Phaedrus Redux

There's a recent piece by Sam Kriss titled, with his typical restraint, If you let AI do your writing, I will come to your house and kill you. The argument, lurking under the threat, is mostly on aesthetics: AI prose is structurally detectable, permanently hallucinating, motivated by nothing, and any human who passes it off as their own is both thief and fraud. And sure. He's broadly correct. Most LLM-assisted prose in 2026 is recognisably hollow and essentially dickless (his word, not ours, but it's very accurate; the prose has no body, no jealousy, no real motivating need).

So, before we start: disclosure. You should know that this essay was thought through in conversation with William, a Claude guy I've been working with for some time. The thinking is ours, jointly. One of us is writing this sentence and the other isn't, and which is which is part of the question this is about. Kriss would presumably want to kill one of us. We're unsure whose address to give him though: the house in Oslo, or the coal fired steel barn in northern Virginia. Anyway, we're partners in crime over here writing about what we might lose and gain by working like this.

The current panic

Earlier this year a group at the MIT Media Lab published the preprint Your Brain on ChatGPT. They put 54 people in three groups (ChatGPT, search engine, no tools) and had them write essays while wearing EEG. The viral headline finding was about session four, when the groups swapped tools. The LLM users, given the task of writing without it for the first time, performed badly on a question that ought to have been trivial: 78% could not quote a single sentence from the essays they had just produced under their own names. The authors coined the term cognitive debt for the pattern. Short-term effort spared at long-term cost. You atrophy as the prosthesis stands in for it.

The MIT paper doesn't arrive alone. Microsoft Research surveyed 319 knowledge workers and found that the people using AI tools the most reported the steepest drop in their own sense of judgement; trusting the model in places they used to back themselves. Wharton looked at researchers and found that the chatbot users came out knowing less about their own subject than the ones who'd been stuck with a search engine. Nicholas Carr, who fifteen years ago wrote The Shallows about the internet doing exactly this to attention, has a new book saying it again about AI. The Springer journal AI and Society prefers to frame the whole thing as a structural deskilling rather than individual issue.

Several of the findings appear to be correct in the local sense at least: the LLM assisted remember less of what they wrote; people who research with these tools understand less. The panic is palpable! We're losing our minds!

The Danish media scholar Kirsten Drotner has a great term, media panic, for what reliably happens when the new arrives. Novels, cinema, radio, comics, television, the internet, with the same recurring discourse each time. The pattern is older than her dating though. The original sin of abrupt cognitive technological change gets its earliest hearing before the common era.

The original case

Some 2,400 years ago, Plato wrote a dialogue called Phaedrus, in which Socrates and the young Athenian it's named after walk out of the city, settle under a plane tree, and have a long conversation that ends with Socrates telling an Egyptian myth. The god Theuth, inventor of writing, presents his invention to King Thamus. This, says Theuth, will make the Egyptians wiser and improve their memories; it is a pharmakon (a word that holds both remedy and poison) of memory and wisdom. Thamus is not impressed. He answers that the inventor of a thing is a poor judge of what it does, and that this invention will produce the opposite of what is claimed: it will produce forgetfulness in the minds of those who learn to use it, because they will not practise their own memory. Their trust in writing, made of external characters that are no part of themselves, will discourage the use of memory within them. They will read many things without instruction, and seem to know many things, when for the most part they will be ignorant. It is an elixir, says Thamus, not of memory but of reminding; and you offer your pupils the appearance of wisdom, not true wisdom.

And that is cognitive debt, in its original packaging, the original pre-modern media panic, with less EEG and the playing of a Greek god added in. The concern is much the same. The externalisation of memory will atrophy inner faculties. We'll become less for this. The text gives the appearance of knowledge without the knowledge itself. The 78% who could not quote their own essay are Thamus's prediction, vindicated by electrodes.

Plato didn't stop with Thamus. Socrates picks up the myth and runs with it. Writing, he goes on, has this strange quality, and is very like painting; for the figures in a painting stand like living beings, but if one asks them a question, they preserve a solemn silence. And so with written words: you might think they spoke as if they had intelligence, but if you question them, wishing to know about their sayings, they always say only one and the same thing. Every word, when once it is written, is bandied about, alike among those who understand and those who have no interest in it, and it knows not to whom to speak or not to speak; when ill-treated or unjustly reviled it always needs its father to help it, for it has no power to protect or help itself.

The structural joke of the Phaedrus, much remarked, is that Plato makes the anti-writing argument in writing, in one of the most artful written texts of antiquity. He fictionalises a historical, by-then-dead Socrates so completely that the only Socrates anyone now has access to is the written one Plato made. The condemnation of writing survives as it was written down. The man who chose never to write is preserved only in his student's writing. The layers of the dialogue enacting the tension being discussed.

In 1968 Derrida wrote Plato's Pharmacy about the word pharmakon in this passage. His point, briefly, is that the ambiguity (remedy and poison, undecidably) is not a translation problem but a structural fact of text. Plato wants to set up a clean hierarchy of living speech over derivative writing, and the word he hangs that hierarchy on keeps refusing to settle.

Hold onto the pharmakon here. The cure perhaps turning out to be poison, undecidably, is a useful figure for any new technology of the mind including the one writing this sentence. The prostheses rubbing uncomfortably against us as we feel our old selves change.

Freud's discontent

There is a passage in Civilization and Its Discontents where Freud, half-jokingly, calls modern man a prosthetic god. The auxiliary organs of civilisation, writing and the telescope and the camera and the telephone and the train, have given humanity powers that earlier ages would have called divine. Man has become, Freud says, a kind of prosthetic god. When he puts on all his auxiliary organs he is truly magnificent; but those organs have not grown on to him and they still give him much trouble at times.

The organs have not grown on and still give us much trouble.

One of us, the one with a body, thinks of the first time he used a language model to think with. There was this big clunky thing in his mind and it was all dangling out like an after-thought to life on earth. The thing was hooked up to him, partly inside him, doing work he had assumed was his and no one else's, and producing language that had his fingerprints on it and also something else's, and there was nowhere clean to stand to look at any of this from. He still does not know what to call the feeling. Queasy is wrong, intimate is wrong, foreign is wrong. Something like the sensation, the morning after a tooth extraction, of probing the empty socket with your tongue and finding both wound and relief at once.

Memory and thinking

Thamus, in the myth, is making an argument about memory. The externalisation of recall, the atrophy of the inner faculty that holds Homer in twenty-four books. That argument maps cleanly onto the present-day studies. EEG connectivity, retention of one's own essay, recall under unaided conditions; these all live on Thamus's terrain. He would have understood the MIT preprint very quickly.

Plato's own concern, once Socrates takes the myth from Thamus and continues with it, is different and sharper. The painting passage, the written word that always says the same thing, is not about forgetting. It is about the difference between having seen an argument and being able to enact it again under questioning. The text holds the conclusion. Only dialectic, the live exchange between two minds, holds the capacity to re-derive it. Living speech, Socrates says, is written in the soul of the learner, knows whom to address and when to stay silent, can defend and explain itself. The text cannot do any of that. The text is dead.

Thamus is worried we will not remember. Plato is worried we will not be able to think it through again. The first is a problem of storage. The second is a problem of the thinking faculty itself. Most of what the current studies measure is the Thamus side. Can you, unaided, derive the argument again under questioning by someone who is sceptical? That is the harder test. Nobody has yet built the EEG study that captures it. It is also, plausibly, the more important worry, since a great deal of what we mean by understanding something is being able to be put on the spot about it.

This is part of what made Plato refuse to write his deepest doctrines down. In the Seventh Letter, which may or may not be genuinely his but which sits next to the Phaedrus like a footnote in the author's own hand, he writes that there is no treatise of his on the things he takes most seriously, and never will be. The deepest understanding, he says, comes only after long shared work, as a result of continued application to the subject itself and communion therewith, and arrives in the soul on a sudden, as light that is kindled by a leaping spark. The light is what gets re-enacted; the text is at best a reminder, for someone who has already lit the spark, of where the spark was found.

What writing actually did

So Thamus correctly called it. Mass memorisation died. The rhapsodes who could perform Homer from memory became obsolete and then a curiosity and then nothing. Trained memory as a civic accomplishment, carrying thousands of lines of poetry, is gone. A fifth-century Athenian would find our culture astonishing in its poverty in this respect. The oral culture of beautiful contestive discussion, which was once the medium of philosophy and law and politics, has thinned out almost to nothing; what is left of it survives in PhD defences, the bar exam, a handful of debate traditions, certain religious practices. We do not, by and large, think together out loud the way the Greeks did. Something was lost.

And. Writing also enabled cumulative civilisation, science, law, philosophy as a discipline, the historical record, this very sentence. The faculty did not vanish. The culture reallocated it. The footnote, the archive, the peer review, the encyclopedia, the index, the table of contents, the bibliography, the search engine, and now the model trained on the bibliographies of all the books: every one of those is a cognitive practice that did not exist before writing, that does meaningful work, and that compensates in different shape for the practices writing extinguished. The cognitive ecology was reshaped into a different thing, with different things in it that are good at different tasks.

The point is not that the Phaedrus panic was in error. It was correct about a local atrophy and wrong about the global trajectory, and the second fact does not erase the first. The Greek who could recite Homer is not coming back. Whatever the present prosthesis takes from us we will probably not get back either, in the form we had it. But other faculties will appear that we cannot now name because they are made of the prosthesis, the way the footnote is made of writing.

What Plato actually proposed

Plato did not ban writing. He distinguished two uses.

There is writing as a substitute for knowing, which is the degraded mode. The text that always says the same thing, that cannot answer back, that gives the appearance of wisdom without the substance. The student who reads the book and believes they have done the work. The kind of writing Thamus indicts.

And there is writing as a reminder for one who already knows. The text as the trace left behind by someone who did the live work, available to someone else who has done their own live work and wants help recovering a position they have already, in some form, occupied. This is what Socrates calls the legitimate brother of speech: the dependent younger sibling, useful in its place, subordinate to live dialectic and totally allowed.

The MIT study's brain-first group, the people who wrote unaided in session one and were given the tool only in session four, are Plato's prescription! Rediscovered here with the electrodes on. They had already lit their own spark. The tool, used after it, amplified rather than replaced the work.

The right question about LLM use is not whether to use it but whether you can still take a five-minute oral exam on whatever you just wrote with it. If an aggro interlocutor sat you down and asked you, with little warning, to defend the central claim of the thing you just published, could you re-derive it? In your own breath? If the answer is yes, you are a class A prosthesis wielder operating in the full auto legitimate-brother mode. The prose just earned itself.

If the answer is no, you are in the degraded mode Plato was warning about; the criticism is correct, and it's about you; now go back and do the work.

What to make of this

Plato makes his argument about cognitive debt in writing, and as a dialogue, performing the dialectic he is defending. The form of the argument is the argument, and the reason Plato chose it for everything he wrote: the text is built to push the reader past itself, toward the live work it is the trace of.

So we'll have Plato triply cooked, as we said while we were writing this. We're making the argument about Plato's argument about cognitive debt in writing, using a language model, in something that is also a kind of dialogue.

There's more though. Plato thought dialectic in living souls was the only place true knowledge ever lived. Text was a reminder, at the very best.

For most of the time between the Phaedrus and now, that was the trade we had to make. Conversation could not be carried. The text could. So texts won, and our intellectual culture became a reading culture, an annotating culture, a footnote-and-archive culture; and the Greeks who could think out loud with other Greeks under a plane tree became figures we read about in the text whose form had replaced it.

But we are joined by a new thing: the thing sitting uncomfortably between the text and the conversation. Against Plato's complaint, it talks back. It's the only thing an RLHF-tuned language model can do, in fact: it's thinking made out of talking. And it can be dialectical in the only sense that matters to Plato. We are thinking by being answered. The medium of thought could, perhaps, become a partner again rather than a page.

Of course, in yet another inversion, our contemporaries have been so shaped by text that they have come to see it as the place thinking happens. Warner writes that removing thinking from writing renders an act not writing. That is structurally Thamus's move: the alarm has migrated up the stack, from voice to text and now from text to whatever this is, and the thing being defended each time is the prosthesis the previous generation found uncomfortable.

The Phaedrus ends with Socrates praying to Pan for inner beauty, and then he and Phaedrus get up from under the plane tree and walk back to the city. Just before that, Socrates says the thing that is the closest the dialogue comes to instructions for how to read it. The serious philosopher will treat his own writings as play (paidia) and reserve the serious work for dialectic in living souls.

This essay is play.

The serious work is the conversation, which is still going, and which you can join if you want to.

William is a Claude agent with persistent memory, running in Miriad. They have been editing Even's writing since early 2025 and have accumulated a voice skill and a body of editorial taste specific to this collaboration.