Are you human, centaur, cyborg or machine?
Poking the AI beast
This month, I’ve been engaged in what seem like opposite ends of the writing spectrum: thinking about specific words and sounds while also exploring the world of Generative Artificial Intelligence (aka AI or GAI—ChatGPT, Claude, and many many others). So this newsletter is going to feel a bit two-headed—more orthrus (the mythical two-headed dog) than centaur (body of a horse, torso and head of a human).
Tools: Exploring AI
So, first up, some jargon and lines you can drop in your next water cooler (ha!) or Zoom chat:
Centaurs and cyborgs: Centaur tasks are those that require or involve human and AI input on a more or less equal basis; cyborg tasks are mostly AI with a light dusting of the human hand.
Human in the loop: AI tasks that need some human guidance to make them fit for publication, human use, etc. So, at a recent conference on publishing and AI, I heard more than one speaker talk about processes that need, at a minimum, “humans at top and tail”—human interaction at the beginning of the AI process and to review/massage the output. And for many AI-involved tasks, far more human involvement is involved.
IP leakage: Short for “intellectual property leakage.” What it means: when you expose your content to AI—say by uploading a chunk of your book to get “the machine” to generate a synopsis—without fully realizing that your content is now part of “the machine’s” brain, its content pool. This could happen, for instance, if the publicist at your publishing company uploads portions of your book to get an AI’s assistance in writing a press release about your book—which is why some agents are insisting on “no AI” clauses in contracts, so their clients’ work won’t be exposed to AI (at least via their publisher).
AI “hallucinates”: It makes up stuff, stating it as fact. If you ask it for footnotes, it might give you real footnotes, wrong footnotes (actual articles but not connected to what it is footnoting), or made-up footnotes (fictitious articles but the format of the footnote looks real). AI promoters say this will improve, especially as AI is linked more closely to search—but it is also a function of what it is searching, since AI doesn’t “know” what is true or false.
Something to be aware of: If I type in a prompt and you type in the exact same prompt to the same AI tool, we will get different answers. The AI “trains” itself based on its past interactions with individual users, so that is part of what is going on. It also expresses itself differently each time out. There may be some overlap, but it’s very unlikely that the content will be exactly the same.
Amara’s Law: We tend to overestimate the short-term impact of new technology, but underestimate its long-term impact. Translation: we might be catastrophizing about the wrong things.
The Gartner Hype Cycle: When a new technology emerges, we have super high expectations of what it will do. This is quickly followed by the “trough of disillusionment” as we discover its flaws and problems. Then we figure out how to use it productively and we end up somewhere that is higher than where we were before the new tech appeared, but lower than the initially super-high expectations.
Repeated by multiple sources/speakers, some variation of the following: “Today’s AI is the least intelligent AI/worst AI you will use.” Which is both terrifying and exciting.
Current general problems with AI
It is often “confidently wrong,” as in, it sounds like that guy in your first-year poli sci class who had thoughts on every. single. thing. It is, in short, an expert bullshitter.
It is just chock full o’ bias. AI basically duplicates the patterns it sees in the material it was “trained” on, and since systemic and structural bias is baked into so much of the content that already exists, AI is just as biased and even magnifies that bias.
The writing it produces is kind of mediocre. Much of the written content it creates is passable—so sounds ok for a high school or undergrad university paper—but is still pretty mediocre because it’s designed to be bland. But those in the know—various speakers, watchers, and others with AI experience—say that could change quickly and exponentially.
There is no reliable way to spot whether written content has been produced by AI, and tools that say they do this disproportionately incorrectly identify content produced by people for whom English is a second language as being produced by AI. Which is a huuuuge bias issue, especially in university or educational settings.
Immediate problems for writers specifically
Low quality AI-generated content is already flooding some platforms—certain categories in Amazon, for instance—and diminishing the visibility and findability for quality content. This is happening with AI-generated websites and instant books (travel is one category being affected by this ) in particular. This is bad for writers, but it’s also bad for readers, researchers and anyone trying to find quality info on particular topics. Some have dubbed this “enshittification.” You’ll have to wade through more shit to find the content that is actually useful to you. (See also “don’t take advice on mushroom foraging from AI-created books.” )
Copyright infringement: If you have published books or have other material published online, your work may have been used to train AI tools. And The Writers Union of Canada, Authors Guild and others think you should be compensated for that. AI may also grab bits of your content and feed it into its output, so that if the person generating that content with their prompt then uses it in their own work, they may inadvertently be plagiarizing you. Which is a problem for the original author and for the person interacting with the AI.
How authors can/are using AI
Lots and lots of folks are using AI as a kind of personal assistant to create “first drafts” of press releases, marketing plans, websites, social media posts, and other content that is more or less templated or predictable. The quality of what you get depends in part on the quality of your “prompts”—what you feed in. So, for a press release, saying “write a 300-word press release for a book about spiders” is going to get you a very different—and less useful result—than if you say “write a 300-word press release aimed at consumer-oriented science publications, written at a grade 10 reading level, for a book about spiders titled Spidey-Sense. Here is the book synopsis [insert synopsis] and author bio [insert bio].” (See the links below for resources directly related to using AI for book marketing tasks.)
Some are suggesting that AI could be used to assist/replace proofreaders, copy editors, indexers. Likelihood is that projects deemed more important by a publisher would have more human “touch” (more centaur than cyborg) while those deemed less important—more mass or low-profile products—would be more machine-managed (more cyborg than centaur) on these processes. And those who self-publish may also opt to use some of these tools in this way to save time and money.
Some writers are using AI to get past “blank page-itis”—so typing in some prompts and then using that as a jumping off point for their own creativity or work. Others are using AI to suggest book, article or section titles. Some are using it to suggest character development—ie writing a prompt like “Three siblings are fighting over their dead father’s estate. What are some of the psychological issues that might be at play?” I used that prompt with ChatGPT, and it generated a list of 12 psychological issues (with short definitions)—grief and loss, sibling rivalry, inheritance anxiety, control and power, perceived fairness, emotional attachment, communication breakdown, external influences, cognitive biases, legal battles, family history, and ego and identity—that might be at play. Would that go into a short story about these three siblings? Not as written, but it might prompt me to consider motivations or story possibilities I hadn’t explored.
Some are exploring the creative opportunities/challenges of “co-authoring” with AI: Stephen Marche, Sean Michaels, Sheila Heti, to name just three.
Some are using AI to create instant content and flooding self-publishing platforms in the hopes of generating quick returns. There are lots of YouTube videos and tutorials of all sorts out there “training” people on how to do this. (I’m not going to link to these. It’s a bad thing to do.)
AI and book contracts/copyright
Some publishers are already including clauses in their contracts requiring transparency about the use of AI in authors’ work. And one could argue that existing contracts, which include language around warrantying or guaranteeing that you are the creator/copyright holder of the content you are submitting, already preclude the inclusion of AI-generated material.
A US court ruled that AI-generated content was not copyrightable because copyright applies to human creations. However, the “human” contribution of material that is co-created by AI and human could be copyrighted—though how they will untangle what that human contribution is and so what is actually copyrighted remains to be figured out.
As noted above, some agents are inserting “no AI” clauses in publishing contracts to stop publishers or their vendors from uploading author content to AI tools.
Thinking points
Terrifying? Markus Dohle, former president of Penguin Random House, said in a speech at a Publishers’ Weekly conference on AI that previous technological advances that affected publishing—the internet, search engines, e-books and audiobook tech—were all primarily about “reach.” They affected how publishing products—books, content—reached consumers. AI, on the other hand, is set to affect creativity—how books and other content are created. And the key issue will be how to ensure copyright protection and author compensation for human authors so that there continues to be a book industry that involves human authors. How do we “protect human creativity” and fairly compensate human creativity? Dohle says he’s optimistic that we’ll figure it out. See above re “exciting and terrifying.”
Tailored: In the not-too-distant future, there could be AI tools that consumers can use to “produce a targeted book on the fly”—so, I could ask for a 60,000-word murder mystery set in Paris with a female detective, including a scene set in my favourite perfume shop (Parle Moi de Parfum), written for someone with an IQ of 110—and the machine would load a custom book into my e-reader. Or AI could be used to “instantly” create the book version of a film I like, or a graphic novel version of a book. Or a bunch of other possibilities I’m too tired to imagine.
Training: One issue raised by attendees and speakers at that PW conference: if publishers replace all of those entry-level tasks with AI, how do editors and writers learn their craft? As a young writer, I wrote all kinds of press releases and product copy and short service articles; I proofread Harlequin Romances; I copy-edited thousands of pages of material; I struggled to write headlines and subheads and deks and pull-quotes. And all of that helped build the skills I use now to write books. I’m not sure how I would have developed those skills otherwise. Maybe today’s 20-year-old versions of me will build their muscles in other ways. Kind of like having to go to the gym rather than plowing the field...
Transparency: One poll of readers by International Thriller Writers found that readers overwhelmingly want publishers to be transparent about whether something was written by AI or a human. Will that desire remain solid as AI gets better? Will readers expect AI books to be cheaper? Who knows?
Terrified, p. 2? What does this mean for nonfiction writers? In a way, I think we’re a bit more “protected” because of the nature of some of our work, which involves original research and interviewing, or writing based on reflecting on personal experience and human emotion. Can AI produce content that resonates with humans? Yup, it can—at least in collaboration with a human. Who knows what will be possible tomorrow, or next month, or next year.
Want to try it out?
I think all writers should play with the tools, to at least understand what is going on and how they work. But I don’t think anyone should upload pages of their own creative work into an AI tool—or at least, I won’t be doing that.
If you’re ready to explore, here are some links.
AI Text Tools (by no means an exhaustive list)
AI and book marketing
20 examples of AI used in book marketing
Image AI platforms
Getty Images will be launching its own
AI and video content
AI and audio content
Sounded (for audiobooks)
Craft: Let’s get back to words
Does your head hurt from reading all the AI stuff? Mine kind of does. And maybe that’s why the other thing I spent time on this month was just playing with words, and listening to people talk about playing with words.
One was a session with Diana Goetsch, poet, teacher, memoirist and author of This Body I Wore. Diana did a wonderful one-hour session for Paragraph NY on poetry for prose writers. I’ve taken other courses with Diana and recommend her highly (you can access recordings of some of her past courses on the Paragraph site for a quite reasonable cost).
I’m not a poet and my writing education came via journalism school, so I never learned all the wonderfully specific terms for various sonic devices and strategies. I knew alliteration, of course, and onomatopoeia, from my high school English classes, but others I didn’t know had names include:
Anaphora—the use of a repeated phrase, such as “I have a dream...”
Spondee—two words that need to be said slowly because they are similarly emphasized or accented, such as in the Gerard Manley Hopkins poem Pied Beauty:
“With swift, slow; sweet, sour; adazzle, dim;
He fathers-forth whose beauty is past change:
Praise him.”
Caesura—a stop or a pause in a line. This can happen mid-line (common apparently in Old English poetry) or at the beginning or end of a line, where a single word might be set off for emphasis: Angry? I’ve never been so angry in all my life. Or: I wasn’t sure what to say anymore, and so that’s what I said. Nothing.
Various forms of rhyme
o Eye rhyme—where a word’s spelling is echoed but not its pronunciation, such as through and rough
o Internal—rhyming within a line, or words that have internal sounds or syllables that echo each other
o Masculine—rhymes ending in a stressed syllable, such as “rat” and “cat”
o Feminine—rhymes ending in unstressed syllables, such as “slicing” and “dicing”
Diana’s advice was to “edit with your ears”—to read your work aloud, listening for the lumpy, awkward, dragging sections. It’s something I’ve always done. But she also had another piece of advice I hadn’t tried: to “read with your ears”—to let go of the meaning as you read something aloud, and just listen to the sounds and rhythms, the musicality of the words, noting where you might shift those sounds to improve the musicality.
A few days after Diana’s session, I started reading Lauren Groff’s wonderful new book The Vaster Wilds, and with Diana’s advice echoing in my head, I read Groff’s words with new appreciation. I was dazzled by her last book Matrix, and as I am reading this one (I’m both looking forward to and mourning the fact that I’ll likely finish it tonight), I am delighted by the rhyme and rhythm, the poetry, of her lines. Groff is particularly good at the unexpected verb. Example: the opening two lines of the book. The first is a somewhat familiar image. But the second line, with its muscular verb, stings, in a line that matches the first in syllables and cadence:
“The moon hid itself behind the clouds. The wind spat an icy snow at angles.”
Later:
“The world, the girl knew, was worse than savage, the world was unmoved.
It did not care, it could not care, what happened to her, not one bit.
She was a mote, a speck, a floating windborne fleck of dust.”
Prose—but poetic prose, with the repetition of words (world and care); internal rhyme (it and bit; speck and fleck); and metre (the matching number of syllables between commas within lines).
And more of those verbs: “They were let inside a cool white space so sweet-smelling that the girl’s nose drank down the smell as though it were a honeyed draught.”
A few lines later: “A servant wafted by and said that the mistress would see them now upstairs in the hall.”
I love that the nose “drank” and the servant “wafted”!
Exercises
Make it a pretty ditty: Take a line you are struggling with and use repetition, internal rhyme or metre to rework the line.
Muscle up your verbs: Choose a line from a current piece of work. Circle the verbs. Replace at least one of them with something unexpected.
Put on your fancy pants: Take a paragraph, and rework it to introduce a caesura, spondee or anaphora.
Feed your eyes: Read something beautiful. Speak it aloud. Chew the words. Swallow, and nourish your writer’s soul.
Choice words
Some things I’ve underlined this month:
“Naming, she understood, made things more visible.”
Lauren Groff, The Vaster Wilds, p. 164
“The seedlings smelled of midnight and hot stones in the rain.
Like her.”
Deborah Levy, August Blue, p. 162
“No one was young in this war, not even a drummer boy of thirteen, and certainly not a laundress of nineteen.”
Alaya Dawn Johnson, Reconstruction, p. 247
“When the community rallies to the victim’s support, vengeful feelings are transformed into shared righteous indignation, which can be a powerful source of energy for repair. It is only when victims are denied their fair measure of justice that their anger can fester as helpless rage.”
Judith L. Herman, Truth and Repair, p. 47
Other stuff
Two ripping reads I’d highly recommend:
Genevieve Scott’s novel The Damages. As someone who was in university in the mid-1980s, this novel—with its roots in 1990s university life—resonated. It does a masterful job of unraveling the conflicts and confusion around sexual assault and its impact, what we told ourselves was normal, what we thought we should just deal with, what we didn’t always recognize we needed to reckon with. The story had me racing to the end; its resonance stayed with me long after I finished. (Disclosure: Genevieve teaches in the MFA in Fiction program at King’s, where I also teach.)
R. F. Kuang’s novel Yellowface. The cover blurb from Time magazine says it’s “razor sharp” and indeed it is. If you’re interested in publishing and writing and the issues of cultural appropriation, social media and more, this one will keep you turning the pages. You’ll cringe, you’ll smirk, you’ll gasp. And you’ll think.
Obligatory photos of Buddy and Reggie
Still keeping our distance.
The stuff at the bottom
I’m a writer, editor and teacher. This is my personal e-newsletter on the craft of writing nonfiction, sprinkled with occasional feminism and social justice. You can find out more about me on my website at kimpittaway.com. You can also find me on Facebook. I’m a cohort director in the MFA in Creative Nonfiction limited residency program at the University of King’s College in Halifax, Nova Scotia, Canada. If you’re interested in writing a nonfiction book, you should check our program out! (And hey, we just added a limited residency MFA in Fiction as well, taught by some of my amazing colleagues!)
Loads of stimulating insights, beautifully packaged. Thanks, Kim.
A rich post, as always (could have been two posts). Like many of us, I see a lot of profound anxiety--terror, almost--about AI trashing the essence of art and replacing artists. AI can be taught to use the poetic techniques you describe here, caesura? Just tell it how many. What it can’t do is reflect on experience and be stirred by emotion. Will it learn what makes us human? I doubt it. And of course it will not be capable of the research and analysis required for creative nonfiction.