Is it possible to create a machine that will write literature, or merely duplicate its structure? When it comes to predicting, remembering and the sticking power of the human residue after each draft.
The Wound
A writer of whom I’m acquainted always has two pieces of paper open when she is writing. There are two, one is the manuscript. The other she considers “the wound”. The other she refers to as “the wound” which is a HIDDEN list of the grievances, half-memories and half-solved misunderstandings that she is dealing with in the story, but never mentions on the page. This second file shouldn’t be accessible to any editor, or to anyone reading the book. But she says that it’s essential for the novel to be empty.
In a nutshell, this is the best possible argument against the notion of a machine being able to compose literature. It is not the objection that the machine can’t put sentences together. The thing is, nobody is at stake with a machine: No life, No memory, No unfinished argument that presses against the page.
Fluency Without Experience
Nevertheless, they continue to pour out the sentences. Just give a large language model (LLM) a premise, and in seconds, it will churn out a decent, well-rounded short story featuring a beginning, a middle, an end, dialogue, which sounds like real people, and even a metaphor that works. This is one point that many writers never thought would be reached in their lifetime.
The first question, thus, has a simple answer. Is it possible to create text which resembles fiction using a machine? Clearly, it can. The more difficult question is whether resemblance equals literature, whether the distance between them equals technique, whether technique can be improved in a way that will bridge the gap, whether it’s a deeper distance.
Story as a Way of Thinking
There is always a mind and there is always a moment; literature is always the record of a certain mind at a certain moment. In The Language of the Night (1979), Ursula K. Le Guin maintained that storytelling is not frills on life, but one of the oldest instruments that the human mind created to gain insight into the world.
Understanding assumes a thing is not understood, a tool for understanding assumes that. The author starts out in uncertainty, and then goes about creating clarity in the language. That reaching style of literary style: the hesitant clause, the metaphor that is taken back, the sentence that is given a second thought by the author halfway through a thought. These attributes reflect mind and the here and now and a stake in the game.
Prediction, Not Confession.
In this sense, a language model doesn’t understand. It does one very complex thing: Predicting the most likely next word in the text, based on the given text up to that point. It is given a vast library of people’s writing until it gets more and more difficult to distinguish between the plausible and the eloquent. It is not a trivial procedure and one of the greatest engineering feats of the modern era.
Prediction does not constitute confession, however. The end of the bell jar, (or the final movement of The Fire Next Time), is not the most probable continuation of The preceding of either The Fire Next Time or The Bell Jar. They are the least safe and/or most expensive words they could have used, as they are the most expensive to use. There is never a cost to a model for a sentence. The one thing that really acutally holds it back as a literary writer is that.
An Unhelpful Object, Not a Bad Person
Of course this does not mean a writer can’t or shouldn’t use the technology, nor is proficiency a trick. Novelists have taken up the typewriter, the word processor and, now, the language model as tools of craft. They use them for testing out different cuts on a scene, see if the plot mechanism works, or to get the tone of a manuscript from one language to another quicker than any editor. The best is a model who is well read, who has a lack of self and is never tired.
Some novelists request these to be given them alternate endings that they do not wish to use. Next to the one rejected, the one they actually want is more apparent, like a photo negative showing the picture.
Solutions to the Problem of Smoothing
Technically, there’s a reason to suspect that a model can’t create style. A system trained to predict text is somewhat similar to a lossy copy: it retains the general pattern, and averages out the rest. The result is a photocopy of a photocopy, smooth and believable, but with some very small flaws.
It’s the details that make up style. A writer’s crackle, her odd rhythm, her unexpected silence and her most personal habit of phrase are just the outliers a predictive system is designed to filter out.
A Contract of Trust
Another issue is not one of authorship and but one of reader. The book has always been an agreement about two conscious entities. In reading a memoir, however, a lyric poem and even a profoundly autobiographical novel, we assume that someone in the real world lived something we were told about and selected these words to tell us about it. E. M. Forster expressed this in his novel Howards End (1910) in two words: Only connect. In his opinion, literature is a link between two lives within one person.
Take one end of that bridge away, and it could still be a bridge! However, what it means to cross it has been changed. That’s why many readers do not feel happy that a piece of writing was written by a machine, even if there is no change in any word. The reason is not due to appearance. They find that there was no one on the other side of the contract of trust and find out that it was never in effect.
Fear and Renewal
This should not cause us to demonise the technology and shift the blame away from literary culture. Every new technology that has affected the way people communicate, record and write, from the printing press to the typewriter, word processor, and internet, has raised the concern that the human voice would cease to exist. In each instance, there was some writing which was flattened, and there was some new writing which was not possible before.
This is the advice that Toni Morrison gave to writers, and it’s a fantastic response: If there’s a book you want to read that hasn’t been written, then you need to write it. But a machine which can yield a reasonable first draft does not affect that responsibility. It reinforces it perhaps even more, by making it evident which types of writing could have been replaced at all times, and which could not.
Conclusion
Hence, the answer to this question is ‘layered’. A machine can write sentences that, in a blind reading, can’t be distinguished from the work of the competent published prose. It can copy structure and voice; it can simulate “feeling” convincingly, for it has read a lot of humans feeling in words. It cannot yet tell from a mortal, particular and vulnerable life like a human could tell when something is at stake in the telling.
This separation might come to a close in time. It hasn’t gone away. Until then the wound is open and the writer will also have to sit with it.
The writer is a student of BS English Literature at Govt. Girls Degree College Turbat, Kech.

