The unsettling future is not the one where a machine writes a better novel than you do.
It is the one where the machine finishes your novel before you have discovered what you were trying to say.
In “The Dark Night of Mathematics,” Kirwin Hampshire imagines a demonic Master Librarian watching an author begin a story in private. The instant the first words appear, the Librarian reads the writer’s mind, completes the story in a million different ways, and uses an oracle to select the best version for publication.
The thought experiment is meant to describe a spiritual crisis in mathematics. If every worthwhile theorem could be produced instantly, what would happen to the human experience of discovery? Hampshire turns to creative writing because writers recognize the same wound immediately. The terror is not simply that the machine produces something excellent. The terror is that it arrives before the author has had the experience of finding it.
“The author still writes. But why the hell would we do this?”
Kirwin Hampshire, “The Dark Night of Mathematics,” July 25, 2026
That question deserves more than the usual answer about human taste, human soul, or human authenticity. The author probably does keep writing. The harder question is what kind of writing remains possible when every direction can be completed before the writer has traveled far enough to understand why one direction matters.
The nightmare is an answer without a journey
Writers rarely begin with a fully articulated meaning waiting to be expressed. They begin with a disturbance. An image will not leave them alone. A character behaves in a way they do not yet understand. Two people enter a relationship that seems to contain more conflict than the premise can explain. An ending appears before its cause. A sentence carries a charge that the rest of the story has not earned yet.
Writing is how they follow that disturbance until it begins to make sense. The writer follows one choice until it exposes another, discovers that a compelling scene belongs to a different story, and realizes that the argument underneath the work has been organizing itself all along. Meaning becomes visible through the act of trying to express it.
This is where the Master Librarian becomes truly demonic. Its million endings do not merely compete with the writer’s ending. They collapse the distance between intuition and understanding. The author is handed an immense field of finished expression before the unfinished story has had time to ask anything of them.
Abundance without orientation turns discovery into selection. The writer moves from “What is this story?” to “Which of these answers do I like?” Those questions can sound similar from the outside, but they produce very different creative experiences. One forms judgment through contact with the material. The other asks judgment to arrive after the possibilities have already been packaged.
Research into AI-assisted creative writing has started to find evidence of that difference. In a 2025 study of 88 university writers, participants wrote poems about meaningful experiences and then asked generative AI to write about those same events. They rated the AI poems more highly for imagery, language, and form, yet felt significantly greater ownership over their own poems and judged them more faithful to the experiences they were trying to express. The machine could improve the visible artifact while weakening the writer’s sense that the artifact carried their encounter with the material. (Justin Nicholes, Written Communication)
That single study does not settle the future of authorship. It does reveal the fault line. Surface quality and creative ownership can move in different directions.
The Storyform is discovered on the way
Dramatica gives us a precise language for what the writer may be discovering.
A Storyform is the story’s structural argument: the arrangement of conflict across the Objective Story, Main Character, Influence Character, and Relationship Story Throughlines. It describes how those Perspectives work together, where conflict arises within each one, how they progress, and what the story ultimately demonstrates through its resolution.
That definition can make a Storyform sound like something the writer ought to know before writing the first page. Sometimes they do. More often, the Storyform begins as a hypothesis hiding inside the material. The writer senses parts of it long before they can name the whole.
They may know that the Main Character cannot give in, but have no idea yet whether that persistence reflects a Steadfast Resolve or a defense the story will eventually break. They may know how a relationship ends while remaining uncertain about what is changing the relationship from within. They may have four beautiful sequences and only later recognize that each one has been circling a different phase of the same conflict.
Dramatica helps the writer recognize those relationships as they emerge. It can reveal that a personal crisis has been mistaken for the Objective Story, that an Influence Character has plenty of personality but never truly challenges the Main Character’s way of seeing the problem, or that two possible endings complete entirely different arguments. The model gives the writer something more useful than a verdict. It gives them a way to understand what their choices are doing.
This is why the phrase Narrative First still matters. It does not mean deciding the meaning in advance and forcing every creative impulse to obey it. It means keeping narrative meaning in view as it develops. The images, scenes, characters, and accidents of composition remain alive, while the writer gains a language for noticing the argument taking shape beneath them.
The journey is still the writing. Dramatica makes the terrain legible.
A thousand endings on a rail of meaning
Now return to the oracle.
Suppose a generative system offers a thousand endings. Presented as a wall of polished alternatives, they become Hampshire’s nightmare: an abundance so complete that the writer’s own movement through the story begins to feel irrelevant. Even choosing among them becomes exhausting because the system has supplied expression without consequence.
The experience changes when those endings sit on a rail of meaning.
One ending may give the Objective Story a Success while leaving the Main Character personally unresolved. Another may produce Failure and Good, allowing the external effort to collapse while the Main Character resolves what mattered internally. One ending may require the Main Character to Change. Another may reveal that the character’s refusal to abandon their way of seeing the problem was the source of their strength all along.
Those endings may share locations, characters, and even dialogue. Structurally, they are different stories. Each one makes a different claim about what resolves the inequity, which costs matter, and whether the effort was worthwhile.
A Dramatica-guided system can communicate those differences. It can tell the writer that one possibility intensifies the Relationship Story while starving the Objective Story of a conclusion. It can show that a satisfying reversal quietly changes the Story Goal, or that a powerful sacrifice implies a different Story Judgment than the writer has been developing. It can trace how a choice made in the final scene reaches backward through the Signposts and changes what the audience understands the whole journey to have meant.
That kind of assistance does not remove discovery. It creates more surfaces on which discovery can occur. The writer can try a possibility, see its consequences, reject the system’s interpretation, revise an earlier assumption, and gradually recognize the Storyform that feels native to the work.
The rail does not choose the destination. It lets the writer feel where each destination leads.
The oracle needs to explain itself
This points toward a more demanding standard for creative AI.
Generating possibilities is the easy part. A serious story-development system should be able to explain how those possibilities alter the story’s meaning. It should distinguish a cosmetic variation from a structural alternative. It should recognize when several versions are simply different expressions of the same Storyform and when one small change creates an entirely different argument.
Most importantly, it should know when the writer is still discovering.
Research with professional writers suggests why that boundary matters. A 2025 CHI study found that writers used generative AI in sharply different ways: some expanded their workflows through brainstorming and alternatives, while others delegated difficult tasks central to their professional identity. The researchers warned that handing over those identity-bearing tasks could contribute to skill erosion, and proposed more structured, process-oriented guidance instead of systems that deliver final answers in one move. (Rama Adithya Varanasi, Batia Mishan Wiesenfeld, and Oded Nov, “AI Rivalry as a Craft”)
For Narrova, that suggests a clear creative responsibility. It should generate when generation helps the writer inquire, compare, test, or move. It should slow down when fluent output would hide an unresolved structural question. And when it offers a direction, it should expose the consequences clearly enough for the writer to argue back.
The ideal exchange is not “Here are ten endings. Pick one.” It sounds more like this: “These endings fall into three different arguments. Here is what each one asks the Main Character to prove. Here is what each one does to the Relationship Story. Here is the version your current material appears to be moving toward, and here is the evidence. Which argument feels closest to the story you are discovering?”
The system brings structure, memory, alternatives, and the patience to examine consequences at scale. The writer brings instinct, lived experience, taste, resistance, and the authority to decide when an interpretation has missed the living center of the work. Neither half of that relationship needs to pretend that the Storyform was fully known at the beginning.
It is discovered through conversation, drafting, rejection, recognition, and return.
Hampshire’s Master Librarian is horrifying because it knows the ending before the author has lived the question. A narrative intelligence worthy of writers would behave differently. It would open paths without pretending they were destinations. It would make the underlying meanings visible without claiming one as inevitable. It would help the writer understand why one possibility belongs and another merely shines.
A thousand endings do not have to become a nightmare. They become useful when the writer can see the story each one creates.
And then continue the journey themselves.
Sources
- Kirwin Hampshire, “The Dark Night of Mathematics”
- Justin Nicholes, “Ownership, Accuracy, and Aesthetics: University Writers’ Perceptions of GenAI Poetry”, Written Communication, 2025
- Rama Adithya Varanasi, Batia Mishan Wiesenfeld, and Oded Nov, “AI Rivalry as a Craft: How Resisting and Embracing Generative AI Are Reshaping the Writing Profession”, CHI 2025