Ok, for some of you this will seem harsh, but bear with me. I don’t know any professional writers who have not, at some point, been approached with the line: “I have a great idea for a book. I give you the idea, you write the book, we split the profits 50/50, ok?”
There’s a reason no writer will accept this offer. First, most writing does not earn enough money to be splitting it! More importantly, the “idea” part is pretty much worthless. Any writer worth their salt can come up with ten pitches in ten minutes. All of us have a queue of books in our heads – the time and effort to get them on the page is the real challenge.
So what does this have to do with AI?
All over social media at the moment (particularly if at some point you expressed an interest in writing) you will find adverts for specialised AI that will write your novel. All you have to do is type in your idea, genre, and character names, ten “poof” you have a novel. Follow the YouTube guide on the Amazon KDP website and you could have it published before the sun sets. Well done! You are a published author!
But are you, really? And what have you gained from this process?
- Writing a novel isn’t about having one big idea. Don’t get me wrong, big ideas help. But a book is constructed of many, many, many little ideas woven together to make a coherent whole. Every page is full of decision to make, choices, all of which require new ideas. If all you’ve contributed to a novel is an initial concept you haven’t really experienced writing a book at all. In fact, if most of the book is made up of the little ideas, is your novel really yours?
- Large Language Models, broadly speaking, work by churning through enormous amounts of words and counting how often the words are precedent or antecedent (next) to each other. Do this often enough, and it starts coughing out sentences. Choose a data set (say, all the novels written in 1962) that contains the sort of sentences you want, it starts giving you something that looks a bit like a novel written in 1962. Have someone training it, telling it whether it is producing “good” or “bad” sentences, it will get even closer. But it isn’t “intelligent”. It isn’t writing. It’s spitting out a generic mess of whatever data it was fed. If that sounds like a good novel to you, then, respectfully, you need to read some good novels.
- We’ve established that AI is really good at sorting through lots of data and finding things that frequently link up. That’s fantastic if you want a list of novels written in 1962. But it’s terrible if you want originality, creativity, or spark. Those are things for which you need actual intelligence. AI models are going to get really good at producing something that looks like a novel, but they will always miss that special something that demarks literature from slop.
The bottom line is, do you want to write a novel, or do you want to be able to say you’ve written a novel (of course, you can do both). Really, actually writing a novel is about going through the process; it’s about being able to turn to any page and remember how you struggled to decide the colour of a tie or whether the character should say “ok” or “sure”. When you finish the novel, good or bad, bestseller or hidden somewhere on amazon, you have made something that is truly, genuinely yours; an expression of your truest self. An AI can never do that. All it can do is give you the most generic text possible based on the prompts you give it. And don’t mistake prompts for writing. You’re just the guy saying “hey, I’ve got a great idea for a novel, you write it and I’ll split the profit 50.50”
Leave a Reply