worldbuilding with neural network Writing

Worldbuilding with Neural Networks

Worldbuilding is one of my least favorite activities, but it’s a prerequisite for a rich, detailed setting, which in turn is a prerequisite for a deep, creative story. I used to think about creativity as a magical process. I would try mystical things to achieve “inspiration”: staying up all night, stimulating myself with nicotine and caffeine, altering my perceptions with alcohol and other drugs, distracting myself with strange music, using randomness to prod my imagination, etc. Now I think of creativity as a practical, comprehensible, trainable process.


  1. The best stories seem inevitable. Effects follow causes, and the story barely scratches the surface of a world that feels deep, complete, and whole.
  2. In contrast, bad stories are thin and artificial. Characters exist only as collections of the exact traits and motivations the plot requires to proceed, and the world seems disconnected and mechanical.
  3. Deep, detailed worldbuilding is usually an exhausting process that kills my enthusiasm for the entire project.

Neural Networks

Neural networks have their root in biology. Our brains are literal neural networks, composed of a hundred billion neurons. I know the concept better from the field of computer science, where, using the brain as a model, programmers create networks of virtual neurons and use them to solve complex problems. The draw of neural networks is that each individual neuron is governed by simple rules and doesn’t need to understand the entire system to function. Voice recognition software like Apple’s Siri and photo comparison apps like Google Photos use neural networks that learn and adapt to their difficult tasks, requiring no human input. I’m no expert on the subject, so here’s a good Gizmodo article for people who want to dive deeper.

The coolest thing about a neural network is its ability to be more than the sum of its parts. While we still don’t know how a human brain creates consciousness, one theory is that any neural network of sufficient complexity and interconnectivity results in consciousness as an emergent property.

I’m going to be stretching the definition of a neural network as a metaphor, but the key components I want to take from this are:

  1. The individual neurons are simple and easily described and comprehended.
  2. Interactions between these simple neurons result in complex behavior.
  3. A network of sufficient size and complexity can result in emergent behavior that exceeds the sum of the inputs.

Enough pretending to be a scientist, although neural networks are a fascinating subject in their own right that have already birthed hundreds if not thousands of pieces of science fiction (although Google already has neural networks learning to write fiction, so it’s not all good news for writers).

The Metaphor

In this metaphor, the neurons are characters, places, even objects. The complex behaviors that result from their interactions are events. And if you get lucky, the emergent behavior is a vital, engrossing world. If you observe that world, you might end up with a great story – no marathon worldbuilding sessions required.

Start with one neuron: a single character. No name, no gender, no eye color, no likes and dislikes. Observe that character. What happens? Nothing. You don’t even know if this character is human, so he/she/it can’t even walk or talk.

Time to make something up. But there’s no need to wait for inspiration, or to write fifty pages of worldbuilding notes. Just choose some mundane things from your life experience. Human, female, named Barbara, right-handed, medium height, frizzy red hair. Job: bank teller. Now observe Barbara. There’s no trick here. All you’re doing is picturing the person you just invented, and imagining them. Maybe you imagine Barbara opening her front door, which looks a lot like your front door. Maybe you imagine Barbara passing money through a teller window at a bank that looks exactly like your bank. That’s fine. We don’t expect more from a network of a single neuron.

We could force a story at this point by just making something up. Let’s try it. Okay, Barbara…gets held up in a bank robbery, and after that powerless experience she’s…humiliated in the grocery store after work when she doesn’t have her wallet. Something inside her snaps, and she goes home and plots an elaborate campaign of revenge against everyone in town who’s humiliated her. As a twist ending, she saves the bank robber for last, but when we think she’s going to kill him, she thanks him for setting her on this path of self-actualization.

That’s a plot, at least for a short story, and it wouldn’t be hard to outline a dozen chapters by stapling together tropes and following some simple rules (are things too slow? put someone in danger, etc.). But that kind of paper-thin adventure-building is what I want to avoid. Whatever you might think of this off-the-cuff plot, it’s certainly not inspired. I know, because I made it up in thirty seconds as a thought exercise.

Worldbuilding For Inevitability

Inevitability is the metric by which I want to measure our story. Ignoring the triggering event of the bank robbery, how much of what follows is inevitable? Not very much. I just threw a couple simple ideas together, probably accidentally plagiarizing a Quentin Tarantino movie in the process, and invented some motivations after the fact.

Let’s take the next step in our worldbuilding exercise and add a second neuron: Ben, the bass player for a world-famous indie rock band. And a third, our setting: Eccleston, a small, poor, rural town nestled in the Appalachian mountains.

We’re nowhere near critical mass, but you can already see the complexity. We have two more neurons, but more importantly, we’ve gone from zero relationships to three (Ben – Barbara, Ben – Eccleston, Barbara – Eccleston). Now the complexity explodes: if we add Sarah, the young woman who runs a daycare and has a mysterious scar and past to match, we go from three to four neurons, but we jump from three relationships to six. Adding a fifth neuron brings us to ten relationships. A sixth, fifteen relationships.

But what’s the point? Here it is: good stories are built on relationships. You can fake it, by coming up with your plot first and fitting characters in later, but those stories feel strained and artificial. Your characters and their relationships should cause the events, and readers will notice when you cheat. If you follow your characters, everything you write will make sense. Doing your worldbuilding this way lets you stick close to the characters and see what they do.

Let’s pick a relationship: Ben – Eccleston. Ben’s a rock star. Why is he in a sleepy Appalachian town? I’m sure every single person reading this paragraph just came up with a totally different explanation. Maybe he’s on the run from the police after killing a security guard at a show. Maybe he’s searching for a secret wishing well that lets you summon a demon and trade your soul for one wish. Maybe he got stuck in town when his tour bus (individual tour busses for each band member, of course) broke down. None of this is revolutionary, but contrast this experience with trying to write a story for Barbara the single neuron. The relationships create questions (why would a rock star be in Eccleston?), which create answers in your mind, without even trying.

What about Sarah and Barbara? Maybe Sarah watches Barbara’s four year old son (look, another neuron!) at her daycare. That’s pretty mundane, so let’s keep worldbuilding. Barbara, arriving early one day to pick up her son Brandon (hey, a name! that was easy), saw Sarah through the window striking her son. Sarah denied it, and Brandon did too, and didn’t have a mark on him. Barbara pulled Brandon out of Sarah’s daycare anyway, and got drunk one night and complained loudly about Sarah at the local bar (wow, more neurons). Now Barbara is suspicious of Sarah, but also feels guilty and confused, and Sarah dislikes Barbara, and they avoid each other in public.

Mental Overhead

There’s a reason so many authors use the real world as their setting: it lets them skip the worldbuilding step. In a fantasy world, when a character needs to travel from one city to another, you have to figure out how people travel in your world, and how long it takes, and whether the fact that your character has to spend two weeks traveling by steampunk zeppelin ruins your plot’s timeline. In the real world, not only do you know how people travel, you’ve known it for years, and when your mind was making unconscious connection while you slept and worked and mowed the lawn, it just knew how long it takes to get from New York City to Boston, and took that for granted as it fit the rest of the story around it. You’d never come up with a premise that relies on someone getting from Los Angeles to Moscow in half an hour, because you know how the world works.

Writing in the real world reduces mental overhead – you’ve been doing the worldbuilding your entire life. You can just go ahead and write your scene about socialites at a rooftop Manhattan party. Maybe you’re worried that the party is wrong, because you’ve never been a socialite at a rooftop Manhattan party, but you’re not worried that the material the skyscraper is made of is actually too weak to support itself. You know the fundamental laws of your world.


This is my attempt to take the complex, beautiful, unsolved problem of creativity (and the drudgery of standard worldbuilding) and find a programmatic solution. I enjoy the crude blue-collar simplicity of it: if your world isn’t real enough, if your story isn’t alive enough, it’s just a problem of quantity! Pour in another load of neurons, let them bump into each other like tropical fish in a tank for long enough, and voila! a perfect novel.

It isn’t that simple, of course. The individual decisions (what is Barbara like? what is Ben like? why does Ben stand outside Barbara’s house every night whistling the theme from 28 Days Later?) take creativity, and maintaining the whole network still takes some overhead: remembering or recording each neuron and each relationship, keeping continuity, etc. But I think the worldbuilding you end up with will be relatively easy to remember because it’s vivid and inevitable. Sure, not everyone’s Barbara and Sarah will have the same relationship that mine do, but they will all feel real because we derived them.

Barbara doesn’t have to be your main character. She’s just a random starting point. You may find out that you don’t care about Barbara’s suspicion of Sarah, or that you do care, but that’s not the kind of story you want to write. You can let Barbara fade back into the background as you find new characters and new relationships you like more. As you swim through the sea of possibility, see what hooks you, and let it drag you up into the sunlight. And who knows? Maybe your new main character will decide to hold up the town’s bank at some point, and the teller might be a familiar face with frizzy red hair.