The Rise of the Robot Writer

This piece of writing is certified 100% the product of an organic human mind, free of any aid from artificial intelligence. You can tell because it was supposed to be a quick 4-paragraph blog post and it turned into a sprawling 2,500-word essay that took me a whole week to write. You really need a human for that level of inefficiency.

If you’ve been on the internet much lately, you’ll probably have seen some of the ongoing fuss around ‘AI art’. Social media has been awash with slightly creepy-looking AI-generated pictures and apocalyptic predictions that the human artist will be very shortly rendered totally obsolete. In the speculative fiction space, the venerable magazine Clarkesworld has been forced to close to submissions because a deluge of AI-generated stories has flooded their system. I’m a slush reader for Apex Magazine and, while we haven’t seen a huge uptick in submissions, artificial or otherwise, we have had a great deal of conversation in our discord about how we could spot such stories and what we should do about them. Meanwhile, the book ‘Fractal Noise’ by Christopher Paolini is being published with a cover featuring artwork generated by AI, despite the controversy this has caused. Will we soon see a world where all books consist of AI-generated words with AI-generated pictures on the front, and no human intervention needed at any stage? Is this the end of human creativity?

I don’t think so.

Now, there’s no doubt that this is an interesting moment in the history of art. New technology is enabling a rapid expansion of possibilities, and that technology is evolving much faster than the ethical and legal frameworks. There’s an awful lot of discourse out there, not all of it very illuminating. Shout out to the Fansplaining podcast and their episode ‘Artificial Fandom Intelligence’, and to the SFWA (Science Fiction & Fantasy Writers Association) which has handily collated a number of responses from its members on this web page.

If, like me, you’ve read too many science fiction novels for your own good, you may be imagining some kind of android, (possibly resembling Murderbot, the entertainment-feed-loving robotic protagonist of the book series by Martha Wells) sitting with a headset on, spinning artworks out of raw pixels from the depths of an electronic brain. Or, if you’ve been doomscrolling artists’ social media, you may be imagining a viral plagiarist, stealing pictures straight from the web and putting a computer-generated gloss on top, a bit like nicking someone else’s photograph, putting a few filters on it, then claiming it as your own. (There are some programmes that work a bit like this for text, substituting synonyms for common words with sometimes hilarious results, but that isn’t the same thing as generative AI).

The truth is both more prosaic and more powerful than either of those things. I think it’s important to remember that programmes like Lensa for visual art and ChatGPT for text are tools, used by humans. They haven’t achieved sentience, or anything like it. They’re essentially algorithms, trained on a vast corpora of artworks or text (the ethics of training these tools on these creative works without the permission of the original creators is an interesting question, and not one I am equipped to answer). A human puts in a prompt, and the AI tool works pixel-by-pixel or word-by-word, making predictions about what should come next based on what it’s seen before. The tools have no actual independent understanding of what it is they’re doing. The reason so many AI-generated pictures have things in the bottom right-hand corner that look kind of like signatures is because the tool has been trained on pictures with signatures in the bottom right-hand corner, and so they generate a squiggle that goes there. It’s a simulacrum of creativity without a creative mind behind it — except for the human mind putting in the prompt.

These programmes produce works that look superficially slick but which on closer inspection have issues. The pictures often have extra fingers or other weird details, while the text lacks any real insight or coherence. Even if — when — the technology improves substantially, it will still be incapable of generating anything truly original, just endlessly regurgitating the stuff it’s been trained on. Maybe one day there will be an artificial intelligence capable of true creativity, but we aren’t there yet. For now, it’s still humans behind the curtain, pulling the strings.

So who are these humans? Since words are my thing, I’ll concentrate on the humans using the writing tools. There are many different uses for the technology — catalogue descriptions, SEO filler, cover letters — but in terms of producing creative work, I think there are essentially three different groups: stuck writers, would-be scammers, and consumer-creators.

Stuck writers are those who are using the likes of Chat GPT to get over their writers’ block (or meet punishing deadlines) by feeding in a prompt and then using the text it generates as a starting point for their writing. Once the writer has gone over the text a few times, editing and expanding, it’s not really distinguishable from something created entirely organically. Where the precise line is between ‘AI-generated’ and ‘human-created’ writing and whether those two categories can be or should be distinguished in law — eg whether AI-generated text can be copyrighted and if so, who should hold the copyright — are yet more interesting questions without easy answers. In one sense, you could say that this is just another tool for writing, and shouldn’t be treated differently from, say, using a word processor. If someone is using this as part of their creative process, and the end product is as good as something written purely by a human, then should we have a problem with it? Or will we start seeing two types of fiction: AI-aided, and fiction certified as purely human, organically produced and AI-free? And each year a random selection of organic authors compelled to show evidence of their AI-free creative process, like having their taxes audited? While I don’t know how all the details might actually work, to be honest, I wouldn’t be surprised to see a bit of a two-tier system emerging.

The stuck writers are at least using the technology for an honest purpose, unlike the second group, the would-be scammers, who are ones gumming up the works at Clarkesworld. These guys have got hold of the idea that there’s easy money to be made generating a whole bunch of stories and then sending them in to fiction magazines. There are two problems with this scam. The first is that the quality of the stories they’re producing is… really bad. The AI can produce a string of sentences that sort-of make a story, but it can’t produce narrative tension, characterisation, or beautiful imagery. Check out this twitter thread and this review of an experimental all-AI magazine for some examples. As a slush reader myself, I can confirm that the competition for magazine fiction is intense. We receive close to a thousand stories a month, and publish about half-a-dozen. I’ve personally rejected hundreds of stories that were good — really good, far far better than anything the AI is close to producing — but they weren’t quite right for Apex and so into the reject pile they go. And I can’t see AI-generated work reaching our standards any time soon – however much the writing improves on a sentence-by-sentence level, what the speculative fiction world values more than anything else is ideas. Fresh, interesting, chewy ideas, explored with beautiful prose and a deep sense of humanity. Yes, it turns out that the robot genre is the human genre after all.

The second problem with the scam is that, even if the stories improve exponentially in quality, there just isn’t much money to be made from magazines. Even the top writers are only publishing a few stories a month, and making hundreds of dollars from their short fiction, not thousands. It’s not a rich potential source for the scammers. And there’s something a bit self-defeating about the whole enterprise: if AI-generated stories get good enough to be published, why would Clarkesworld (or anyone else) pay humans to submit the stories? Why wouldn’t they just generate their own, and act purely as a curator for AI fiction? Or would you even need the magazine at all? Wouldn’t readers just grow their own?

This brings us to the final group, the creator-consumers. This is currently a minor part of the fiction scene, although I believe the visual art world is ahead of us, with people who might previously have commissioned fan artists to make them icons and pictures of their favourite characters now turning increasingly to AI-generated artwork instead. I think we may well start seeing something similar happening in the fiction space.

The fact is, there is a market — a large market — for bad fiction. Or rather, for fiction that is bad in the specific ways AI-generated text is getting increasingly good at replicating. There are hundreds and thousands of books out there with passably competent writing and totally formulaic stories and characters. AI text generators are soon going to be great at producing books that are passably competent and totally formulaic, much faster than humans can manage it. Will these books still have a human creator/curator behind them? Or will they be 100% artificial, an algorithm set free to generate text and put it up on Amazon (or wherever) without any human intervention? Will they share shelf space with AI-aided books and organic human-produced books, in a three-tier publishing system? Will all the people who are currently earning a living producing these kind of formulaic works get squeezed out altogether? What would stop publishers from cutting those pesky authors out of the equation and just generating thousands of books, then delivering them straight to readers? Are publishers even necessary to the equation? If you can simply plug in your preferred tropes to a text generator, and instantly get something tailored to your personal tastes, do you need anything resembling a separate ‘creator’ at all?

I’ve even heard predictions that, before too long, each individual consumer will be reading/listening to/viewing a completely personalised selection of AI-generated work, with no two people ever consuming the same work. Are we heading for the total atomisation of art?

To be honest, this scenario seems unlikely to me. Sure, I expect there will be a lot more of this kind of creator-consumer, but also — humans like to share art, not just consume it in a sealed vacuum. We get excited about something we’ve seen/read/heard, we want to tell our friends and family, and chat about it on the internet, speculating about what might happen next.

Also — we sometimes want to be surprised. If everything is all formulaic pablum all the time, there are no surprises — or not interesting surprises at any rate, just glitches in the matrix. So, while I obviously can’t predict how everything will turn out, I really struggle to see a future in which all new media is fully computer-generated, customised to personal tastes of the consumer, and enjoyed alone.

What I can see — easily — is a future in which it’s impossible to make a living as a writer of passably competent formulaic fiction, and nearly impossible to make a living as a writer of any kind. It’s not the scammers, ultimately, who are the issue — they’re a temporary annoyance who I think will soon give up once they realise they’re not actually coining it as fake writers, because hardly anybody is coining it as real writer. The issue is an ever-increasingly supply of written work, coupled with static demand. AI text generators make it much easier to produce passably competent work, and so we’ll inevitably see much more of it, whether created by the consumers themselves, by individual ‘writers’, or by companies churning out vast quantities of artificial media.

Where does this leave us? Well, to be frank, it’s already nearly impossible to make a living as a writer, or as an artist more generally. Taking a long view, I think this AI stuff is the latest in a long line of innovations that make art easier to produce and reproduce, and hence lead to an over-saturation of artwork and a winner-takes-all marketplace where the most popular art gets everywhere, and the vast majority of art goes nowhere. The easier it is to make art, the less appreciated the artist.

We’ve come a long way from the days of hunter-gatherers, sitting round the bonfire feeling bored, until someone starts telling a story. First writing, then printing, typewriters, word processors, audiobooks, ebooks… now AI. All this stuff makes it much easier to tell a story, and much easier for stories to spread. The most popular storytellers, whether human or otherwise, can now reach a global audience — while the vast majority of storytellers, absent a local bonfire and bored group of hunter-gatherers, struggle to find an audience at all.

So it’s all doom and gloom? Well, I suppose you could look at it that way, especially if you’re a writer who currently scrapes a living from Kindle Unlimited and you feel your livelihood is under threat. In all honesty, I think we’re due a broader reckoning with the impact of artificial intelligence on all of our lives and workplaces, and I think the long-term solution may well be some kind of universal basic income.

That’s all way beyond my paygrade though. What is my paygrade? Am I getting paid to write these words, right now? Nope. Did I mention that neither I, nor anyone else on the Apex Magazine slush team, gets paid a dime to do it? That’s right — we do it all for the joy of discovering new voices in fiction. And, while the writers whose stories make it into the magazine do get paid, they didn’t know that at the point they sat down at the keyboard to make their ideas come to life. They had no guarantees they’d get anything out of the process except their own creative fulfilment.

From the days of boredom around the bonfire until now, humans have always loved to create stuff. And sure, if we can make a living out of our creativity, that’s great, but even if we can’t, we do it anyway. I don’t see that very-human urge to create going away. In the dawning age of the robot writer, then sure, I expect they’ll be a lot of computer-generated writing around, but don’t think for a moment that humans will stop writing. And I think there will still be people who want to enjoy the products of purely human creativity. After all, we can listen to thousands of hours of electronic music on Spotify, but we still pay money to go and see humans stand in front of us and make noises with their mouths.

Humans not only love to create and consume, we also love to share experiences with each other. So, no matter how much robotic pablum is out there, I think there will still be a place for art made by 100% organic human brains. And that will be the case, until civilisation crumbles to dust — and then we’ll be gathered round the bonfire again, telling each other stories.