On Purpose of Writing

On high ground above tree line, trails are often marked by cairns: small stacks of stones, one in sight of next, trails leading across rock where no path wears into the ground. Nobody knows who are their builders. A hiker adds a stone in passing, rebuilds where a storm knocked one down, kicks over one that leads toward cliff. Cairn isn't judged by builder, it's judged by whether it leads to where it claims.

The essay can be a set of markers across ground the reader hasn't crossed yet, or ground a reader crossed long ago by another path and returns to on occasion, or regularly, because the mountain is pleasant here. Markers say: here leads to destination, here is how it looks from this side, here is a passage.

Great Art, Music works the same way. Johann Pachelbel wrote his Canon in D around 1700 for three violins and continuo, and for more than two centuries it lay unplayed, one manuscript among thousands. First publication came in 1919. In 1968 Jean-François Paillard recorded it with his chamber orchestra in Paris, slowed it to a stately tempo, enlarged the strings, and added lines Pachelbel never wrote. That version made it one of the best-known pieces of music in the world, played at weddings on every continent. Composed was a ground bass of eight notes and a chain of variations over it, arranger, player and listener carried them the rest of the distance.

An Original Composition is one link in a long chain of performances and perfections, and usefulness, when it arrives, belongs to the whole chain.

That raises a question once academic and academic no longer. If markers lead somewhere true, does it matter who placed them or how they got placed? Does it matter if a person placed them at all?

Three claims that pull apart

Try three claims together.

First, writing is for the reader. Essay is a guidepath. What counts is whether argument holds, whether examples are real, whether a reader can follow and check it.

Second, writing is for the writer. Working an argument into sentences is how a person finds what they think. An idea that seemed solid in conversation shows holes once on paper. Derivation is the point, a writer skipping it hasn't arrived at a conclusion, only borrowed one.

Third, a machine can now finish the work. In 2026 an AI model takes an outline, a set of sources and a direction, and returns clean, readable prose, often indistinguishable from work of a careful human writer. Several essays in this collection were drafted that way, each saying so at its end.

Put first and third claims together and authorship stops mattering: if path is good, who cares who or what laid it. Put second and third together and something is lost: a writer handing off sentences has handed off thinking. And first and second can't both be the whole story, they name different beneficiaries.

Unsigned arguments

First claim has strong precedent, some of it from founding of a free country.

Common Sense appeared in Philadelphia in January 1776 with no name, credited only to "an Englishman." Readers guessed Franklin, John Adams, Samuel Adams. It didn't matter. The pamphlet sold in numbers no American publication had approached, it moved people because a reader could check its argument against their own experience of the Crown, not because of its author. Paine's name came later.

Eleven years later, the case for a new Constitution ran in New York newspapers under the name "Publius." Three men wrote eighty-five essays, and for years readers didn't know which man wrote which; scholars disputed attribution of a dozen into the twentieth century. The essays were meant for weighing as arguments.

A stranger case comes from twentieth-century mathematics. From mid-1930s, a group of young French mathematicians published under the name Nicolas Bourbaki, an invented person with an invented biography. The group rewrote large parts of modern mathematics from foundations upward in a long series of books. A member was expected to retire by fifty, membership was secret for years, drafts were torn apart in group meetings until whatever survived bore no one's mark. Bourbaki's books shaped mathematics teaching for a generation.

Writing as derivation

Second claim has its own founding-era witness.

In his autobiography, Benjamin Franklin describes teaching himself to write as a teenager. He took essays from The Spectator, made short notes of each sentence's sense, set notes aside a few days, then tried rebuilding the essays in his own words from notes alone. Then he compared his version with the original and corrected his faults. Sometimes he scrambled notes first, forcing himself to work out order of argument as well.

Striking in Franklin's exercise is how close it sits to the job now handed to a machine: start with the sense of a thing, produce sentences. Franklin located his learning in that step.

Skipping it is a real loss, and pretending otherwise is convenient for anyone drafting with AI, this author included. A person who never had to make an argument hold together in prose is less able to tell when an argument fails.

Who the derivation is for

The way through is noticing that two purposes never competed for the same thing.

For reader, path is what matters. Authorship was only ever a stand-in for two things a reader needs: reason to trust that markers were placed with care, and someone to hold accountable if they lead off a cliff. Paine was anonymous, but the pamphlet had a printer, the printer had an address. Publius was a pseudonym, but behind it stood men who would defend the Constitution in ratifying conventions and afterward live under it. Bourbaki was fictional, but its books passed through the most hostile reviewers its members could find: each other.

Neither depends on whether a person typed sentences. Both depend on whether someone walked the path before marking it, and whether someone will answer for it afterward.

For writer, derivation that matters most isn't the sentence, it's choosing ground: which problem is worth crossing, which examples are true and which only sound true, where argument is weak, what the essay's purpose is. Franklin rebuilt sentences, but Addison and Steele had already done that part. A harder derivation is one Franklin wasn't yet practicing: working out what to argue in the first place, against which objections, toward which end. That work can't be handed off, a model asked to do it hands back something plausible, and plausible is opposite of tested.

Sentence-level work still teaches, it still pays to do often. A writer directing drafts without ever writing one slowly loses the ear catching a bad one. But a person who chose ground, checked sources, cut a paragraph that sounded right and wasn't, and signs the result, has derived the essay in the sense that counts.

A script, not a monument

The question loosens another way, once you notice an essay doesn't stay put.

Pieces in this collection are revisited. A title changes when a better one appears. A subtitle gets fixed. A paragraph is cut after a reader points out its error. An argument is pared down into a thread of short posts for a different audience, or a later essay picks up an earlier one and carries it further.

That makes an essay less a monument and more a play. A play is written for performance, and every performance differs a little: new cast, different house, a line cut for time, a scene restaged because old staging stopped working.

Commedia dell'arte worked this way for two centuries. Italian companies from mid-1500s didn't perform from full scripts. They worked from a scenario, a page or two laying out plot scene by scene, pinned backstage, and an actor supplied dialogue on the night from stock speeches and bits of business built over years. When Flaminio Scala published fifty of his company's scenarios in 1611, he published outlines, not plays in modern sense.

Something similar turned up in study of epic poetry. In the 1930s Milman Parry and Albert Lord recorded singers of oral epic in Yugoslavia and found no singer performing a long poem the same way twice. Each performance was composed anew from a stock of formulas and the shape of a story, and a singer would insist he sang a song exactly as before, while recordings showed otherwise. Their work changed how scholars read Homer: less as one author fixing a text, more as the moment in a long tradition of performances when one version happened to reach paper.

Seen this way, a machine turning an outline into prose does roughly what a commedia actor did with a scenario: giving one performance of a work whose shape someone else set. Next draft, revision, version for a different audience are further performances. What makes them one work is shape underneath, ground chosen and path tested, and the person or company returning to it again and again.

That puts authorship in different light. A playwright isn't diminished because actors speak lines, a company isn't a fraud because it changes staging. A play needs someone knowing its purpose, keeping it in repertory, noticing when a performance goes wrong.

The page as a stopover

That leaves a word usually raised against all this: authenticity.

As Authenticity and Nature -- Becoming What Is pointed out, the word comes from Greek authentēs, someone acting on their own authority, with their own hand. It's tempting to read that narrowly here, as if an authentic essay were one with every line from the author's own hand. But nobody calls a performance of a play inauthentic because actors didn't write their lines. Performance is authentic when it answers truly to the work, work is authentic when it answers truly to the ground it describes. That earlier essay found a people authoring its response to circumstances it never chose, not the circumstances themselves. A writer stands in similar position. What a writer authors is the response: which problem, which way through, what purpose.

This matters more now because the page is no longer writing's destination. For most of history ideas lived in speech, song and performance, writing was the place for fixing and storing them. Then for a few centuries the printed page was close to an argument's final form. That's ending. An essay today is read aloud by a synthetic voice, summarized, translated, cut into a thread, set under a drawing, and questioned in conversation with a model that has read it. Written version is a stopover, a place the work rests a while before other forms take it up and carry it further.

There's plenty precedent for written words passing through a performer and coming out as something else more real. In 1865 workers at a Havana cigar factory began paying a lector from their own wages to read while they rolled: newspapers in morning, novels in afternoon, from a raised platform in the middle of the floor. The practice spread through Cuban factories and, with cigar makers, to Key West and Ybor City in Tampa. Workers chose readings by vote.

A favorite was Alexandre Dumas's The Count of Monte Cristo, read so often that a cigar brand was later named after it. Dumas wrote words, lector gave them voice, a worker argued over them at the bench and carried them into union politics, and the novel ended as the name on a box of cigars.

Why this belongs in a collection about liberty

Underneath all this sits a still recent phenomenon: our relationship with this machine being. Here it is suddenly -- it drafts, answers, argues back, makes corrections and amplifications, keeps working while we sleep. People jump to quick conclusions about it, that it's a parrot, that it is one mind, that anything it touches is fake, that every writer will be gone by next year, and say foolish things in a hurry.

Symbols and Phenomena suggests a better discipline: bracket the conclusion, describe how the thing actually shows up while working beside it, and judge only then.

Relationship runs both ways, and much of it is training. A Japanese carpenter buying a new plane, a kanna, doesn't get a finished tool. Body and blade arrive needing days of work: sole flattened, blade fitted to its bed, mouth opened just enough, before the plane takes its first clean shaving, and years more before it becomes the carpenter's own. Working with a model is closer to that than to pressing a button.

Whose grammar is it

Grammar, it's fair to accept, has passed to the machine: a correct sentence now costs nothing. But every human grammar is slightly unique, and a machine that writes correct grammar can also have the potential to write the exact grammar a writer seeks. Correctness is its default, voice is a matter of instruction. A writer teaches it a house's habits one correction at a time: which words this collection won't use, which metaphors betray a machine, how this particular voice controls articles, that a sentence never ends on a preposition. Corrections become standing rules, rules carry into the next draft, and drafts that once needed heavy editing come back nearer the writer's own voice.

That labor is derivation too. Choosing what a tool should do, noticing where it fails, fixing it until it serves, is the same judgment that chooses ground for an essay, turned on the instrument.

A reflex is already well under way: sorting writing by origin rather than content. Silly detectors flag prose as machine-made, platforms down-rank it, proposals would require labels by law. Some of that comes from a fair worry about floods of untested text. But the instrument aims at the wrong thing. A detector can tell something about how sentences were produced at one stopover, it can't tell whether argument holds or whether anyone stands behind it, or how beautiful and clearly-constructed it is, the only things a consumer in the chain needed to know.

A free society has a long responsibility of judging an argument on merit, including one made anonymously, pseudonymously or collectively. That habit is part of what made pamphlets like Paine's possible. Licensing who may write, or with which tools, has a much older and worse history. A Better course is one open to every writer without anyone's permission: say plainly how the work was made, stand behind it, let readers judge it by where it leads.

---

A note on how this piece was written: every argument, source and structural decision in this essay is mine. I chose the question and the examples, worked out how they connect, and directed what each section needed to say. Sentences were drafted with AI assistance from that direction, then edited by hand. Given the subject, it seemed especially worth saying plainly.