Writing to understand
There’s a dual interpretation of the title in this post. First, I’ve been using my own writing as a tool to understand what AI can and can’t do. It provides a useful medium for doing this as I have a lot of it, and it’s something I’m deeply familiar with so I can gauge the quality of the output.
I tried feeding a couple of writers tools, including Sudowrite and Novel.AI as well as Claude and Gemini with both my academic text (Metaphors of Ed Tech) and fiction writing (my in progress psychological horror novel).
The writing specific sites were good at generating content if you could describe scenes, characters, and style. From a purely technical perspective, this is pretty impressive. I asked it to generate a new chapter in the novel, and what it came back with wasn’t bad. I didn’t need to the chapter, this was just a test, but I did make an attempt at working it into something that felt appropriate. It reminded me of when I have taken my own blog posts and used them as the basis for a book chapter. They act as a good starting point even if what I actually end up with contains none of the original text.
The academic text didn’t fare as well here, perhaps it was because the structure was less easy to set out. I asked both Gemini and Claude to suggest a new chapter of Metaphors, based on what they had read. The best was Gemini’s “EdTech as Palimpsest”. Here’s a selection from the chapter it generated:
For this chapter, I want to delve into a metaphor that challenges this notion of constant, clean-break novelty: the palimpsest. Imagine educational technology, and indeed education itself, not as a fresh page for each new innovation, but as an ancient parchment. A palimpsest is a manuscript page, or scroll, from which the text has been scraped or washed off so that it can be reused for another document. Critically, however, the original writing is rarely completely erased; traces remain visible beneath the new text, influencing, haunting, or subtly altering the meaning of what is written on top.
It’s not a bad metaphor, and it does a decent job of matching my style. I felt both it and Claude tried a bit too hard to be humorous in places, like an uncle trying to be down with the kids. But maybe that _is_ my style?

In terms of what this means for writing I had a few thoughts. As I’ve mentioned before, I think the output improves with the quality and quantity of input you can give it. This is not a democratising technology then, as those who have more can get more from it. I noticed that a number of serious published writers endorsed Sudowrite. One can imagine that if you are a decent writer then you are very adept at specifying the structure of a chapter and providing sufficient content to generate similar work in that style.
One impact of this might be that writers can then create and publish books quicker. Instead of a book a year it becomes two or three. Georges Simenon was such a prolific writer, that there is an apocryphal story when Alfred Hitchcock called Simenon, he was told Simenon couldn’t come to the phone because he was in the middle of writing a novel. Hitchcock’s response was, “That’s all right, I’ll wait”. AI will turn every writer into a Simenon. This creates an arms race for authors. Like being a cyclist around the time of Lance Armstrong, even if you didn’t want to take performance enhancing drugs, you were forced to in order to keep up.
This comes to the second interpretation of my title. This kind of AI enhanced approach works well if you know what you want to happen, and especially in more formulaic genres where it can be plotted out neatly. It is sometimes proposed that there are two main approaches to writing fiction: discovery and outlining. Outlining stems from intricate plotting and structuring and then, essentially, filling out the plan. Discovery relies on just writing and seeing where the characters take you. Most authors tend to be predominantly one kind or the other, but will at some point use both elements.
AI enhanced writing works well for the outliner. But for discovery writing you have to do it, the act of writing is itself the means by which the author comes to understand what their novel is about, who their characters are, how it will develop. Both approaches are valid, but I feel there will be fewer unusual novels or surprises in a world dominated by an outliner approach.

One Comment
Alan Levine
NOBODY, no GenAI machine, writes like Martin. All are weak tea imitators.
Also, any use of “delve” is an LLM give away. Might as well toss in a “deep dive” as chattered by the NotebookLlM podcast twins.
Frankly, it’s all going to mix together all of the worlds bright colors of writing into grey goo where we dont know what’s what.. Keep it lavender fuscia , Weller!