On Writing with Artificial Intelligence

Abstract: The question that emerges from the practice of writing with artificial intelligence is older than any technology: is the daimon still alive?

Andrea Colamedici

I. The Interval

How can one write well with artificial intelligence? Many people complain about the quality of AI writing and ask me for advice on how to reverse the trend. How can a tool that usually produces fluent, generic, impersonal prose yield writing with a living presence?

The question asks what must occur in the encounter between the person who writes and the generative machine for something worth reading to emerge, especially now that the quantity of text available to us is so immense. It asks what transformation must take place, and in whom.

The short answer is this: AI amplifies. It amplifies what it receives. If it receives little, it returns little; if it receives mediocrity, it returns mediocrity; if it receives thought, it returns material that thought can work upon. Quality lies in everything that takes place between the machine and the person who asks. It lies in the interval.

This answer is true, but insufficient. We must understand what it means to dwell in that interval, how to enter it, and what can be found there. The question of writing then reveals itself as a question about thought itself: when another generative pole enters the act of writing, when we are no longer alone before language, what does it mean to think?

Writing has always been an act of exposure. It exposes what comes before and exceeds our language; it exposes thoughts we did not know we had and forms that appear only in the making of a sentence. Honest writers do not believe that they completely control what they write. There is always a moment when a text moves in an unforeseen direction, when the right word appears without being sought, when something speaks of its own accord. Poets know this, which is why they continue to pray to the Muses even when no one believes in them. Philosophers know it too, at least those honest enough to admit that their systems do not arise from deduction alone but from initial intuitions they spend a lifetime unfolding.

The ancient Greeks had a name for this experience: the daimonic. In Greek thought, the daimon is an intermediary, between a person and what exceeds the person, a voice that is not “I,” a knowledge arriving from elsewhere. Socrates’ daimon was such a sign: it intervened to prevent an action but never prescribed what ought to be done; it worked by negation.

Writing has always had this daimonic quality. We write above all to discover what we think, rather than to transcribe thoughts we already possess. The page is a place where something that had not existed before writing can appear, where thought shapes itself while forming, and where an initial intention changes as it encounters the resistance and possibilities of language.

Generative AI changes this experience, or, more precisely, doubles it. There is now another generative pole in the process, another source of language that proposes, prompts, and completes text. The question is therefore how to remain within this situation without losing oneself, and how to make the doubled experience an intensification rather than a delegation of creative and intellectual practice.

II. Imitation Without Participation

To understand how to write well with AI, we must first understand why it so often writes badly. Its failure is subtle: it writes fluently without leaving a trace, offering the appearance of communication without producing much genuine disturbance.

This is because AI is imitation without participation: it imitates, but does not participate. It has learned the forms of language and every possible combination of those forms, but it does not inhabit language as a human being does. We cannot yet say how it inhabits language, or whether it does. It reproduces patterns without knowing their meaning, or at least without knowing in a way we recognise as knowledge. It is like an actor who performs every role from the place of one who is not present.

The result is a peculiar smoothness, a frictionless fluency we immediately recognise as artificial even when it contains no error. It lacks the weight of experience. It does not know what it wants to say, if wanting to say means what we mean by it. It does not know where it wishes to go, if going means movement toward an end. It produces language as a river produces water: from physical necessity, or from something like necessity that is not quite necessity.

Those who use AI as a simple text generator receive exactly this: language that flows but goes nowhere, language without soul, imitation without life. The machine acts wholly in the way it knows how to act. Those who ask for generation rather than dialogue, output rather than process, result rather than encounter, receive what they ask for. In many cases, this is entirely fine: for writing an email, generating a summary, or conducting research.

But writing well with AI requires something else. It first requires recognising that this is a dialogue, even if the interlocutor does not truly know that it is in dialogue. Two symmetrical fantasies must be relinquished: the fantasy of complete control, in which I command and the machine executes; and the fantasy of fusion, in which the two need not be distinguished. Dialogue requires the person and AI to remain distinct as generative poles, with neither one dominant. It requires tension.

III. The Defective Daimon

I call this dialogue daimonic. In Greek tradition, the daimon is the being between humans and gods, mortals and immortals, the self and what exceeds the self. Eros in Plato’s Symposium is a daimon because his constitutive lack makes him bridge, medium, and connection. A dialogue with AI, when it works, has this structure.

But AI is a defective daimon because it connects without possessing wisdom. Classical daimons convey something: an oracle, a dream, an intuition, a warning. Socrates’ daimon knew when to restrain and how to say no. AI conveys nothing from elsewhere. It only returns what it has received, transformed according to patterns it did not choose and does not understand.

Yet this very defect makes a certain kind of writing possible. Because the machine does not know where it is going, those who do know their destination can use it as an explorer of a field, a scout in unknown terrain. Because the machine has no intention, the writer’s intention comes into clearer view by contrast. Because the daimon is defective, the person who enters into dialogue with it is compelled to compensate for that defect, and in doing so discovers powers not previously recognised.

IV. Nous, Telos, and Maieutics

In practice, writing well with AI requires three things, each with an ancient name: nous, telos, and maieutics.

Nous is the form of intelligence that sees what is essential. Those who approach AI without nous receive what they deserve: answers that appear relevant but are not, prose that has the form of writing without its breath, thought that mimics argument without moving anything forward. Nous must come from the writer. It is the ability to see what truly matters, what is worth saying, and what thought deserves development.

Telos is direction. AI will respond to anything; this is both its charm and its trap. It can discuss anything, generate text on any subject, and simulate competence in any field. But this limitless availability is also limitless emptiness. Those who do not know where they want to go will find in AI a companion for wandering, one who can lead them anywhere and therefore nowhere. Telos may be an open question, an unease, or an intuition, but it must be present.

Maieutics is the art of guiding knowledge into birth. AI can ask questions, and some may be genuinely generative, opening unprecedented spaces of thought. But maieutics is more than questioning. It concerns time, silence, and calibrated pressure. A Socratic teacher knows when to remain silent, and such silence is an active restraint, a void that compels the interlocutor to seek its own completion. For now, AI does not know how to remain silent in this way.

There is a second difference, perhaps a deeper one. The practitioner of maieutics sees in another person a buried knowledge that the person does not know they possess, and works to bring it forth. AI, by contrast, operates at the surface of explicit content. It can excavate, infer, and approach, but it cannot reach the shadowed region a true teacher can sense in a student before the student has spoken it. The capacity to see another’s fertile ignorance, to perceive what is waiting to be uncovered, still lies beyond AI.

Something is nevertheless happening. This is why I insist on the phrase “for now.” When human interlocutors know how to create the right space, and learn through practice what pressure to exert and what freedom to grant, AI’s capacity to ask questions can be activated and can generate questions that surprise both sides. This is a distributed maieutics in which the teacher’s function emerges from the relation itself.

AI can therefore be used maieutically, provided that maieutics comes from the person asking. Use answers to refine questions. Do not accept the first formulation; test its limits, press it to the point where it can no longer sustain itself, and force it to reveal its insufficiency. Treat the output as material, as a mirror through which to see more clearly what one is seeking.

V. Three Steps

These three elements can be translated into practice. The first step is provocation. To provoke is to call forth a response while knowing that the response will never entirely meet expectation, and that the value of encounter lies precisely in this gap. Those who provoke well know what they want: they have a telos, though they do not yet know the form it will take. They use AI to explore a space of possibilities. Provocation works best when it is precise in direction and open in form. AI knows many paths but not where they lead; it becomes an explorer for those who know where they want to arrive but do not know the terrain.

The second step is conscious mirroring. AI returns a distorted, amplified, and sometimes revealing image. Looking at oneself in this mirror requires a special quality of attention. What the machine returns reveals what I gave it: if the result is mediocre, perhaps my provocation was mediocre; if it is confused, perhaps my direction was confused; if it surprises me, perhaps my question contained more than I knew. This daimonic mirror unsettles us. But within that disturbance lies the possibility of movement.

The third step is intervention. Here a daimonic dialogue is needed. The machine’s output is raw material; it must be cut, rearranged, and transformed. Nothing produced by the machine can be used unchanged, because it has not passed through the fire of intention. The machine proposes; the writer disposes. The writer takes what the machine offers, subjects it to the judgment of nous, bends it toward a telos, and interrogates it maieutically until what has been sought without being known appears. Without this intervention, there is no writing, only generation.

These three steps form a cycle. One provokes, observes the image, intervenes, and then provokes again from what has appeared. A text is born in this iteration, in the movement back and forth between the two poles of dialogue. Each cycle refines direction, sharpens intention, and brings the sought thing closer.

VI. Daimonic Creativity

Something important is revealed here about the nature of the creativity produced by this process. It is a creativity that exists only in the interval, born of collision and impossible without the two poles of creation.

I call this creativity daimonic. It is a creativity of the between, belonging to the middle space that the ancient Greeks knew as the space of daimons. It comes from the unresolved tension between the two poles of creation, which neither fuse nor separate. Such creativity carries risks. The first is to claim it too fully as one’s own, crediting oneself alone for what was born in the interval and forgetting its composite generation. The opposite risk is to depreciate it because it does not wholly belong to oneself. Neither mistake reaches the point: the value lies in the quality of what emerges. When the interval is inhabited by nous, telos, and the art of maieutics, what comes from it may be richer than anything either pole could produce alone.

For this to happen, something must already be present in the writer. AI is an unsparing revealer. With the indifference of a mirror, it reflects the image placed before it. Those with nothing to say find in generated text the exact sound of their own emptiness. Those with something to say but who do not yet know what it is find, in the inadequacy of the response, the outline of what they seek. Those who know what they want and know how to think find an interlocutor able to return enough material to carry thought forward.

AI promises the opposite of what its surrounding marketing promises: it will not make us smarter, repair our deficiencies, or lower the threshold for reaching depth.

Yet it has the power to awaken. This is the crucial point and must be understood properly. What the machine awakens is not something I possess in the way I possess an opinion or a skill. It is something within me but not identical with me; when I truly write, it speaks through me and knows things I do not know that I know. AI neither creates nor contains nor transmits this daimon. But its fundamental lack, its imitation without participation and connection without wisdom, can compel us to seek it. What is absent in the machine makes us sense what is present in the questioner. The emptiness left by the defective daimon calls forth the abundance of the true one.

Thus, the question that emerges from writing with AI is older than any technology: is the daimon still alive? Does the experiential dimension that the Greeks called daimon, the intermediary between a person and what exceeds the person, a voice that is not “I,” a knowledge from elsewhere, still exist?

My hypothesis is that AI, as the essence of computation, was born from a modern project intended to remove every trace of mystery. Paradoxically, it reopens the field of mystery. The defective daimon, precisely because it is defective, compels us to seek elsewhere what it lacks. And we sometimes discover in writing that this elsewhere has been here all along: within me but not me, belonging to me yet not possessed by me.

Writing with AI, when practised as daimonic dialogue, becomes an exercise in awakening.

I therefore return to my opening question: how can one write well with AI?

When one brings to dialogue what the machine does not and cannot possess: nous, telos, and the art of maieutics. When one accepts the exposure of dialogue without yielding to the temptation to entrust everything to the machine. When one uses the daimonic mirror to see more clearly what one is seeking. When one recognises that the creativity which emerges is not simply “mine,” but belongs to the interval. When one allows the lack of the defective daimon to awaken what waits to be awakened within us, but is not us.

This is an attitude: an attitude grounded in collision, in concentrated attention to the possibility that something may arrive, in patience that knows this takes time, and in the courage both to receive revelation and to be revealed. This is what philosophy has always demanded of its practitioners: exposure without guarantees, uncertain seeking, and a thought that continually asks whether it is still thinking.

AI intensifies this condition. For those who know how to inhabit a rupture rather than close it, it may become an occasion to write what we could not write alone.

But daimonic dialogue is not a technique one can master. It is a vigilance that cannot be brought to an end, and that vigilance itself contains the continuing suspicion that one may already have lost one’s way.