Imagine you read a short poem on a quiet morning. It has four lines. It describes a grandmother's kitchen after the power goes out, when the only light comes from a pot of rice steaming on a gas ring:
“The rice keeps steaming after the power is gone, and my grandmother counts the spoons by feel, the way she once counted us when we were small, so that no one is left in the dark.”
The last line makes you stop and read it again. You feel something close to grief, the good kind that makes you want to call a parent or a sibling. The poem seems to know what it is like to love someone who is getting older.
Well, what if no person wrote it? What if a language model produced it in a few seconds after someone typed a short request? No grandmother counted spoons. No child was counted. No kitchen existed, and no power went out in a particular year that someone still remembers. So ask what changed. The words are the same words. The rhythm is the same rhythm. Your eyes moved across the same lines in the same order. Yet something seems to have left the poem. You might feel a little cheated, or you might feel nothing at all and wonder why you ever felt anything. Either way, the poem is now a different kind of object. The real question is who was speaking when the poem spoke to you, and whether anyone was there to speak at all. Roland Barthes argued that the reader is the real place where a text finds its meaning.
Machine-written text seems to fit his argument perfectly, because it appears to have no author standing in the way. But I want to argue that this is a mistake. Barthes freed the reader from the author, but machine text removes the author without freeing anyone. A reader cannot judge a text whose intention belongs to nobody. The missing author does not give the reader power. It takes away the one party who could be asked to answer for what the text claims.
Roland Barthes published “The Death of the Author” in 1967. It is a short essay, and it is one of the most cited pieces of literary criticism of the twentieth century. Its target was a habit of reading that was common in schools and in journals. Critics would read a novel or a poem and then go looking for the author's life to explain it. If a poem seemed sad, the critic would find the poet's sad years. If a novel seemed to hide a secret, the critic would dig through letters and diaries for the secret's source. In this approach, the author was the key that opened the text. Once you knew the author's intention, you thought you knew what the text meant. Barthes thought this approach was wrong. He wrote that a text is made of many writings, drawn from many cultures and set in conversation with each other. Its meaning does not come from a single origin.
In his words, “the birth of the reader must be at the cost of the death of the Author.” This sentence is the center of the essay, and it is often quoted without its context. Barthes was not claiming that real writers do not exist or that their lives are useless. He was saying that the writer's intention should not be the final judge of what a text can mean. He also gave a positive idea in place of the old approach. He wrote that “the unity of a text is not in its origin but in its destination.” The text gathers its unity at the point where it is read. A reader brings memory, expectation, and mood to the page. Two readers of the same poem may find two different poems, and neither is wrong. For Barthes, this was a kind of freedom. The reader is no longer a passive student taking dictation from a master. The reader becomes an active maker of meaning.
Barthes also described writing itself in a way that matters here. He said that writing is “the destruction of every voice, of every point of origin.” The text is a surface where many sources meet and none of them governs the whole. He wanted critics to stop searching for a hidden owner behind the words and start paying attention to the words and to the reader who receives them. This is an idea worth taking seriously. It asks the critic to be humble about claiming to know what the writer meant. It also gives more room to readers who come from different backgrounds and see things the author never imagined. Barthes did not want to shut down interpretation. He wanted to open it.
The case for machine text begins with that same picture of writing as a surface of many sources. A large language model is trained on a huge body of human writing: books, articles, forum posts, letters, product descriptions, and poems. When it produces a new text, it draws on patterns from all of that material. The output is, in a plain sense, a tissue of quotations drawn from countless places. Barthes used that exact image for literature. A machine seems to be the purest case of it. The machine also seems to have no intention in the way a human writer does. When you read a poem written by a person, you can wonder whether the poet was sad, angry, bored, or trying to impress a teacher. Those questions matter to the old way of reading, and they can distract from the text. With machine text, those questions have no clear answer. There is no inner life to consult.
There is no biography to check. The reader is left alone with the words. If Barthes wanted the text to belong to the reader, machine text appears to hand it over completely. Consider how often this situation now arises in ordinary reading. A person searches for a recipe, a medical explainer, a product review, or a news summary and finds text that reads smoothly and sounds confident. There is no name at the top that the reader recognizes. There may be a site, a logo, or a byline that belongs to no real person. The reader is free to take what the text offers and fill the gap with their own sense of what it means.
The concrete examples are not hypothetical. In early 2023, the technology site CNET was found to have published dozens of financial explainer articles written with the help of an automated system. Several of those articles contained factual errors, and CNET later issued corrections on many of them. Later that year, Futurism reported that Sports Illustrated had published product reviews under author profiles with invented names and stock photos, where no real reviewer appeared to exist. In both cases, the texts were read and believed by ordinary people who had no way to find the person who had stood behind them. The texts were free of an author in one sense, and the readers were left without a way to check them in another.
Seen this way, machine text appears to be the completion of Barthes's project. It has no author to block the reader, and it offers the reader a field of meaning with nothing in the middle. The reader seems to gain the freedom that the critic in 1967 wanted to give them. This is why the case is attractive. It is the death of the author without any ceremony, and it looks like progress for the reader.
The French philosopher Michel Foucault answered a version of this argument only two years after Barthes. In a lecture given in 1969, later published as “What Is an Author?”, he agreed that the author should not be treated as a simple source of meaning. But he added something Barthes did not stress. He argued that the author is not only a person who writes. The author is what Foucault called an “author function,” a social role that controls how texts are classified, shared, and trusted.
Foucault gave a sharp description of this role. He wrote that “the author is the ideological figure by which one marks the manner in which we fear the proliferation of meaning.” In other words, a society that has many possible readings of a text uses the author's name to narrow them. The name tells us which texts belong together, which ones are serious, and which ones are dangerous or false. The author becomes “the principle of thrift in the proliferation of meaning,” as Foucault put it. Without some limit, meaning would spread in every direction, and no one would know what to believe.
Foucault's point is easy to see in ordinary life. A medical leaflet with no company name, no doctor, and no date would not be trusted, even if the words were clear and accurate. A scientific paper without authors or institutions would be hard to cite. An anonymous legal notice would be ignored by most people, and rightly so, because a law needs a source that can be identified. In each case, the name is not just decoration. It tells the reader what kind of text they hold and how much weight it should carry.
This means that readers still depend on something outside the text when they decide what it means and whether to believe it. They look for a name, a publisher, a purpose, or an institution. When the name is missing, they look for another way to place the text. The reader does not stand alone with the words. The reader stands with a set of assumptions about who would produce such a text and why.
Machine text does not remove this need. It only removes the name that would normally meet it. So the reader begins to look for a substitute. They may trust the platform that displayed the text, the website that published it, or the company that built the system. They may assume a human editor checked it, even when no editor did. They may trust the tone of the writing, which is often calm and sure of itself. These are weak replacements for an author, and they can be wrong. But they show that the author function is still working, even when no author exists.
Foucault therefore gives a sharper version of the problem. The death of the author did not end the social role the author played. It only left that role empty. When the role is empty, the reader must fill it with guesses, or they must accept whatever the machine offers. Here is my position. The missing author does not free the reader. It leaves the reader with no one to hold responsible for a claim.
Barthes's reader is a reader of pleasure and play. A poem can mean many things, and the reader can enjoy the variety. But many texts are not poems. A text can say that a medicine is safe, that a bridge was inspected, that a candidate took a bribe, or that a historical event happened on a certain date. These are claims about the world. A claim is a kind of promise. It says, in effect, that someone stands behind these words and can be asked why they are true. When a text makes a claim and no one stands behind it, the claim is still on the page, but the promise has nobody to keep it.
This is the difference between reading for pleasure and reading for trust. In a poem, a missing author costs us a little. We may enjoy the text anyway. In a news report or a medical explainer, a missing author costs much more. The reader cannot ask the writer what the source was. The reader cannot ask whether the writer checked the number. The reader cannot press the writer to correct a mistake or punish a lie. The reader can only accept the claim or reject it, usually with very little information.
Responsibility is not a cold bureaucratic idea. It is the mechanism that makes trust possible. A journalist who signs a story can be questioned, corrected, challenged in court, or dismissed for dishonesty. An editor who approves a story takes on part of that responsibility. A reader who sees the name of a reporter they have trusted for years has reasons to believe the story, because they know who would pay a price if it were false. Trust is built on a chain of people who can be held to account. Remove the people, and the chain breaks.
Machine-written text breaks that chain in two places. The first is the writer. The system that produced the words cannot be questioned in any meaningful way, and it cannot feel shame or accept blame. The second is the publisher. A publication that uses machine text may not disclose it, may not check it carefully, and may not own the errors in the way a human writer would. The CNET corrections show what happens next. The errors were fixed, but the reader who had already acted on a wrong explanation had no one to hold accountable for the harm. The fake reviewers in the Sports Illustrated case were not even a real person who could later say what they had tested.
This matters most in journalism. A news story is only useful if the reader can judge how it was made. Was the source a government report, a rumor, or a guess? Did a reporter speak to the people involved? Did an editor review the claim? A machine text can imitate the surface of a news story without any of these steps. Its fluent sentences look the same whether the facts are checked or invented. The reader is left to guess, and the guess is made harder by the very smoothness of the writing.
Education is another place where the missing author causes damage. When a teacher reads a student's essay, the essay is evidence of what the student understands. The teacher asks whether the argument is the student's own, whether the sources were read, and whether the student can defend the claims in conversation. If the essay was produced by a machine, the evidence is false. The teacher cannot judge what the student learned, and the student has not learned what the grade says they learned. The essay is a text with no one behind it who can be asked to explain it. In this case, the death of the author does not free the student. It takes away the process that education depends on.
Some will say that these are problems of bad actors, not of the text itself. A dishonest writer can always lie, whether a human or a machine. That is partly true. But the old system of authors, editors, and names gave readers a way to tell honest work from dishonest work. It was imperfect, and it failed often. Still, it gave readers something to check. Machine text, used without disclosure, removes that check from the reader's hands. It makes the lie harder to trace and easier to repeat.
So I do not think readers should celebrate the death of the author as a simple gift. Barthes was right that a reader makes meaning, and right that a text should not be closed by a single intention. But a reader who makes meaning from a claim about the world still needs to know where the claim came from. The reader should not be freed from this question. The reader should be given a harder duty, which is to ask who stands behind the words, what they would lose if the words were false, and how the claim can be checked. Barthes's reader cannot do this alone. They need institutions, names, and a clear line of responsibility, even if those things are imperfect.
Return now to the four-line poem about the grandmother and the rice. On the first reading, it moved me, and I think that reaction was real. The image of a family in the dark, held together by a counting that is almost a prayer, is powerful. Nothing in my argument changes that the lines work on the senses and the feelings. Barthes would say that the poem still belongs to me, and that my feeling of grief is the meaning it has for me now.
But the argument does change how I read the poem. When I learned that no one wrote it, the grandmother stopped being a person who once counted children in the dark. She became a character invented by a system that had learned how such images are usually built. The line “so that no one is left in the dark” stopped being a moment of love and became a pattern that moves a reader. I can still enjoy the pattern, and I may still feel something when I read it. What I can no longer do is treat the line as a sincere statement from someone who knows what it costs to keep a family together. There is no one to ask what the line meant to them, and no one who would stand behind it if it were false.
That is the loss the missing author leaves behind. Some pleasure may remain, and I do not think that pleasure is worthless. But trust does not move with the words. Trust needs a person, and a person needs to be answerable. When the author is gone and nothing takes the author's place, the reader is alone with a claim that no one will answer for. That is not freedom. It is a kind of abandonment, and the reader should see it clearly before accepting the words at face value.
