AI and Creative Writing

A standard question on literary agent query forms is whether the author uses AI to create any part of the manuscript or submission materials. I have an answer that is simple in my head but not so easy to explain. On top of that, even though I find AI extremely helpful, I share the widespread anxiety about its risks.

If AI is used to create, it takes over from thinking. I worked for decades without AI being available, and so I’m in the habit of creating everything I write. Today’s young people are bound to find it hard not to be seduced into having AI doing the thinking for them. Schools are employing a variety of techniques to discourage this practice, from barring the use of AI for students under a certain age to applying AI to uncover a student’s use of AI. I’m not sure how this threat can be addressed adequately. However, whether we like it or not, AI is here to stay. It’s like petroleum: We can use it to fuel cars or blow up buildings. Except, as everyone now knows, the risks of AI are even greater—far greater. We must keep figuring out how to harness it.

For me as a creative writer, AI’s most obvious benefit is its ability to detect typos. Without AI, it’s a time-consuming process for any author, but impossible for people who cannot see print. I can’t find every typo because not every item in a manuscript is spoken by my speech synthesizer. I can use spellchecker to find some errors, and a certain program picks up spacing and a few other flaws, but neither catches everything. Even if my speech synthesizer did catch everything, the process of reading character by character would be a monumental task undermining my productivity. The paid ($20 a month) version of ChatGPT finds everything.

Another way I use AI is to find redundancies. I do so by asking ChatGPT to assess a manuscript for strengths and flaws. Obviously, the results can be broader than redundancies. For example, AI not only recommends changes, but how I should word them. I tell it to stop it. I instruct it, instead, to state the issues and identify the passages it urges me to delete.

Once AI and I agree on this protocol, in the process of identifying flaws, it finds passages where I’ve repeated a detail or idea already adequately addressed. It might also suggest how a novel moves too slowly in the opening chapters, a perennial concern when there are several main characters. AI’s recommendations for how to get the story moving along can be helpful. This aspect of AI is what I mean by redundancy. If I don’t need to say something, don’t say it!

Overall I use AI the way I’ve approach a creative writing class, where the teacher and fellow students respond to a story with all kinds of suggestions, many useful, many not. I select those that work for me and set aside those that do not.

Even so, I hesitate to check “No” to the question whether I used AI to create any part of my submission. I do so anyway, but add a sentence to my query letter explaining, as well as I can, what I mean by that “No”: I say I use AI to subtract, not add.

When I was first testing AI two or three years ago, I entered an idea into the search field and asked the bot to write a poem around it. The results were remarkable: lines with meter and rhyme. However, the poem as a whole was awful, filled with predictable phrasing and wooden in its sentiment. AI has no feeling for a truly fine poem’s impact on the human heart. Even if I’d wanted to use an AI-generated poem as a starting point, its artificiality destroyed my original inspiration.

I write because I have things I want to express. Having AI express those things would defeat my reason for writing. My writing voice may appeal to some and irritate others, but individuality matters. It’s from our own experiences, our own reflections and the consequences of our actions that we hope to move readers.

For now, I hope my use of AI is not a bar to getting my work out in the public arena.