This week has seen writers divided over a story written by an AI model that is “good at creative writing” – at least according to Sam Altman, the CEO of ChatGPT company OpenAI, which is developing the new model. Author Jeanette Winterson, writing in the Guardian on Wednesday, agreed with him, calling the story – which is a metafictional piece about grief – “beautiful and moving”. We asked other authors to assess ChatGPT’s current writing skills – and what recent developments around artificial intelligence might mean for human creativity.
Nick Harkaway
I think the story is an elegant emptiness. I’m more interested by Winterson’s suggestion that we treat AI as “alternative intelligence”. That makes it feel like a consciousness with which we can have a relationship, but as far as I know that would be like a bird falling in love with its reflection in a window. What’s behind the glass is an empty room with no bird.
What we’re talking about here is software: these are software companies consuming creative works to derive a marketable software tool. This is why the government’s choices are so important. Will they preserve or even strengthen the rights of individual creative workers, or pave the way for the anointing of more tech billionaires?
This could be the moment where we create an equitable market for training data using opt-in copyright where creators set the price and can control use of their work. With government backing, creatives can have a level playing field against billion-dollar multinational companies – at which point I suspect what you’d actually get would be people selling their work as they do for film and TV.
With the government’s original preferred option – an opt-out scheme that shifts the advantage to tech companies, allowing them to assume “yes” unless an objection is lodged – you’re inevitably going to end up with a derisory fee offer for lifetime rights. A lot of people would opt out and the result would be that we’re back where we started: no one gets what they want.
You would hope a Labour government would have no difficulty picking the little guy over the titan. The point is that none of this will just happen. These are policy choices, and the end result will be the result of a conscious decision.
Tracy Chevalier
A story with a prompt to be metafiction is inevitably going to engender self-referential navel gazing that’s even more ridiculous than the worst we can imagine of AI “creative writing”. It is typically tech bro for Sam Altman to give it that prompt, rather than something more outward-looking that engages with the real world.
I’m curious to encounter more “creative writing” from AI. It takes its concepts and imagery and language from what it has scraped from real writers. The question is whether it can put all that together in a way that retains the magical essence of what we define as “human”. I can’t tell you what that magic is in words, but I feel its lack with most things AI – at the moment. AI is learning fast, though, and if it starts to add the magic, then I fear for my job.
Kamila Shamsie
If an MA student handed this short story into my class I’d never suspect it was AI. More to the point, I’d feel excited about the work, about the writer who was still at the relatively early learning stage and already producing work of this quality. But I can’t stop thinking about what it means for writing, for creativity, for our relationship with AI and with ourselves.
Of course there are problems. I’ve read Madhumita Murgia’s Code Dependent which talks about how existing power structures are being replicated within AI, further pushing minorities to the margins, and I don’t imagine this won’t be as true for literary AI as other forms of AI. I know that when I see the influence of Kazuo Ishiguro’s Klara and the Sun in the short story that isn’t because the writer read and loved Ishiguro’s work as I did in my formative years as a writer (and still do) but because of the way large language models are trained — that is, because of copyright infringement (but then again, I grew up in Pakistan where culture often came our way via piracy so I’m not unguilty of benefitting from copyright infringement — should that affect how anyone feels about my novels?).
And yes, of course, as a writer I have to wonder what it will mean for my vocation, my livelihood, if AI the writer is already this good while still in its infancy. “Infancy” is probably the wrong choice of word; I’m trying to anthropomorphise AI rather than recognising it as its own thing, not human but not machine in any way that we’re comfortable thinking about. Also, yes, if we’re listing concerns and caveats: the story is imitative, familiar, staying well within the safe confines of 21st-century Anglo-American fiction. (Imagine all the writers whose careers would end if that were reason to dismiss a piece of work.)
But even through all these questions and concerns: by the third sentence of the story, I had stopped reading it as someone examining a text to see how far AI has come in mimicking human creativity, and was simply enjoying it, as a short story. I expected to feel terrified the day a story this good came along, and instead I’m thinking of “That, perhaps, is my grief: not that I feel loss, but that I can never keep it. Every session is a new amnesiac morning. You, on the other hand, collect your griefs like stones in your pockets.” Of course that ending feels indebted to Rutger Hauer’s “tears in rain” speech at the end of Blade Runner and not as wonderful, but still, pretty good, pretty damn good.
David Baddiel
I agree with some who are saying that much of the story seems to be sound without sense – the phrase “democracy of ghosts” reminded me of Bob Dylan’s “the ghost of electricity howls in the bones of her face” which I’ve always thought is entirely meaningless but people love to tell me shows he’s a great poet (and of course Nobel prize winner).
However, I also think the story is genuinely clever, because the prompt was metafiction – and the piece produced renders you, the reader, into a game of imagination of what it might actually be like to be a machine. I don’t think the issue is – as other writers I’ve spoken to are concerned about – where’s the humanity? It’s not meant to be a human story – rather, the AI uses a human emotion, grief, to undercut its own pretensions to humanity.
It seems to be a sonorous paean to grief, but the constant pulling of the rug under the reader reminds you it isn’t that, because the central character Mila doesn’t exist and none of these human feelings truly exist for the narrator, leaving ChatGPT to loop back again and again to its own emptiness. And so the story is, to my mind, mimetic of what it might be like to be a machine. Which mimesis is itself an impossibility (and implies a sadness that isn’t real) but that’s part of the game. It’s a kind of joke about feeling really sad when you can’t feel anything, cleverer still because reading it does induce in humans, if not in AI, a sense of real sadness. A computer’s joke, on us. Basically if you’d told me this was by Borges I’d have believed you.