Note: This column is also available free to the public. If you like my writings on genre issues, consider backing my Patreon or subscribing to my Substack.
Last month I explored the deceptive language used to describe machine learning. This deceptive language starts with calling machine learning "artificial intelligence" when there's no intelligence behind the systems, which instead use algorithms crafted from data samples.
Since then I've tracked other examples of deceptive language in discussions around AI. This time the poor language usage comes from Publishers Weekly, which tweeted "Sci-fi author says he wrote 97 books in 9 months using AI tools, including ChatGPT and Midjourney."
That tweet linked to a Business Insider article rehashing a Newsweek essay by author Tim Boucher, who said he "created 97 books in nine months with the help of AI." But the key words come later in the article, which reveals that Boucher's books are "between 2,000 to 5,000 words and feature 40 to 140 AI-generated images." As Kat Howard pointed out, "ASIDE from everything else, which is a lot, the third bullet point in PW's own article says the 'books' are 2000-5000 words. My friends, that is a short story."
And Boucher's self-written promo essay in Newsweek, titled "I'm Making Thousands Using AI to Write Books," also placed the most revealing factoid way down. Not until the fifth paragraph does Boucher reveal that "I sold 574 books for a total of nearly $2,000 between August and May."
So over 9 months, Boucher created 97 books and sold 574 for nearly $2000. But that equals at best just under 64 books a month, and spread across 97 titles that's less than one copy per book per month on average.
As Lincoln Michel said, "Okay I'm sorry but the ethics of 'A.I' aside, why is an author who has earned $250 a month being profiled in Newsweek?"
If the headline had said "I Sell Less Than 1 Copy of Each AI Book a Month," the reaction to that article would have been totally different. Or as Django Wexler said, this kind of "breathless stuff" about AI gets clicks. "But the actual product is pretty thin gruel.
And then there's the press release from Hachette Book Group around the upcoming publication of I Am Code, a book of "poetry" claimed to be written by a machine learning system. But as Shaoni C. White pointed out, "It's being sold as 'written by AI', and the sales pitch is that the LLM is a 'poet' with a 'mind.' But if you read closely, you can see that's wildly misleading to the point of being a straight up lie. They had a LLM generate over 10,000 'poems' — sections of text in the general shape and length of poetry — and the HUMANS involved cherry-picked the 1% that sounded cool. When you make a random text generator spit out 10,000 poem-like texts, some of them will be passable."
White's entire thread is worth a read but make special note of this point: "What the programmers did here isn't really anything groundbreaking, in terms of the history of poetry. Selecting cool bits and pieces out of a mess of uninteresting stuff is a whole genre: found poetry. When you cut out lines from a newspaper and glue them into a letter you don't say "Look, guys, this newspaper can write letters now! Any day now, writers will be replaced!"
In the end, the release of I Am Code is yet another example of manipulative language being used to promote machine learning systems.
Never forget the language we use matters, especially when the companies using manipulative language to describe machine learning have their own goals that won’t necessarily benefit the general public.
Or as Shaoni C. White wrote, "If your end goal is to pay writers less for their labor, then pretending that the LLM is the creative genius is a clever strategy."
Great piece - thank you! I wish more people weren't such absolute suckers with this stuff.