I'm writing a novel without using AI – and I can prove it

Spectator

7 days ago

Everyone's seen stories about the creep of AI into art of all kinds. Recently the people behind the music-fabrication website Suno have been making outrageous statements to the effect that people don't enjoy learning musical instruments and writing their own songs, so why not let AI do it for them? This is very new, very disturbing and very consequential. I could talk about graphic art and video and film-making, but you'll know what's been going on there. I'll just cut to the chase and get to how AI tools are impacting and will continue to impact the writing of fiction.
I anticipate a future in which human authorship will need to be proven. A few years ago I simply wouldn't have believed that this landscape could be possible. In 2017, a team called Botnik fed the seven Harry Potter novels through their predictive text keyboard, resulting in a chapter from a new Harry Potter story: Harry Potter and the Portrait of What Looked Like a Large Pile of Ash. With some human selection what emerged were extracts such as: ''If you two can't clump happily, I'm going to get aggressive,' confessed the reasonable Hermione.' 'To Harry, Ron was a loud, slow, and soft bird.'
Things have come on since then. Now, if you ask ChatGPT or any of the other engines to write about the moon landings in the style of Finnegans Wake, which I have done, it will produce something pretty plausible, possibly not better than you could have done yourself given an hour or two, but rather compensated for by the fact that it took two seconds.
As a result, novelists are already writing novels with AI. Are they as good as human novels? No, not yet. It's a process, probably, of gradual supplantation. First the writer uses AI to brainstorm ideas, then gets the AI to write a scene based on the most promising idea, then gets AI to supply a whole chapter, then the whole of the book. Gradually human oversight is reduced and then eliminated. In 2024 the winner of Japan's most prestigious literary award, the Akutagawa prize, admitted that she had written her novel with the help of artificial intelligence, though this confession was made after she received the prize money. She was praised for her honesty. Perhaps the majority of serious current novelists are experimenting with it, because it is just too tempting. I would guess that in future most novels will be written with AI help, because authors have deadlines, they are weak, and they fear the blank screen.
There are people out there saying: never fear, AI writing is just autocomplete on steroids, it will never have emotions, it will never write creatively, it will never be original and it will never truly engage a human reader. I used to say things like that. Now I don't. AI probably can't think and probably isn't conscious – although Geoffrey Hinton, who helped make it, argues that it can and is – but that doesn't matter. All it needs to do is convincingly mimic thought and consciousness, as well as mimicking creativity and originality. After all, who's more likely to be original, a human or a machine that has access to every book every written? Is there anything new under the sun? If there is, won't an infinitely resourced machine be able to shine its own light on it? That's when human novelists will be completely, irrevocably superseded.
The terrifying thing is it doesn't matter if AI machine novelists are not very good, or even if they never get as good as a human writer, since for a majority of people they will be good enough. They will out-compete, and out-autocomplete, human writers, just as AI bands are mimicking human bands with enough success to suck revenue away from human musicians on Spotify. Writers' livelihoods are at stake because consumers won't care enough.
Except… what if there is a market for novels if they are demonstrably written by humans? What if there is, in ten years' time, a market for an artisan novel, quaintly written on the premise that no machine had a hand or a robotic arm in its creation?
How, though, could this be proven? It's possible at the moment to detect AI text, but only if the writer has been careless, and the tools to do so are clunky and sometimes inaccurate. After generating the text, the writer can 'humanise' it, either by hand, or by employing a humanising program. So I'm proposing something. I want to write one of the world's first provably, demonstrably non-AI-assisted novels. And this is how I'm going to do it. In fact, this is how I have already started doing it.
During every writing session I livestream my desktop and have an additional camera on my workspace and keyboard. I have a main novel file, some character files, a plot file and a scrap file. I may also have other files. All these files are in one folder and accessible to pull out. This bringing up of files from the main folder is viewable on screen. There is no access to the internet, and certainly nothing AI-generated. At the end of each writing session in Google Docs, I save a named version. At the next writing session I open Google Docs and identify that last version at the top of the list, date- and time-stamped as it is, demonstrating that it is the last version I worked on and hasn't been altered. Then I go back to Google Docs and start working, live-streaming and recording. At the end of the session I save the version so I can return to it.
This protocol I call Maximal Human Authorship Protocol or MaxHAP. It, or something like it, is going to be required in future, because if we don't have it, no one will ever be able to say again, and be believed: 'I'm a writer.' Does that matter? It matters to me, because I've been writing for a long time, and writing is among the things I value most in the world. I want to protect the notion of a verifiably human author, of the dignity of that author.
In future, the writer will have only a little dignity. Let's not make it none.