When a novelist announces that a chapter was drafted with the help of a large language model, the reaction is rarely neutral. Some readers feel cheated, as though a machine's fluency were a kind of counterfeit; others shrug, arguing that a word processor's autocomplete or a thesaurus already mediates every sentence a writer produces, and that the new tools merely mediate more. Underneath this disagreement sits a question that copyright law, literary criticism and ordinary readers have answered differently for centuries: what, exactly, do we mean when we credit a person with authorship?
For most of literary history, authorship was inseparable from labour. To be an author was to have wrestled a text into being, sentence by sentence, and the marks of that labour, revision, false starts, a recognisable style built up over years, were taken as evidence of a mind at work. The Romantic idea of the solitary genius sharpened this picture further, treating the text as an emanation of an individual interior life, so singular that no one else could have produced it. Even when that picture was complicated by editors, ghostwriters and collaborative writers' rooms, the fiction of the single originating mind proved remarkably durable, because it gave readers someone to credit, blame and admire.
Generative language models unsettle this picture not by removing labour altogether but by relocating it. A writer who prompts, curates and revises a model's output is still labouring, but the labour looks different: less like inscription and more like editorial selection among possibilities the writer did not individually generate. Critics who dismiss such work as unauthored tend to assume that only the generation of raw text counts as authorship, and that selection, arrangement and revision are secondary activities. Yet this assumption sits awkwardly with how much of literary history already depended on exactly those secondary activities. Editors have long shaped manuscripts into publishable form; translators produce texts that are read as literature in their own right despite never having generated the original ideas; even conventional writers routinely borrow phrases, structures and plot devices from a shared cultural reservoir without forfeiting the title of author.
What may be genuinely new is not the presence of borrowed material but its scale and its source. A human writer's borrowings are refracted through an idiosyncratic biography, so that even derivative work carries a personal signature. A language model's output, by contrast, is drawn from a statistical compression of an enormous, largely anonymous corpus, and it carries no comparable interior history against which to read the text. This does not make the model an author in the way a person is an author, but it does mean that the writer who deploys it cannot simply claim the model's fluency as proof of their own accumulated craft. The honest position, then, is neither to dismiss AI-assisted writing as fraudulent nor to award it the same interpretive weight as work built entirely from a single hand, but to develop a vocabulary, still largely missing from criticism, for degrees and kinds of authorial contribution that existing categories were never built to describe.
Publishers and prize committees, forced to rule on individual cases well before critics have settled the theory, have mostly reached for disclosure as a stopgap: require writers to declare the extent of machine assistance and let readers decide how much weight to give the label. This is a reasonable interim measure, but it treats a philosophical question as though it were a matter of consumer protection, assuming that a simple percentage or checkbox can capture something as entangled as the relationship between a writer's intentions and a model's suggestions. The deeper task, of rebuilding our concept of authorship so that it can accommodate distributed, machine-assisted composition without either romanticising the old solitary genius or flattening every text into interchangeable output, remains largely undone.
For most of literary history, authorship was inseparable from labour. To be an author was to have wrestled a text into being, sentence by sentence, and the marks of that labour, revision, false starts, a recognisable style built up over years, were taken as evidence of a mind at work. The Romantic idea of the solitary genius sharpened this picture further, treating the text as an emanation of an individual interior life, so singular that no one else could have produced it. Even when that picture was complicated by editors, ghostwriters and collaborative writers' rooms, the fiction of the single originating mind proved remarkably durable, because it gave readers someone to credit, blame and admire.
Generative language models unsettle this picture not by removing labour altogether but by relocating it. A writer who prompts, curates and revises a model's output is still labouring, but the labour looks different: less like inscription and more like editorial selection among possibilities the writer did not individually generate. Critics who dismiss such work as unauthored tend to assume that only the generation of raw text counts as authorship, and that selection, arrangement and revision are secondary activities. Yet this assumption sits awkwardly with how much of literary history already depended on exactly those secondary activities. Editors have long shaped manuscripts into publishable form; translators produce texts that are read as literature in their own right despite never having generated the original ideas; even conventional writers routinely borrow phrases, structures and plot devices from a shared cultural reservoir without forfeiting the title of author.
What may be genuinely new is not the presence of borrowed material but its scale and its source. A human writer's borrowings are refracted through an idiosyncratic biography, so that even derivative work carries a personal signature. A language model's output, by contrast, is drawn from a statistical compression of an enormous, largely anonymous corpus, and it carries no comparable interior history against which to read the text. This does not make the model an author in the way a person is an author, but it does mean that the writer who deploys it cannot simply claim the model's fluency as proof of their own accumulated craft. The honest position, then, is neither to dismiss AI-assisted writing as fraudulent nor to award it the same interpretive weight as work built entirely from a single hand, but to develop a vocabulary, still largely missing from criticism, for degrees and kinds of authorial contribution that existing categories were never built to describe.
Publishers and prize committees, forced to rule on individual cases well before critics have settled the theory, have mostly reached for disclosure as a stopgap: require writers to declare the extent of machine assistance and let readers decide how much weight to give the label. This is a reasonable interim measure, but it treats a philosophical question as though it were a matter of consumer protection, assuming that a simple percentage or checkbox can capture something as entangled as the relationship between a writer's intentions and a model's suggestions. The deeper task, of rebuilding our concept of authorship so that it can accommodate distributed, machine-assisted composition without either romanticising the old solitary genius or flattening every text into interchangeable output, remains largely undone.
Question 1
Single correct option
According to the passage, the Romantic idea of the solitary genius was significant mainly because it:
- A. proved that only unassisted writing could be considered literature.
- B. gave readers a single identifiable figure to credit, blame and admire for a text.
- C. was the first theory of authorship to consider the writer's biography.
- D. showed that editors and translators had no legitimate claim to authorship.
Show answer and solution
Answer: (B) gave readers a single identifiable figure to credit, blame and admire for a text.
The passage says the Romantic picture 'sharpened' the idea of the author as an individual mind and that 'the fiction of the single originating mind proved remarkably durable, because it gave readers someone to credit, blame and admire.' Option 2 restates this directly. Option 1 overclaims (the passage never says unassisted writing is the only literature); option 3 is not stated; option 4 contradicts the passage, which says editors/ghostwriters/collaborators complicated but did not erase authorship.
Question 2
Single correct option
The author brings up editors, translators and conventional writers' borrowings primarily to:
- A. argue that all literary work is essentially collaborative and machine assistance changes nothing.
- B. question the assumption that only raw text-generation, and not selection or arrangement, counts as authorship.
- C. prove that translators deserve more critical recognition than they currently receive.
- D. show that language models perform exactly the same function as human editors.
Show answer and solution
Answer: (B) question the assumption that only raw text-generation, and not selection or arrangement, counts as authorship.
These examples follow directly from the sentence: 'Critics who dismiss such work as unauthored tend to assume that only the generation of raw text counts as authorship...Yet this assumption sits awkwardly with' these cases. The point is to challenge that narrow assumption, not to erase all distinctions (option 1 overstates) or to argue for translators' status as an end in itself (option 3), and the passage later explicitly distinguishes models from editors (option 4 is wrong).
Question 3
Single correct option
Which of the following best captures the distinction the author draws between a human writer's borrowings and a language model's output?
- A. Human borrowings are always original, while model output is always derivative.
- B. Human borrowings are filtered through a particular personal history, whereas model output is drawn from a vast, largely anonymous corpus with no comparable interior history.
- C. Human writers borrow far less material than language models do.
- D. Model output cannot be revised or curated by a human writer in a meaningful way.
Show answer and solution
Answer: (B) Human borrowings are filtered through a particular personal history, whereas model output is drawn from a vast, largely anonymous corpus with no comparable interior history.
The passage states human borrowings are 'refracted through an idiosyncratic biography' giving even derivative work 'a personal signature,' while a model's output 'carries no comparable interior history.' Option 1 is too absolute and unsupported; option 3 is a claim about quantity the passage does not make; option 4 contradicts the earlier point that the writer's curating and revising is itself labour.
Question 4
Single correct option
The author's attitude toward publishers' and prize committees' use of disclosure requirements is best described as:
- A. fully approving, since disclosure resolves the underlying philosophical question.
- B. dismissive, viewing disclosure as a cynical marketing exercise.
- C. cautiously accepting it as a practical stopgap while noting it does not address the deeper conceptual problem.
- D. hostile, arguing that disclosure requirements unfairly penalise AI-assisted writers.
Show answer and solution
Answer: (C) cautiously accepting it as a practical stopgap while noting it does not address the deeper conceptual problem.
The author calls disclosure 'a reasonable interim measure' but says it 'treats a philosophical question as though it were a matter of consumer protection' and that 'the deeper task...remains largely undone.' This is neither full approval (option 1, ruled out since the question is explicitly unresolved) nor dismissal or hostility (options 2 and 4), but qualified acceptance alongside a substantive reservation.
Saved on this device. Sign in or create a free account to keep it on every device.