As promised, I’ll touch on AI today.
When I decided to begin writing in earnest, back in 2020, AI wasn’t a thing. By the time its age began to dawn, I was spending a lot of time writing and feeling heavily invested in the process. I watched with amusement as the first chat bots came into public use, but later with horror as I saw them begin to write slop and then whole books’ worth of slop.
I am, however, a pragmatist and an opportunist, and turning my back on AI wasn’t going to make it go away. If the writer’s marketplace was going to be flooded with machine-derived slop and make writers’ lives harder, then the best thing writers can do is to harness AI themselves, where they can do so without damaging the quality of their own writing. And so my stance has become to focus on quality writing in the face of AI, but to exploit AI as much as I can in the process to even up that tilted playing field.
In the early days, Grammarly came out and I leapt upon it with joy because I had my first manuscript ready for proofing. But Grammarly had little grip on context and it took me longer to spot Grammarly’s mistakes than to spot my own. Next came Pro Writing Aid and it was streets ahead of Grammarly, although it was very easy to get lost in its many different analysis tools.
It was to be a few years after those first proofing attempts that I was again ready to proof a manuscript, several weeks ago. This time I engaged the AI, Claude. And, my goodness, was it good! It understood context almost perfectly and it knew to look for other faults such as hanging indents and trailing spaces, although all it could do was warn of their presence in the manuscript. I had no intention of making it agentic and getting its hands on my work.
This was simply icing on the cake, however, because as it read the manuscript, it could track characters and turning points. It checked in with me. Did you intend this turn here? Did you intend this progression of stakes here? You have a name switch here. Some of the comments were amusing because, being incapable of emotion, it didn’t understand humour or misleading plot twists, and would earnestly tell me that my narrator had been inaccurate when that had been the whole point. Reassuring.
Not only could it manage that, but I could then demand from it both a synopsis and a blurb, the bane of many a writer’s life. Both were absolute rubbish, but the synopsis did have all the right points in all the right order and the blurb did give an outsider’s overview which I, as the writer, struggled to see. Both made excellent starting points for crafting something better.
It took me less than a week to structurally edit and proof a 70k manuscript, and another week to produce synopsis and blurb. My own efforts on the synopsis and blurb were purely creative because Claude had done all the sifting and sorting. It had been months of painful work before.
ChatGPT does my research well. It can get deeper into internet resources than I can, and it will summarize well. It could never replace the year I spent reading thick history books to seed my first historical romance but it made fact-checking later fast and easy.
I can use both to discuss progress and chat about options. I used to crave being in a group of active authors where we could all bounce ideas around, but never managed to find any suitable resource. Living halfway up a remote Spanish mountainside doesn’t help in making actual contact with other authors. AI helps me focus my thoughts and ideas, almost becoming my muse.
I see all this as some recompense for knowing that the LLMs have stripped the human canon without consent or due payment, and will continue to do so when my work is in there. My real hope is that the ‘human-authored’ tag, which is an essential way forward, doesn’t preclude using AI for the kind of work I have described above. Let’s not tie our own hands behind our backs, but use the resource available with responsibility. Let it underpin and strengthen our creativity wherever it can.
None of the AI use above contravenes the Society of Authors’s definition of a human-authored book. I quote their requirements in relation to this:
The text of the work was created by you as sole author or together with other human beings and that it is your own intellectual creation as the result of the exercise of your skill, labour and judgment.
The text, therefore, represents the expression of your creativity as the human author and it has not been produced by the use of a Generative Artificial Intelligence (GAI) model e.g. by the use of prompts.
The use of GAI or tools with AI-embedded features does not disqualify a work from being ‘Human Authored’ under this scheme provided that the tools have been used for assistive purposes only e.g. the use of GAI-powered tools like a word processing programme, spelling and grammar check, for research, brainstorming or outlining, or any other use that does not generate text.
I see that this definition does not prohibit AI prompt generated imagery either, although many sales platforms ask the simple question was AI used in the creation of the book or the cover? and if you say yes, it can add an AI tag to your listing. Not a great idea. It may be as well to dig deeper into their terms to see whether the ‘use’ of AI in respect of the text means generative AI or assistive AI, as per the SOA definition.
I feel some optimism for the protection of human creation as generative AI now watermarks their output, and in a foolproof way, too. The problem will be to avoid human creative work getting drowned in the sea of slop. I think getting our work registered as human-authored is an important step for us all to take. Here is the link to the SOA human-authored declaration. You do need to be a member of the Society of Authors to make the declaration.