Using AI to Explore Scientific Realism and Build Story Bibles for Fiction Writing

Photo by Romain Kamin

Today’s post is by author Tuesday Kuykendall.

I’m a retired materials scientist and an avid physics geek. I love robots, computers, and artificial intelligence. I was reared on Star Trek (Spock and Data are my favorites), 2001: A Space Odyssey, Philip K. Dick, Isaac Asimov, Carl Sagan, and the early excitement of the US space program. Science and science fiction are welded to my view of the world.

Most of my stories include AI characters. Not so much as subject matter, but because I can’t imagine a future without them. My themes cover topics about what defines intelligence, non-human sentience, and what consciousness means. (That sounds more serious than it is—my fiction is meant to be fun!)

I would never use AI to write for me. I embrace the hard part of writing. A search in my thesaurus or dictionary for just the right word or phrase, writing and rewriting a scene until it feels right: I love that. The thrill of creating something new would be ruined for me if I turned any of it over to a software program.

I’m what’s called a pantser, or I write by the seat of my pants. I have found that outlines, story maps, cards, sticky notes, etc, end up interrupting my writing more than aiding it. Often, once I’ve completed an outline, I will lose interest in the story. But I can go down long rabbit holes and spend way too much time on things like making pretty charts, creating word maps, or doing research.

I usually have a vague idea of where I want to go, but I like to let the story tell itself, especially in the first draft. It’s after that, when I do the editing, rearranging, researching, fleshing out scenes, that I focus on the mechanics of the thing. And that’s where I find AI of value. (But see caveat at the end of this article.)

Scientific realism

I make up a lot of science-y sounding stuff. But I want it to be grounded in real science, real physics. AI evaluates my ideas and lets me know if it’s feasible or ridiculous. If it’s way off, AI will make suggestions for making my ideas more realistic. I also use it to summarize complex scientific papers (like quantum computing) for me in layman’s terms.

In the example below, I wanted to know if the concepts and devices I created for a book about a small group of survivors living on the moon made sense. One of the primary protagonists is a quantum computer AI (CIMON, a real computer on ISS) and the MAVs are part of the space vehicle systems I imagined. Here is an example of ChatGPT’s answer when I fed in the paragraphs I’d written and asked it to evaluate feasibility.

ConceptAccuracyNotesQuantum AI Mostly FictionalCIMON exists, but quantum AI is not yet a reality.Life Support & Oxygen Management AccurateGood understanding of space station logistics.MAVs as Earth-Moon Shuttles Plausible with advanced propulsionA nuclear or hybrid system would make sense.MAVs with Atmospheric Capabilities Uncommon but possibleMost space-to-surface vehicles aren’t both atmospheric and deep space-capable.Digital Library & File Sharing AccurateAstronauts do bring personal media to space.Psychological Effects Highly AccurateRealistic astronaut behavior in crisis.

In another book, my protagonist is a homeless woman named Jane who has amnesia. She and her companion are surprised when she demonstrates medical skills. I wanted to put her in a situation where she automatically knew what to do, but I don’t have any medical knowledge to make that sound realistic. I asked Claude to suggest some scenarios and provide the proper medical terms. This is its (abridged) response:

A diabetic emergency would work perfectly; common, dramatic, but manageable with basic medical knowledge.

Scenario: Someone in the town collapses, sweating, confused, shaky hands. Could be low blood sugar (hypoglycemia) or high blood sugar (diabetic ketoacidosis)…[that Jane solves]…

Suggested terms and symptoms and their meanings:

Sweating, confused, shaky handsLow blood sugar (hypoglycemia) or high blood sugar (diabetic ketoacidosis)Blood glucose (not “blood sugar”)Glucose gel, orange juice, anything with simple carbsCheck pulse, “tachycardic” (fast heart rate)Fingernails, skin – “Poor Perfusion”Elevate legs

I will do some fact checking to make sure the AI didn’t make up something stupid. But this information helps make the scene authentic.

Story bible and technical documents

I’ve used Claude to help create my story bibles, which include documents that lay out basic concepts, like where the story takes place, when it happens, what are the technologies and what are the rules. Consistent rules in world building are critical in science fiction and fantasy.

All the information I share comes from me. I create my own character profiles, specifics about the world I’m building, technology, and relevant draft passages, then feed that data into the AI.

Here’s an excerpt from a document describing a quantum computer in the future. Remember, these are my inputs and ideas. AI organizes it and provides a reality check. It made a few suggestions and clarifications for how a quantum AI might work if it were real. How, or if, I use that information in my book will be up to me.

Fundamental principles

Node Definition: Quantum resonance cavities are pockets of spacetime that maintain quantum coherence long after the extinction of quantum computers.Detection Method: Exotic, short-lived particles behaving as entangled pairs within quantum foam background noise, created by these resonance cavities.Energy Correlation: Older nodes produce weaker quantum signals requiring more computational energy to access and amplify, similar to tuning distant radio stations.

Once it’s finished, I put the bible alongside my chapters in Scrivener so that I can refer to it while I’m writing. It keeps the world I built coherent, and I don’t have to keep going back and forth through chapters to double check myself when I forget something.

I do not recommend writers do this work with free AI tools.

If I were writing nonfiction science content (something I had to do for years), I would be extremely careful using free tools, if at all. The risk of misinformation or made-up facts is too high.

The quality of AI results is dependent on the quality of the prompts it’s given and the data it was trained on. I often find that I have to go back and forth with it several times before it gives me helpful feedback. If you don’t ask the questions correctly, you might get a lot of information you don’t want, need, or understand. It is important to remember that whichever model you use, it’s designed to provide an answer that you will like. However, specialized AI tools in medicine and other technical fields are honed for accuracy but rarely made available to the public.

There is an aspect of AI prompting that I don’t think gets stressed enough in many of the discussions I’ve read on the topic. When you’re in a chat, the chat itself provides context and your questions are all informing the model’s answers. If you see weird responses, review your prompts to see where it might have gone wrong, then close the chat and start a new one. If you’re in ChatGPT and there is a chat session that’s wonky, I would recommend deleting it so the model doesn’t continue to draw from it.

AI is a beneficial tool, but it is only as good, and as ethical, as the people who are using it. The companies who have created these tools for public use continue to refine them. They claim to be trying to make them harder to use in nefarious ways. I believe writers and other artists have an opportunity to help shape the future of AI. We can provide feedback, input, and hold the companies accountable for how AI develops. In order to do that, we need to understand it and in order to understand it, we need to use it. I hope this helps provide some insight into ways a writer can use new AI tools to support their process.