Chatgpt is coming for books
And for journalism. And probably harder than you think
This is the 185th edition of SHuSH, the official newsletter of The Sutherland House Inc. If you’re new here, hit the button:
We are pleased to announce that the second edition of Sutherland Quarterly is now available for order. From the inimitable author and journalist Paul Wells, An Emergency in Ottawa: The Story of the Convoy Commission:
On Feb 14, 2022, Prime Minister Justin Trudeau made what might be the most controversial decision of his tenure, invoking the Emergencies Act to end a three-week occupation of downtown Ottawa by truckers protesting mandatory COVID-19 vaccine mandates. Proclaimed in 1988, the Emergencies Act is designed to give federal officials extraordinary powers in the event of threats to Canada's national security that can't be managed under existing laws. Trudeau used it to make the protest illegal, freeze the accounts and cancel the vehicle insurance of participants, requisition tow trucks to clear protestors from the streets, among other measures. The government defended the first-ever invocation of the act as just and necessary; several premiers and the Canadian Civil Liberties Association called it an assault on democratic rights and civil liberties. As required by the act, Trudeau appointed a commission of inquiry into its use. Last November, justice Paul Rouleau held three weeks of riveting hearings that included testimony by so-called Freedom Convoy organizers, police officials, cabinet ministers, and Trudeau himself. Award-winning author Paul Wells was a regular visitor to the inquiry. Witnesses described layer on layer of dysfunction and acrimony in every organization that converged on Parliament Hill—three levels of government, three police forces, and the protesters themselves. How does a society make crucial decisions when everyone is exhausted, nothing works, and the noise from the truck horns and the shouting is deafening? And how do the protagonists regroup to make their case in the weird and sterile environment of a public inquiry? That's the story-inside-a-story of the Emergency in Ottawa.
Single copies will hit stores in early April, $19.95 (plus HST); the subscription price is 20 percent off the cover price or $67.99 (including HST). I hope you’ll consider subscribing.
These short books are timely, topical, and highly readable. They do not come from one particular worldview or represent one style of writing. Each contains responses to the previous essay to create a sort of rolling conversation from book to book. While our first two authors are established authors, we will also be publishing new voices (and we’re open to submissions).
I ran into Steve Forbes, the two-time candidate for the Republican presidential nomination and editor-in-chief of Forbes magazine, at an event in Washington two weeks ago. Steve was in a panic about Chat Generative Pre-trained Transformer, better known as ChatGPT.
As you probably know, ChatGPT is an artificial intelligence bot that mimics human conversation. It is available for general use and can be asked to write essays and song lyrics and computer programs and conversations about death anxiety between two slices of bread (“Who knows, maybe we’ll end up being part of something great like a delicious sandwich or a warm bowl of soup.”).
The bot does its work almost instantly and usually to an acceptable level. In a blind test, one of its answers to an exam question at the highly selective Wharton School received a grade of B. Of course, mischievous users have bullied it into write instructions for Molotov cocktails, Neo-Nazi propaganda, and convincing yet unsound abstracts of scientific papers. Students have used ChatGPT to cheat on assignments. The service has convinced some parents that their children will never learn to write a paragraph on their own.
We are early on this journey. ChatGPT is not a fully realized product but an initial iteration of generative artificial intelligence—AI that produces original content rather than merely acting on or analyzing existing data. There are lots more to come. Microsoft’s Bing released its buggy, argumentative, and emotional version of ChatGPT a few weeks ago. In conversation with an editor at The Verge, it claimed to have hacked, fallen in love with, and killed one of its developers at Microsoft. We’re now hearing calls for federal regulation of these services.
Google is working on Google Bard, its competitor to ChatGPT. Meta claims to have a version but has yet to decide whether to release it. China’s Baidu corporation expects to unleash its service this month, as does the South Korean search engine Naver. We’ll soon be well supplied with competing, fast-improving generative AI services and swimming in the oceans of content they’ll produce. That’s why Steve was in a panic.
Let’s back up a bit. Forbes was founded in 1917 by Steve’s grandfather, B. C. Forbes. Through most of the twentieth century, Forbes was among the world’s best sources of business news and commentary. It was also one of a handful of publications through which advertisers could target high-net-worth business readers. If you were selling business travel or luxury watches or Cadillacs, you needed Forbes (and Fortune and the Wall Street Journal) and you paid a couple hundred thousand dollars a month to put one of your messages on one of its pages.
Yes, business people also listened to radio and watched television and read the local newspaper, but audiences for those products were mass, meaning that the advertiser would be paying to reach a mess of undesirable readers along with the few desirable ones. That’s an inefficient ad buy. Forbes was super-efficient, with all the right readers and only the right readers. The magazine was a license to print money. It was run for many years by Steve’s father, Malcolm Forbes, who became very rich. Chateau-in-France-private-Boeing-727-best-friends-with-Elizabeth-Taylor-and-one-of-the-world’s-best-Fabergé-egg-collections rich.
Along came the internet and the volume of content available to readers grew exponentially. Advertisers now had an abundance of ways to reach high-net-worth individuals. Ad pages priced at hundreds of thousands of dollars—a difficult sell.
By 2006, the Forbes family sold a minority interest in its company to Elevation Partners. In 2010, it sold its stately headquarters on Fifth Avenue to New York University. In 2013, majority control of the company was unloaded to Hong Kong’s Integrated Whale Media.
By the time COVID hit, Forbes was mostly a website, Forbes.com. It produced some good journalism but churned out a high volume of low-quality content, some of it from contributors paid by the traffic they generated, some written by the site’s advertisers. It made no sense for the company to spend a lot on its digital content because it wasn’t worth a lot to advertisers or to readers. Too many alternatives.
By 2022, the new owners were raising “strategic investment” funds from the cryptocurrency firm Binance (which US Senators this week accused of being “a hotbed of illegal financial activity”).
That, in a nutshell, is the decline and fall of North American newspapers and magazines.
Generative AI has the potential to make this ugly situation infinitely worse. If the internet introduced a flood of new content, ChatGPT and its competitors, says Steve, will produce “a tsunami.” That means way more competition for both readers and advertising dollars and way less revenue for his company.
You might think people don’t want to read stories and commentary by machines. There’s inherent value in human intelligence and the human voice. And you may be right, but to a large extent that doesn’t matter.
We celebrate journalism for its big investigative pieces, in-depth features, and fearless commentary, but most of what journalists do is routine: report a press conference, summarize a court decision or a government report or a company’s annual report. Even before ChatGPT, machines were writing basic earnings reports for business sections and game reports for sports pages. They’ll soon be doing a large percentage of what journalists do, infinitely faster than journalists do it and in infinitely higher volumes, at no cost (or minimal cost) with similar if not improved accuracy and reliability.
Human-generated news and commentary will struggle to be seen in an ocean of increasingly accomplished bot-generated material, all of it optimized to outperform human product in algorithmic searches. The machine story will be good enough for most people and for most purposes. The content churned out by today’s media companies will be somewhere between worth less and worthless, making it difficult for those companies to afford original, high-value human journalism.
Probably there will still be a market for original, high-value human journalism, but it will shrink. Some of us will pay for a 5,000-word investigative piece from the Globe & Mail; some of us will settle for a free full-length replica produced by a bot trained to avoid plagiarism and copyright infringement; some of us will ask ChatGPT for a 500-word summary in the voice of John Oliver; some of us will forego the Globe story to read a dialogue between a vampire and a unicorn on the merits of drinking blood vs. eating rainbows.
This will happen in years, not decades.
After learning that Steve was worried, I wondered if I should be worried. The book industry is not dependent on advertising, meaning generative AI will affect us differently than magazines and newspapers, but the same principle applies: the more content the world produces, the less most of it is worth.
We saw this when the internet hit book publishing.
First, online retailers such as Amazon and Bookfinder.com drastically increased the number of books available to purchasers. If you were looking for something to read in the early nineties, you were limited to the choices available in local bookstores, which might have carried between 5,000 and 40,000 books. The web made many millions of titles available—frontlist, backlist, and long out-of-print—at the touch of a button. Every new book was now competing against every book ever published anywhere. Try raising your prices in that environment.
Second, the internet abetted self-publishing. Go back to 2005, before self-publishing was a thing, and traditional US publishers produced 172,000 new titles. As of 2021, traditional publishers were releasing close to 300,000 titles per year while self-publishing was generating 2.3 million. Self-publishers rarely sell a lot of copies—most of their books aren’t very good—but there are enough of them to claim half of total e-book revenues and a good chunk of paperback revenue. They’ve devastated the genre fiction business (romance, sci-fi, thrillers) at conventional publishers. As noted in SHuSH 169, Penguin Random House saw three-quarters of its sales in those categories disappear to the self-published market between 2011 and 2019.
By some estimates, the US is now producing as many as four million books a year if you count trade publishing, self-publishing, reprints, public domain literature, educational publishing, and so on. That flood of literature is not a problem for Michelle Obama or Prince Harry or JK Rowling or John Grisham. Their books everybody knows about. It’s a problem for almost everyone else because it’s so hard to get your book noticed. Amazon is a literal five-kilometre-wide, 6,400-kilometre-long river of books.
Thanks to ChatGPT and its ilk, that flood of books will become a tsunami. ChatGPT’s next iteration, coming shortly, will produce a 60,000-word book in about 20 seconds. It will not be great literature, but good enough for most people and most purposes and very inexpensive to produce.
I expect ChatGPT won’t be used to write whole books so much as to help authors write books faster. There are romance and fantasy novelists who now produce a book every two months. In the future, they’ll be doing two every month. It helps that ChatGPT can be trained to write in your voice. Simply create a dataset of your own writing and process it through a tool like Hugging Face’s Transformers library or OpenAI’s GPT-3 API. The bot will pick up your use of language and other writing patterns and generate copy approximating your personal style.
(Just this morning, Publishers Lunch reported that The Authors Guild has added a new clause to its model book contract prohibiting publishers from “using or sublicensing books under contract to train artificial intelligence technologies.” The guild is spooked that digital publisher Findaway Voices is working with Apple on machine learning while Bookwire has teamed up with Google Books for AI-narrated audiobooks. It hopes to stop data miners from training AI models to compete with human work. )
On the non-fiction side, ChatGPT will write or help write guide books, basic biographies, basic histories, basic personal finance, basic personal advice, basic diet and health books, puzzle collections, and how-to series. Those categories produce a lot of the ballast at publishing companies large and small.
If we’re now producing four million books a year, we could easily be producing ten or twenty million a year by 2030. Amazon and other sites will be clogged with product, making it more difficult for human books to get discovered (ChatGPT’s metadata is likely to be more algorithmically savvy than yours). That will undermine the value of a lot of what’s published today by traditional publishing firms and make it more difficult for those companies to afford original, high-value human books.
In a nutshell, generative AI has the potential to destroy a lot of value in the literary world without producing a single great work of literature.
Will it happen? One minute I think of Forbes grasping at lifelines from crypto shysters and think it will. The next I remember that Amazon AI has almost thirty years of data on me and has yet to make a decent book recommendation.
Lend SHuSH a hand
A weekly newsletter like SHuSH is a lot of work. We don’t charge for it or present advertising or ask for donations, but that doesn’t mean we wouldn’t appreciate a little support. The best way you could help keep us going is to subscribe to our Sutherland Quarterly.
Our regular reminder to readers to support independent booksellers. Click this link to make the above map come alive.
Or click here to visit the Sutherland House website and order one of our latest:
Our Newsletter Roll (suggestions welcome)
Steven Beattie’s That Shakespearean Rag, a newsy blog about books and reading
Art Kavanagh’s Talk about books: Book discussion and criticism.
Gayla Gray’s SoNovelicious: Books, reading, writing, and bookstores.
Esoterica Magazine: Literature and popular culture.
Benjamin Errett’s Get Wit Quick, literature and other fun stuff
Lydia Perovic’s Long Play: literature and music.
Tim Carmody’s Amazon Chronicles: an eye on the monster.
Jason Logan’s Urban Color Report: adventures in ink (sign-up at bottom of page)
Anne Trubek’s Notes from a Small Press: like SHuSH, but different
Art Canada Institute: a reliable source of Canadian arts info/opinion
Kate McKean’s Agents & Books: an interesting angle on the literary world
Rebecca Eckler’s Re:Book: unpretentious recommendations
Anna Sproul Latimer’s How to Glow in the Dark: interesting advice
John Biggs Great Reads: strong recommendations
Mark Dykeman’s How About This: Atlantic Canadian interviews and thoughts on writing and creativity.
THAT’S IT FOR THIS WEEK. THANKS FOR READING. PLEASE SIGN UP OR CONVINCE SOMEONE ELSE TO SIGN UP, OR SHARE, OR LEAVE A COMMENT:
It's our bodies that will save us. Anything that requires going somewhere, talking to another human, witnessing something (while Mx Chat can read books, can it watch live opera and do up a review? live theatre should shut down online live streaming while they're on time), the engagement of the senses, the occupying of and moving through space. Evaluative mind, emotions. Any discipline that's moved everything online and relies on teleconferencing will be in trouble faster. It's a mistake, I think, that psychotherapists, for ex, are now meeting clients chiefly via zoom when body language used to be a big part of the job (or, two unconscious-es, in the same space, communicating, somewhat unpredictably). The animal bit, the glitchy bit, might yet be our salvation, paradoxically.
I am hoping that I can continue to find curated, quality content by humans by buying direct from small publishers like Sutherland and university presses. And reading directly on The Atlantic, The Athletic, Globe & Mail, The Line, The Hub, The Hill Times...I don’t really see the dreck or understand what people mean when they say that Google serves up the news.