On writing in the age of generative AI

I have a few conflicting thoughts on this, so bear with me. If you’re an author, don’t start ringing the alarm bells in your head and getting all revved up against a fellow author saying anything vaguely supportive about generative AI. The world is a complex system and unfortunately our primitive black-and-white thinking-and-feeling brain apparatuses often fail to grasp nuances of grey.

I use my primitive brain apparatus to write books, but generative AI has become part of my research and editing process for three main tasks: brainstorming, summarising information, and checking grammar. For these tasks, it has mostly replaced Google (although I still use Google if I’m looking for named sources). For brainstorming purposes, AI is usually too vanilla, impersonal, and uncreative, but sometimes the suggestions can still spark a few (better) ideas. It has proven more useful for summarising knowledge on a widespread topic, or to check if I’m using the right word or grammar in a short excerpt, preventing me from getting stuck in language conundrums. However, I avoid checking grammar for large chunks of text, as I don’t want to give away my IP and help big tech monopolies make even more money.

Now, I don’t expect AI to help me with writing novels. Why? Mostly, because AI still sucks at creative tasks. I’m talking about attempting prompts such as “write a scene with character X, taking place in place Y, and where Z happens”.

(Want a quick example? I’ve done it here.)

Generative AI sucks at creative tasks for one (or more) of three reasons: 1. The result is too bland, over the top, or clichéd (“Across it, the hills of the other side shimmered in the weak morning light, promising something cleaner, maybe freer.” - WTF?). 2. The details don’t match the vision for the story or writing style, and they take longer to refine and adapt than if I were to write the scene from scratch (a case in point - the prompt I used above is similar to a scene from The Seed Vault, but my character would never say something as ridiculous as “You make it sound like freedom is just a ferry away"). 3. The result is actually okay but the process defeats the purpose.

Take some time to notice the last reason: the result is actually okay but the process defeats the purpose. Sometimes, AI does come up with a snippet that I could have written myself, and which works for a specific scene. To use the example above, I don’t totally dislike the line “She half-smiled, but it didn’t reach her eyes. The duffel bag at her feet might as well have been chained to the concrete.” I mean, it’s not brilliant, but in a novel with 80k words would it be that bad? Not all lines can be pearls.

And yet, what would be the point of asking AI tools to write scenes or chapters for me? The reason I write is because I enjoy coming up with the ideas myself. Writing feels cathartic, energising, and fulfilling. Outsourcing the creative process to an LLM would defeat the purpose of writing. This is my personal opinion as an author, but unfortunately I don’t have control over other authors’ and readers’ preferences, neither how the market will evolve.

What will happen when AI becomes much better at creative writing tasks?

Here’s an interesting comparison: photography. In the old times, you needed a photosensitive surface, a small box with a hole, and a darkroom with the right chemicals. Nowadays, you only need a mobile phone with a camera. Is photography less of an art, or less of an expressive outlet because the technology evolved? I don’t think so. But the creativity and technical mastery were displaced from knowing the right exposure times and darkroom tricks to understanding camera settings. The “good eye” for composition, the time invested, and the motivation to take great pictures remained. Photography also became more widely available. (I’m simplifying the concept to its bare bones - there is still a lot you need to learn to become a good photographer, but you get what I mean.)

Let’s apply the same logic to creative writing and AI. What will it mean to be an author in 10, 20, or 50 years, once AI becomes better at certain tasks? How will creativity be displaced, to account for this new democratisation in writing tools?

I believe a writer will definitely mean something different from what it means today. The creativity will move from plucking the right words out of your brain to a slightly different skill. And it won’t be about writing good AI prompts - bloody hell no. Prompts are dreadful! They kill the magic of storytelling and the playfulness of language, by turning everything into a set of instructions. No, it will be something else. The new skill will require authors to mix the right combo of memories, experience, judgement, and sensibility. Is that so bad? Maybe I’m being naïve, but I don’t believe readers will want to read fiction generated exclusively by AI. They will still seek a sliver of individuality, a real person behind the curtain. Authors won’t disappear. Books will be more than ever.

——————-

PS 1. I haven’t talked about the shocking tactic of stealing authors’ works to train AI. There is no possible excuse for that. The perpetrators should be forced to pay royalties to authors for the rest of their lives. End of.

PS 2. I also haven’t touched on the awful carbon footprint of AI. I might as well have thrown in a grenade and burned this post to cinders. Something is terribly wrong with the logic of replacing human neural networks with artificial ones. We’re very energy-efficient thinking machines, despite all our flaws and biases.