I’m a speculative fiction writer, and I hang out with speculative fiction writers. I think that we are in an important position in our culture: we can think deeply and seriously about issues in society and technology, while we also also have the freedom to speculate and project thoughts into the future or onto alternate worlds. When we have something to say, we have unique opportunities and methods by which to effectively speak.

A lot of writers have strong opinions about artificial intelligence: they want to talk, think, and write effectively about it. That isn’t the same as a desire to use it. Writers in the 80s who never touched a computer still wrote intelligently and incisively about how computers could affect society. They wrote for audiences who knew those technologies, though, the same way other readers know horses, cars, medicine, or guns. Writers who eschewed cell phones still wrote for readers who no longer accepted “the protagonist wasn’t at home” as an obvious obstacle to reaching them by phone. Writers today are likewise writing stories about characters encountering or creating AI, for readers who will be thrown out of a story that gets technical details wrong.

At the same time, many of the writers I know have come to instinctually recoil from anyone who really wants to talk about AI, because the conversation is becoming exhausting.

It seems like learning about AI ought to be easy. It’s hard to spend more than an hour online these days without something about AI flashing across your eyeballs: You have to use it! You must never use it! It’s going to uplift humanity, cure cancer, and you’ll never have to work again! It’ll destroy society and the environment, put everyone out of work, and maybe literally kill you. You will be told in all confidence that it’s critically important technology, and then in equal confidence that it’s all a scam. And that’s all aside from the deluge of computer-written posts veering from saccharine to enraging, and faintly Ghibliesque cartoons in mostly poor taste. Very little of that deluge tells you anything useful.

I’m a fiction writer, but I’m also an engineer. I have a PhD in machine learning, and I’ve been following developments in AI and machine learning for a long time now. I’ve been on in-person and online panels asking “what the heck is this?” and while those panels are well-attended, there’s never enough time to cover enough of the topic to provide any understanding. I’ve had plenty of more in-depth conversations over drinks, but those only go so far. So, I’m introducing this series of blog posts explaining the current technology underlying chatbots and AI coding systems. I’m not going to argue for or against using them in real life, I just want to explain the technology well enough that writers can incorporate that understanding into the plots of their (human-written) stories.

I want to give writers enough understanding, concepts, and vocabulary to write their stories and to effectively do their own research so that whatever they have to say, gets across as clearly as possible.

While programming these networks requires math (if surprisingly little), I’m going to assume very little math knowledge, and I’ll dispense with it altogether where I can. I want to give you the language to describe the math; I won’t expect you to do any. Conversely, that means that if you are comfortable with math, some of my descriptions will feel elementary; I’ll do my best to keep that to a minimum, and provide references for you to dig into.

Where these posts are going to diverge more from the usual explainer is that I intend to make space in each of them to talk about how to translate the material into speculative fiction. For example, what a neural network might look like in a fantasy world, or the more social aspects of how your scientist characters might go about doing and communicating their work.

I might change my plans depending on feedback, but I’m planning six posts:

• What is a neural network / deep learning?
• What does it mean to train a model?
• What are tokens and embeddings?
• What specifically is a large language model?
• How does a large language model become a chatbot?
• How does a large language model become an agent?

What I’m not going to talk about:

• How and whether to use AI (especially in writing fiction)
• What the impacts of AI on society are
• The legality or morality of how particular systems were built or used

It’s not that these topics aren’t important. It’s that the entire point of these posts is to give readers the means to express their own beliefs as effectively as possible, and I don’t think I can do that if I’m also pushing my own opinions, or if the comments get filled up with arguments.

That said, given the topic, I’m going to say this one thing up front and then that’s it: If I’m going to ask human beings to spend their time and read these posts, then I feel the need to write every word of them myself, except where I need to show machine output (explicitly labeled) to illustrate a point. That goes likewise for artwork. There may be mistakes and poor phrasing and weird drawings, but they will be mine. That is not advice or a recommendation or judgment. If you take my personal approach as some kind of criticism or endorsement of how you personally do things: that’s on you, pal.

See you in the first post, on neural networks!