By Ben Williams
Ben Williams is the Publisher of The Johannesburg Review of Books. He’s formerly the Books Editor of the Sunday Times and the General Manager for Marketing at Exclusive Books.
Thanks to artificial intelligence, we are sprinting into territory where reading, writing and publishing will never be the same again.
Idon’t mean to alarm you, but these dying months of 2022 represent the last moments that we’ll collectively be able to trust the words we read.
This column will accordingly count as one of the final relics of the sliver of time between the end of the Gutenberg Age, which closed in the early 2000s, and the beginning of an as-yet unnamed era, which will likely stretch on until civilisation itself splutters out, and which will be stamped by an uncountable number of texts written by artificial intelligence.
Call this sliver – the slimmest possible band in the history of writing since the first glyphs were scratched into clay in the Indus Valley some 5,000 years ago – a mere twenty year stretch. For all the fretting I’ve done, during the last two decades, about the demise of the “Gutenberg mind” that I grew up with as a Gen Xer, looking back from the current precipice, I’m now incredibly grateful for the taste of mental rewiring this period has given me. The unique sensation of muddledness, of unfocus, that’s produced by our constant state of partial attention and context switching has been an exquisitely privileged kind of meaning-making all along. For we’ve been grappling with a modern whirlwind of speech and ideas still constrained by thought processes native to the human condition – our condition. That all ends shortly. We’ve lived through the frenetic twilight of human sense. As I say, it’s been a privilege.
Lest there be any doubt about what I mean by “texts written by artificial intelligence”, I’m speaking of the complete works: SMSs, tweets, Facebook posts, emails, abstracts, essays, news articles, columns (!), lyrics, poems, short stories, plays, screenplays, novels – literally every piece of copy we lay our eyes on, or hear sung or read aloud.
Soon – it’s a matter of a handful of weeks – everything about these forms will change. In 2023, we will say goodbye to original writing forever. Or, at least, our day-to-day ability to discern original writing will be nipped in the bud. There’s little practical difference between these two scenarios.
The main culprit driving this epochal shift is an algorithm called GPT-3, which stands for Generative Pre-trained Transformer, version 3. This program can put together strings of words on any subject that read like they were written by an educated person slightly pressed for time.
The technology was created by a company called OpenAI and is closely guarded and parsimoniously licensed. But the leakage from its public-facing access nodes has already proven that there is no escape from its gravitational pull. I personally have read, in the last thirty days or so, several texts belonging to the forms mentioned above, whose main purpose was simply to shock me when it was revealed that GPT-3 wrote them. I was indeed shocked, each time. The only conclusion to come to, after seeing the power of this machine, is that all writing will, in future, bend to AI’s event horizon.
That meaning disappears along with human intention when words are produced by machines, no matter how cogently, should be self-evident. (Is the machine writing what it believes to be the right words, or is it writing what it hypothesises are the right words for making a human believe they’re the right words?) The same goes for art: no doubt some AI-produced texts will have aesthetic value, but I’m inclined to hold that original works they are not.
Starting in … January? February? AI will leave us on utterly unsure footing when it comes to all the reading we’ll do for the rest of our lives: How much of the text I’m scanning was written by a person, how much by a machine? Further, as AI grows inexorably into the world’s master author – if a mediocre one, as generative processing tends toward outputs that represent the average, rather than the exceptional – certain responses by those of us who remain in writing and publishing seem equally inevitable.
Here are a few variations on these possible responses from my post-Gutenberg, pre-AI imagination.
First, get ready for books written by algorithms to be marketed as such, and turned into best sellers. Genres with set formulas, like Romance and Thriller, will be especially susceptible to algorithmic fracking: publishers will leap at the chance to commission novels written by machines with certified track records of mining hits. “From the unfathomably twisted cyber depths of The Archangel, release flavour 4.71, whose bestselling code expressly guarantees you sleepless nights, a new terror rises to confront Jack Babbit. Your next download begins on October 31st.”
Beyond that, expect publishers to offer custom-made reading experiences. Do you like Romance? Adjust the chivalry-to-erotica sliders on the screen, scroll through a series of yes/no questions on the plot elements you prefer, name your characters and await immersion in a bodice-ripper composed just for you, delivered to your Kindle in under ten minutes. Want to go deeper? “Subscribe for a personalised series that features you at the centre of the action. ALL the action.”
On the same note, watch out for collectives of “meta writers” who craft original generative algorithms instead of stories, selling the former to publishers, or self-publishing their algorithm’s outputs; and editors remade as “AI whisperers” who are able to put the finishing touches on generated texts by dropping the right nudges in the right places, elevating the final product to premium status. Conversely, we’ll also need “AI detectives” on the staff of our publishing houses and news and academic organisations, dedicated to sniffing out, with their suite of advanced tools, any generative prose dead spots within submitted work. (Good luck to them: with AI, the house always eventually wins.) And lawyers will get involved, too, crafting attestations for students and authors to sign indicating that their work is original and “AI-free”. (Again: we’ll only have their word to go by.)
Among the next generation of literary Luddites, meanwhile, expect an especially creative response to the vexed question of how to establish originality and authenticity of authorship. Step one will probably involve the creation of a universal symbol or mark, Creative Commons-style, which writers will use to stamp their work as “100% human”. Step two will be more experimental. Imagine, as a means of enshrining the creative spark dwelling in our organic brains, a new tradition that involves crafting poems in real time while being livestreamed and supervised. Or – extreme writing contests that borrow from chess matches, in which human authors pit their wits against a machine, aiming to predict the text that should follow a series of prompts, with the “engine correlation” establishing the level of the human writer’s knowledge and mastery of rules-based composition.
There will also likely be a turn to the absurd and incomprehensible, in aid of establishing modes of writing that machines can’t mimic, because they make no sense – the problem being that a machine can learn to make no sense fairly readily. It may follow that literary outliers like the Bad Sex in Fiction Award turn the tables on humans and machines alike, in that intentionally writing an extremely terrible sex scene, one that is much worse than it was thought possible for a sex scene to be, which is quite hard to do, becomes the apogee of high literary art. Text that is amateurish, super-dense, or intentionally broken then becomes the mark of a virtuoso.
We may also see a new flourishing of artistic writing, à la Japanese calligraphy, in which the quirks and flaws of the artist-author’s brushstrokes act like the cracks in a work of Kintsugi, amplifying an idea’s value in analog form. We may see a new language produced and circulated only on paper, scrivener-style, to a sect of devotees who are sworn to prevent its digitisation. Or, a resurgence in oral storytelling, with the bards and griots who can recite the great pre-AI texts from memory revered as latter-day priests. Or even the veneration of everything published before 2023 as the emblem of an archaic, Promethean genius – of the time before muses of our own creation descended from a lesser Parnassus to prophesy in base, endless utterings.
One of GPT-3’s most astonishing capabilities is so-called “zero-shot” learning, in which the machine draws correct conclusions about things that are unknown to it. Writers, used to confrontation with the futility of our daily undertakings, understand zero-shot scenarios better than most. We mostly have zero shot at achieving our dreams as writers, for example. Now the futility deepens, because no writing, any longer, will be accepted at face value.
“Zero shot” thus works rather nicely as a moniker for the coming age of AI. Abbreviate it using a numeral: “0s”. The Gutenberg Age, succeeded by 0s. 0s forever.DM/ML
Ben Williams is the Publisher of The Johannesburg Review of Books.
Visit Daily Maverick’s home page for more news, analysis and investigations