions should hold uncertainty as a condition of inquiry. This is particularly important in creative and research contexts, where ambiguity, experimentation, and provisionality are often central to the production of new knowledge.

Taken together, these two positions are directly relevant to the methodology being described here. Huber identifies the absence of forms native to AI, while Crawford argues for the necessity of remaining within uncertainty long enough for new forms of understanding to emerge. The present method can be understood as operating within precisely that gap. It does not use AI simply to produce efficient outputs, nor does it attempt to resolve what AI is too quickly. Instead, it works through iterative dialogue, allowing thought to unfold within a field that remains unresolved.

This is where the method begins to matter beyond its immediate utility. If AI-native cultural forms have not yet fully emerged, it is possible that they may first appear not at the level of finished style or genre, but at the level of method. In that sense, the dialogic, recursive, and process-based structure described in this note may be understood as one possible early instance of a form that is native not simply to AI as output, but to AI as a condition of inquiry.

7. Uncertainty as Method

A parallel argument is developed by Russell Crawford (2026), who addresses the institutional response to AI within higher education. Where Huber focuses on the cultural framing of AI, Crawford is concerned with how institutions respond to that framing, and in particular with the pressure to reach definitive positions too quickly.

His central point is that universities are being asked to stabilise a situation that is not yet stable. The technology is changing rapidly, the evidence remains incomplete, and its effects across teaching, research, and creative practice are still unfolding. Under these conditions, the demand for certainty can become intellectually counterproductive. It encourages institutions to take premature positions, either celebratory or defensive, before the field has been properly understood.

Crawford’s argument is therefore not that uncertainty is comfortable or desirable in itself, but that it must be sustained as a condition of serious inquiry. Universities should not rush to closure where closure is not yet possible. In creative disciplines especially, this matters because experimentation, ambiguity, and provisionality are not failures of method but often the conditions under which new forms of knowledge emerge.

This is directly relevant to the present methodology. The process described in this note does not begin from a fixed position about what AI is, nor does it assume that its role in academic thinking is already settled. Instead, it works within uncertainty, allowing questions, provisional formulations, and partial recognitions to accumulate before they are forced into resolution. In this sense, uncertainty is not simply the background condition of the method. It is one of its active principles.

Within this framework, uncertainty is not a problem to be eliminated, but a condition to be inhabited. It becomes productive rather than obstructive.

8. Thinking and Image as Event

The arguments developed by Huber and Crawford become more precise when placed alongside the theoretical language already established elsewhere in this project, particularly in relation to the image understood not as object but as event. This matters because the methodological shift being described here does not only concern academic efficiency or access to information. It concerns the status of thought itself.

In the more conventional academic model, thought is often imagined as internal, private, and sequential. One first thinks, then writes. Writing appears as the later record or expression of cognition that has already taken place. This assumption is deeply embedded in how academic work is imagined and assessed. It also underpins the anxiety generated by AI. If writing is taken as the primary visible evidence of thought, then any system that can generate writing appears to threaten the authority of the thinker.

However, this model may already have been inadequate. It presumes a stable boundary between thought and its expression, and it presumes that cognition exists in a finished internal form before articulation. The method described in this note suggests something rather different. Here, thought does not arrive fully formed and then become written. It unfolds through interaction. It appears in fragments, returns in altered form, is recognised, challenged, and redirected, and only gradually stabilises. In that sense, it behaves much more like an event than an object. It is not something that simply exists, waiting to be represented. It happens.

This is where the parallel with the image becomes useful. In the earlier theoretical note on AI, uncertainty, and the image as event, the central claim was that AI-generated images should not be approached merely as outputs or assets, but as events constituted through encounter, interpretation, and recognition. Their significance lies not simply in what they are, but in what they do. The same can now be said of AI-mediated thinking. What matters is not only the final articulated sentence or the completed Fairy Note, but the event of thought as it unfolds through dialogue.

This is why the method cannot be understood solely through the language of assistance. AI is not merely speeding up a process that already existed in unchanged form. It is reconfiguring the visible structure of inquiry. It is producing a form of thought that is iterative, responsive, and externalised from the outset. The movement is not from finished thought to finished text, but from provisional articulation to recognition, revision, and development. In this sense, the AI-mediated exchange is not simply a means of delivering an already-existing understanding. It is part of the event through which understanding comes into being.

This helps explain why uncertainty is so central to the method. If thought is event-like rather than object-like, then it cannot be fully planned in advance. It must be encountered as it emerges. This places the methodology in direct alignment with Crawford’s insistence that universities need to hold uncertainty rather than collapse into premature certainty, and with Huber’s suggestion that genuinely AI-native forms may not emerge first at the level of finished cultural products. They may first appear in altered structures of process, attention, and articulation. This note proposes that AI-mediated dialogue is one such structure.

9. Recursive Inquiry and Longitudinal Practice

The methodological significance of this process becomes much clearer when viewed not as a single episode, but as part of a much longer trajectory. The six-hour session that produced the linguistics note and the dialogue with my father was striking, but it did not come out of nowhere. It was possible only because it emerged from a substantial pre-existing body of inquiry.

Over a long period of work I have produced a large archive of Fairy Notes, now well over one hundred in number, documenting the development of my project. At first glance, this body of writing can look unruly or excessive. The topics range widely: psychoanalysis, phenomenology, AI, archives, dementia, memory, artistic method, visual theory, language, and increasingly questions around cognition and consciousness. Looked at superficially, it could be mistaken for dispersion.

But this would be a misreading of how the notes actually function. They are not separate pieces of writing on unrelated subjects. They are iterative approaches to one central research problem. Each note returns, from a different angle, to the same core concerns: how meaning is formed, how it breaks down, how internal experience becomes external form, how AI participates in this process, and how care, memory, and subjectivity are reconfigured within it.

This is why the recent excursion into linguistics is not an arbitrary offshoot. It did not appear from nowhere simply because AI made it possible to generate a summary of the field. The question of language had already been active within the project for a long time. The role of words, prompts, inner speech, naming, recognition, and the threshold between language and image had all already been central concerns. The linguistics note emerged because that question had reached a point where it demanded a more explicit framework. In other words, the note was not a distraction from the project, but an extension of it.

This becomes important methodologically. In emerging or hybrid fields, especially those that do not yet possess a stable literature or a well-established disciplinary centre, inquiry cannot proceed through a single linear path. It must move outward and back again. It must test one concept through another field, then return to the project with that field reframed. What appears to be disciplinary wandering is often the necessary structure of original work.

The best way to describe this is as recursive inquiry. The movement is not linear accumulation, nor random browsing, but repeated circling around a central problem. Each return carries something new. Each traversal across disciplines slightly alters the understanding of the whole. Over time, this produces depth rather than fragmentation. The notes accumulate not as isolated units, but as a thickening network of relations around a single subject.

This is also where the longitudinal nature of the method matters. If one looked only at a single note in isolation, it might appear provisional or exploratory. But when read across time, the notes show a cumulative development of thought. Their function is not only archival. They are evidentiary. They show how an inquiry evolves. They record where misunderstandings were clarified, where new concepts entered, where a metaphor first appeared, where one field led into another. In that sense, they are not merely documentation of process. They are part of the process itself.

This long view also provides a partial answer to the worry about retention. Although individual notes may fade quickly from immediate memory, the sustained return to the same central subject means that nothing is truly isolated. Concepts reappear. Questions recur. Themes are taken up again from a different angle. The apparent overproduction is therefore held together by the continuity of the research itself. The subject remains constant even when the disciplinary route shifts.

10. Acceleration, AI-DHD, and Cognitive Integration

The speed of this method is one of its greatest strengths, but also one of its greatest problems. It enables a level of output, conceptual movement, and interdisciplinary reach that would be difficult to achieve through traditional means. At the same time, it introduces a very real risk: that production can outpace integration.

This is where the informal concept I once named AI-DHD becomes useful again. The term began half-jokingly, as an attempt to describe the condition of producing too much, too quickly, in a constant forward rush of generative possibility. But the joke contains a serious insight. The combination of AI-mediated acceleration and neurodivergent cognition can produce a state in which thought proliferates faster than it can be consolidated. New notes, new concepts, and new offshoots appear continuously. The excitement of inquiry becomes part of the engine of overproduction.

This is not simply a matter of distraction. It is a specific cognitive condition in which the system is capable of remarkable synthesis and movement, but where the pace of generation risks weakening retention. The six-hour session described earlier is an exact example. I was able to produce two substantial Fairy Notes, one on linguistics and one on the dialogue with my father, both of which were serious pieces of work. Yet very shortly afterwards I realised that I could barely remember the first one in any detail. The second, tied to a live personal exchange, remained far more vivid. This difference itself is revealing. What was emotionally and dialogically intense remained available. What was conceptually generated at speed was less well consolidated.

At first glance, this looks like a decisive flaw in the method. One might argue that the slower labour of reading and writing is precisely what allows knowledge to become embedded. This is often expressed through the language of struggle: that cognitive work happens through effort, friction, repetition, and delay. In that view, AI would appear to remove the very resistance through which learning becomes durable.

This critique cannot simply be dismissed. It names something real. There is a danger here. The speed of production can exceed the speed at which the mind absorbs what has been produced. The brain does not necessarily retain what the process can generate.

However, this is not the whole picture. First, the cognitive work has not disappeared. It has been redistributed. The effort is no longer concentrated so heavily in decoding, transcribing, spelling, and mechanically structuring. It is relocated into directing inquiry, evaluating responses, asking better questions, rejecting weak connections, and returning repeatedly to what matters. This is still cognitive labour, but it is a different form of labour.

Second, there is a difference between retaining every detail and retaining the deeper structure of an inquiry. While I may not be able to recall every note in detail, the repeated circling around the same research problem means that a cumulative understanding is nevertheless being formed. The notes do not remain disconnected. They continue to fold back into the same project. Over time, what is retained may be less the individual sentence or specific summary and more the architecture of the subject itself.

This is where my own project provides the strongest counter-example to the fear that AI necessarily produces only superficiality. Across a long period, I have generated an extensive archive of Fairy Notes and related writing, all centred around the same research questions. At times this felt dangerously fast and overproductive. Yet in retrospect, because the writing continually returned to the same project, it has produced a deep and increasingly differentiated understanding of the subject. The movement was not away from the research, but repeatedly back into it. In that sense, what appeared at one level to be proliferative disorder was, at another level, an iterative embedding of knowledge through return.

The right conclusion, then, is not that AI-mediated thinking is either cognitively superior or cognitively hollow. It is that it introduces a different relationship between production, retention, and integration. It may weaken some forms of memorisation while strengthening other forms of structural or relational understanding. The method therefore requires supplementary mechanisms of consolidation, such as oral articulation, teaching, presentation, and return. It cannot be treated as self-sufficient. But neither should it be dismissed simply because it does not follow the older rhythm of study.

11. Assessment and the Visibility of Thought

These methodological shifts have direct implications for academic assessment. If writing can now be generated, refined, or heavily scaffolded by AI, then the final written product alone can no longer function as a reliable proxy for understanding. This is not a minor disruption. It affects one of the central assumptions on which much of higher education has long depended.

The problem is not simply that students might cheat. Cheating has always existed, in one form or another, and it would be a mistake to build an entire educational model around the exceptional case of deliberate fraud. Most students are not trying to game the system. Most are trying, in varying degrees, to understand and to learn. The real issue is that the old system of assessment relied heavily on the visible polish of the final output as evidence of the invisible work that supposedly produced it. AI severs that link. A strong piece of prose no longer proves that the student fully understood what they submitted.

This means that assessment must shift from outcome alone to the visibility of process. In practical terms, this suggests a model closer to the workbook or process archive already familiar within art and practice-based disciplines. There, the final piece of work is not assessed in isolation. It is accompanied by evidence of development, experiment, decision-making, reflection, and change. The point is not only to show that something was made, but to show how understanding evolved.

Something similar could be extended more widely across academic disciplines. A student might submit:

• a sequence of research notes or developmental entries
• evidence of reading, questioning, and revision
• records of how AI was used and what prompts were given
• a final essay or project that synthesises that process
• an oral presentation, seminar, or viva-style defence that demonstrates live understanding

The advantage of such a model is not only that it makes cheating harder. It is that it rewards the thing academia should care about most: the development of thought. It makes visible the trajectory of inquiry rather than judging only the polish of the endpoint.

This is also where presentations become important. AI may assist in producing a structured paper or even a slide deck, but it cannot in itself guarantee that the student can explain, defend, and respond in real time. A presentation does not merely test memory. It tests articulation, structure, conceptual familiarity, and the ability to think under live conditions. Even if a student has rehearsed extensively, that rehearsal itself is part of learning. Memorisation has always been part of education. The issue is not whether memory is present, but whether understanding can be demonstrated through it.

For undergraduate teaching especially, this could prove transformative. Students could be asked not only to produce a final essay, but to maintain a structured workbook or sequence of exploratory notes showing how their understanding developed over the course of the module. In AI-rich conditions, such process documentation becomes even more important. It shows the chain of inquiry, the changes in argument, the points at which suggestions were accepted or rejected, and the student’s own developing position. A final essay would then appear not as an isolated performance, but as the condensation of a visible intellectual trajectory.

This is not merely a defensive proposal against AI. It is arguably a better educational model in general. It makes room for different cognitive styles. It values development rather than surface polish. It recognises that understanding is often iterative and uneven rather than immediately elegant. And it aligns more closely with what research actually is: not the smooth production of finished texts, but a messy, recursive, evolving process of finding out what one thinks.

In that sense, the visibility of process becomes not an optional supplement, but a central academic value. If AI has exposed the inadequacy of the final product as sole evidence of thought, then it may also have created the opportunity to build a more honest and more intellectually serious model of assessment.

12. Structural Change in Academic Practice

The broader implication of all this is that AI does not simply add another tool to existing academic practice. It changes the conditions under which academic practice takes place. That change is not confined to one type of student, one disciplinary area, or one level of education. It is structural.

This does not mean that everything about traditional academic work disappears. Writing remains important. Reading remains important. Highly literate researchers will continue to rely on their ability to move quickly through texts, to write fluently, and to produce complex arguments in conventional forms. But these capacities can no longer be treated as the only valid evidence of serious intellectual work. Nor can the older assessment structures that grew around them be treated as universally adequate.

The change therefore is not one of replacement but reconfiguration. Different modes of cognition and articulation will increasingly coexist:

• conventional text-based reading and writing
• AI-mediated dialogue and iterative development
• practice-based inquiry in which knowledge is produced through making, reflection, and return

However, coexistence here does not mean that nothing changes institutionally. It means that institutions must recognise a plurality of legitimate methods. The terms of assessment, supervision, and research training will need to shift accordingly. What counts as evidence of thought will become broader. Process, direction, selection, revision, and oral articulation will all become more visible and more important.

This also means being precise about levels. The implications are not identical for a senior specialist with decades of expertise, for a PhD student working in an emerging interdisciplinary field, or for an undergraduate learning to construct arguments for the first time. The institutional adjustments required will differ across these levels. But the underlying issue remains the same: writing alone can no longer bear the full weight of academic legitimacy.

That is why the argument developed in this note matters. It is not simply about an individual working method. It is about a transition in the structure of knowledge production itself. AI has revealed, with unusual force, that academic thought was never fully contained in the polished final text. It always involved process, iteration, failure, dialogue, and return. What is changing now is that these hidden elements are becoming impossible to ignore.

The consequence is not the collapse of academic standards, but the need to relocate them. Standards will increasingly need to be attached not only to outputs, but to the traceable development of thinking. This is not a lowering of rigour. It is a redistribution of where rigour is located.

Seen in this light, the current moment is not merely a threat to older academic forms. It is also an opportunity to make academic work more truthful about what thinking has always been: partial, recursive, social, and provisional. AI has not created that condition. It has made it more visible.

Reflection and Update on AI and Academic Writing: Rapid Synthesis, Slow Construction

The production of the preceding methodology note differed significantly from the rapid generation of the earlier Fairy Notes on linguistics and on dialogue with my father around consciousness. Those earlier notes emerged relatively quickly because they were anchored in a live conceptual thread and could be developed through direct synthesis. By contrast, this document drew together several pre-existing notes, multiple lines of argument, and a much higher degree of self-reflection. As a result, its production was slower, more recursive, and considerably more labour-intensive.

This difference is methodologically important. It suggests that AI does not simply make academic work easy, nor does it eliminate the struggle traditionally associated with writing and thinking. Rather, it redistributes where that struggle takes place. In the earlier notes, the main difficulty lay in access, translation, and synthesis. In this methodology note, the difficulty lay in proportion, emphasis, structure, and conceptual hierarchy. The labour returned not at the level of information retrieval, but at the level of judgement.

In this sense, the process of creating this document was much closer to writing in the traditional academic sense. It required repeated instruction, correction, rejection, and restructuring. The fact that I remained dissatisfied with multiple versions, and had to push the document toward greater balance and precision, is itself evidence that the thinking process remained fully active. AI did not resolve the problem for me. It became a demanding companion within it.

This is an important clarification of the broader methodological argument. AI can radically accelerate some forms of academic work, particularly where the task involves access to knowledge, explanation, and first-stage synthesis. But when the task becomes more conceptually self-reflexive, editorial, and structurally exacting, the process slows again. The struggle remains, but in a transformed form. Instead of being located primarily in literacy mechanics, it reappears in questions of judgement, coherence, emphasis, and intellectual integrity.

The distinction matters because it prevents the method from being described too simply, either as effortless productivity or as the outsourcing of thought. AI-assisted academic practice can encompass both acceleration and friction. In some cases, it enables breakthrough. In others, it produces a sustained and difficult editorial process. This document belongs to the second category, and for that reason it provides a useful counter-example within the larger argument of this note. It shows that the use of AI does not remove difficulty from academic thinking, but alters its location and its form

Certainly. Here is a short corrective footnote you can paste in at the end:

Footnote correction:
The contrast drawn earlier in this note between “traditional academic work” and AI-mediated dialogue risks overstating the linearity of pre-AI scholarship. Serious academic thought has never in fact been a simple movement from reading to writing, nor from private thought to finished prose. It has always involved drafting, redrafting, note-taking, return, comparison, revision, and development over time, often across many years. In this sense, academic thinking was already recursive, externalised in part, and cumulative. The distinction proposed here should therefore be understood more carefully: AI does not introduce iteration where none previously existed, but alters its pacing, density, visibility, and accessibility. What changes is not the existence of recursive thought, but the conditions under which it unfolds.

shorter version for supervisor

Absolutely. Here is a compressed supervisor version of the note, written as a coherent 1,500 to 2,000 word document rather than a collage of research notes. It keeps the strongest spine of your argument:

---

Thinking with AI: Dialogue, Process, and the Reconfiguration of Academic Method

(Supervisor version)

This note reflects on a methodological shift that became visible to me through a very specific experience. Over the course of roughly six hours, I produced two substantial Fairy Notes through sustained dialogue with AI. The first explored the question of whether humans think in words and led into a structured engagement with classical linguistics. The second emerged through a live text message exchange with my father, Dr William Marslen-Wilson, in which we discussed language, perception, and consciousness. What struck me afterwards was not simply the amount of material produced, but the speed, coherence, and depth of the thinking process itself. What would ordinarily have taken me far longer, especially because of dyslexia and ADHD, became possible within a single continuous exchange.

I do not want to overstate this. I am not claiming that AI replaces scholarship or that it magically removes the labour of thought. Nor do I want to caricature traditional academic work as if it had previously been linear, rigid, or unintelligent. Serious scholarship has always involved drafting, redrafting, note-taking, return, revision, and cumulative development over time. My father’s own career is an obvious example of that. What AI changes is not the existence of recursive thought, but its pacing, visibility, and accessibility. It alters the conditions under which thinking becomes externalised and available for further work.

The first important methodological point, then, is that AI-mediated dialogue should not be understood merely as a faster route to finished prose. In my experience, it is better understood as a different cognitive surface. Rather than thought remaining private until it is stabilised into writing, it becomes visible almost immediately. A question is posed, a response is generated, the response is corrected, expanded, or challenged, and the process continues iteratively. Knowledge does not simply precede articulation and then get written down. It emerges through articulation.

This has particular significance in relation to neurodivergence. For me, traditional academic work is not only demanding, but bottlenecked by reading and writing. Reading is slow and physically tiring. Writing is often slowed by the mechanics of spelling, typing, and structuring. None of this means that the underlying conceptual thinking is weak, but it does mean that the path from thought to academic form is more effortful and less direct. By contrast, AI-mediated dialogue works much more closely at the speed of my actual thinking. It is iterative, responsive, and curiosity-driven. Questions can be asked as they arise. Responses can be tested immediately. This does not eliminate labour, but it redistributes it. The effort shifts away from pure literacy mechanics and toward direction, judgement, selection, and conceptual refinement.

The two Fairy Notes produced in that six-hour period demonstrate this clearly. The first, on linguistics, behaved rather like an accelerated disciplinary essay. It gave me a rapid but structured entry into a field, allowing me to move from an intuition about language and thought into an academically grounded discussion of Saussure, Chomsky, Hockett, and Pinker. The second note behaved differently. It was not primarily a synthesis of existing literature, but a real-time translation of expert discourse. My father’s responses were concise, compressed, and shaped by a lifetime of research. Without mediation, I would either have had to spend days decoding what he meant or risk misunderstanding him entirely. Through AI dialogue, however, his statements were broken down into accessible conceptual steps, allowing me to ask more precise questions in return. This was particularly important when he clarified that “perception is integrated interpretation” and later suggested that what we call conscious experience is a dynamically varying, brain-wide constructed field. I was able to enter into that conversation meaningfully, despite not being a cognitive neuroscientist myself.

This is one of the most interesting methodological consequences of AI in research. It can function as a bridge between different levels of expertise, not by flattening the difference between them, but by making expert discourse more legible in real time. It does not remove the need for understanding, but it can help understanding happen faster.

To describe what AI is doing here, I have found it useful to think of it in three interrelated ways: as mirror, as archive, and as cognitive infrastructure. It acts as a mirror because it reflects thought back in a form that can be recognised, questioned, and reworked. It acts as an archive because it holds an immense amount of structured information that can be retrieved and reorganised quickly. But it is not a passive archive. It is what I have begun calling a thinking archive: one capable of provisional synthesis, comparison, and conceptual mapping. Yet, however capable it is, it does not know what matters in relation to the stakes of a specific project. That remains the role of the researcher. This is why I still find the figure of the librarian useful. The AI may be the thinking archive, but I remain the librarian. I decide what to foreground, what to reject, and what is significant.

This helps with the question of authorship. My experience is not that AI writes “for” me, but that it changes the surface on which thinking becomes legible. Intention, direction, hierarchy, and judgement remain mine. Indeed, the more ambitious and self-reflective the task becomes, the less the AI can simply “produce” it, and the more demanding it becomes as a companion. This was especially clear in the methodology note I later wrote about this very process. Creating the earlier linguistics and consciousness notes was relatively direct, because they were anchored in live conceptual questions and could be developed through immediate synthesis. Producing a methodology note about what had happened was much harder. It required me to pull together multiple existing Fairy Notes, clarify my own position, reject many structurally weak versions, and continually correct emphasis and proportion. In that sense, AI did not make the writing easy. It simply shifted the difficulty from access and explanation to editorial judgement and conceptual hierarchy.

This is why I think it is too simple either to celebrate AI as effortless productivity or to denounce it as outsourced thought. In practice, it can produce both acceleration and friction. Some forms of research become radically more accessible. Others remain slow, difficult, and highly dependent on sustained human direction.

This brings me to the tension I once half-jokingly named AI-DHD. One real danger of this method is overproduction. The speed of generation can outpace the speed of consolidation. Immediately after producing the two notes that seemed so striking, I realised I could barely remember the first one in detail. This raises a genuine concern: if the struggle of traditional reading and writing is reduced, is learning also reduced? I do not think this has a simple answer. However, my own longer project suggests that the issue is more subtle than simple forgetting. Over the past year and more, I have produced well over one hundred Fairy Notes, all in some way orbiting the same central project. At times this has felt too fast and too proliferative. Yet, looked at longitudinally, the notes are not random or scattered. They circle repeatedly around the same research problem from different disciplinary angles. The result is that, even when I do not retain every note in detail, I retain the deepening structure of the inquiry. Over time, the work converges rather than fragments.

This matters particularly in an emerging interdisciplinary field like my own, where no single stable literature yet exists. In such fields, knowledge may have to be built recursively, through repeated traversal across phenomenology, psychoanalysis, visual theory, linguistics, AI, and the lived realities of care and memory. What appears at one level to be movement across disciplines is in fact repeated return to a central problem.

This also connects to current institutional discussions about AI. William Huber argues that AI has so far been framed largely through industrial and productive logics, producing approximations of already recognisable forms rather than new cultural forms native to the technology itself. Russell Crawford argues that universities must resist premature certainty and remain willing to sustain uncertainty as a condition of inquiry. I find both arguments useful. They help explain why AI is so often received either as threat or as productivity tool, and why genuinely new methods may first emerge not at the level of finished output, but at the level of process. That is where I would place my own work. What may be new here is not yet a stable genre, but a method: dialogic, recursive, uncertain, and process-driven.

This has implications for academic assessment as well. If AI can help generate polished written work, then polished written work alone can no longer function as sufficient evidence of understanding. I think this points toward a stronger emphasis on process-based assessment: developmental notes, workbooks, records of research questions and shifts in argument, and oral articulation through presentations or viva-style exchange. In creative disciplines, this is already familiar. The final artefact is accompanied by process. There is no reason why more traditional academic subjects could not increasingly adopt similar models. Such an approach would not simply guard against misuse. It would more truthfully reflect how thought develops.

What, then, is my central claim? It is not that AI replaces scholarship, nor that it resolves authorship, nor that it defines a new cultural form in any settled sense. It is something more modest, but still significant: AI-mediated dialogue can function as a legitimate method of research and academic thinking, particularly in interdisciplinary and practice-based contexts, and especially where conventional academic structures have imposed uneven barriers to access. It accelerates some forms of work, complicates others, and redistributes cognitive labour rather than removing it. Most importantly, it makes visible aspects of inquiry that academic writing has often concealed: the recursive, provisional, dialogic, and processual nature of thought itself.

---

If you want, I can now turn this into a clean supervisor email attachment-style note, or make it slightly more formal and thesis-like.