Posts Tagged GPT-3
When machines write books: will AI writing threaten authors’ livelihoods?
Posted by Roz Morris @Roz_Morris in How to write a book on February 12, 2023
Chez Morris, we’re divided about AI tools. I’m mildly interested, while Husband Dave enjoys experimenting, just from curiosity. While warming up for a day’s writing, he asked GPT-3 for a description of a character walking along a harbour and posted the result on Facebook.
I started to read, but lost heart after two sentences.
Why? The reason for that interested me much more than the content.
The paragraph was respectably publishable. It even had a whimsical idea, that fishermen were stealing fish from the sea.
But it felt depressingly pointless. It came from a void. There was nothing meaningful or living on the other end.
Novels, memoirs, poetry and creative non-fiction are more than words. They are a bridge to another soul. A soul that notices and feels and has mysteries and questions it needs to share.
If you’re looking for that, AI-generated text is empty calories.
That doesn’t mean it’s not useful. I came across a literary author who used it to write about a deeply personal experience. Which was intriguing.
In an episode of This American Life, novelist Vauhini Vara wanted to write an essay about her sister, who died when she was in college. She never found the right way, so she briefed GPT-3. The AI came back with 100 words about losing her sister, then meeting a guy who made her forget the sadness. Although this wasn’t the direction Vauhini wanted, it showed a grasp of story structure – the essay needed to end somewhere new. She refined her prompt and had several more tries, which became a fascinating discussion between her and the AI. What about this direction? What about this? What do you really want? Finally she realised; she wanted to explore the loss. By briefing and rebriefing, the AI got her there, because it knew millions of examples. Find it here.
Millions of examples
You can’t talk about AI tools without considering what they’re learning from. Work by you and me and everyone we like to read. And not necessarily with the permission – or even the knowledge – of the creators.
Although we all learn our art from the work of others, AI tools are doing it faster. And on a bigger scale.
This is where it gets worrying.
Every day there’s a new and troubling iteration. A tiny example from my Facebook voyages this week – several authors have noticed a clause in their contracts with audiobook platform Findaway Voices, which allows Apple to use audiobook files for machine learning training. You let Findaway distribute your books and they let Apple train AIs on them.
Train them to do what? Probably many harmless and useful things, but one of them must be to narrate audiobooks instead of an expensive human. And authors aren’t given the choice to keep their books out of this great experiment, which will probably make a lot of money for a corporation somewhere. Legally, those are derivative works and the original creators have a right to share in the proceeds. There’s more about it on Victoria Strauss’s blog, Writer Beware.
This clause isn’t new. It’s been in Findaway’s small print for years. One author found it in a contract she signed in 2019. There may be countless other rights-grabs we’ve unknowingly agreed to over the years, or been opted into by publishers who released our work. Or our work is probably being used anyway, whether there’s a contract or not.
The use of creative work without permission is becoming normalised because it’s impossible to stop. The moral boundaries to it are breaking down. That’s not a healthy trend.
We can’t stop the machine
But could we protect works in copyright? Is it too late? Perhaps not. A few years ago, websites used to post cookies on your devices, whether you liked it or not. Often you didn’t know. Now you have to agree to it, and though it’s maddening, you can refuse permission if you want to. (I always refuse, just because.)
If that can be done, creators could be asked for active and expressed consent and could opt out.
I’m not sure what good it would do, except to reinforce the point that creative work is a skill, a craft, a service and a business.
Books will be written by AI, but…
There’s a cynical view that AI could create the pulpier kinds of novel where readers want procedure or plot or iterations of tropes. (Do those readers really exist? I can’t comment. It’s not my world.) There will be literary AI experiments. Collections of poetry. Probably memoirs, just because. (Though who would pay for a memoir written by AI? You could just generate one yourself, free.)
There will be breakout sensations. AIs might entertain us with a whacky fresh juxtaposition, like the fish being stolen from the sea. Some of the output will be weird or moving, because it’s monkeys with typewriters. And also because the reader supplies some of the meaning in a work, often without realising how much they are doing. But they usually do that because they think there’s a guiding purpose, an answer to find.
This brings me to the question of originality. Novelist Ted Chiang talks about that here.
Non-creative people sometimes tell me there is nothing new under the sun. I disagree. There are new things, all the time. Although we all – AIs and people – learn on what has been done before, that’s just the start. Then comes the work. And the art and the craft. We experiment and refine until we find the way to express our own truth, a truth from our unique complications and depths, the new thing that’s worth saying, and for readers is worth reading.
And so I contend that in certain artforms, and that includes creative writing, you can’t cut out the expensive human.
The human is the entire point.
There’s a lot more about writing in my Nail Your Novel books – find them here. If you’re curious about my own work, find novels here and my travel memoir here. And if you’re curious about what’s going on at my own writing desk, here’s my latest newsletter. You can subscribe to future updates here.