AI book witch hunt: BookTok rails v. The Atlantic’s Will Oremus article on Tuhin Chakrabarty’s study that suggests Daggermouth is partly AI-generated book based on AI detector Pangram & rare expressions. H.M. Wolfe denies it. BookTokkers defend her.
Will Oremus’s article on “Is This What Comes After AI Slop?, Daggermouth could be the first best-selling novel partly written by a chatbot.”
BookTok is a fascinating community. The Atlantic’s Will Oremus may soon found out how fascinating–and fierce. This past week, Oremus wrote a 1,500 word essay, provocatively titled “Is This What Comes After AI Slop?, Daggermouth could be the first best-selling novel partly written by a chatbot.“
Oremus was summarizing a pre-print study posted by Professor Tuhin Chakrabarty; his article was also provocatively titled, “Generative AI floods and dilutes the market for books.” Chakrabarty conducted a study on the potential effect of AI-generated books on books written without AI.
To determine if a book was AI-generated, Chakrabarty used a combination of the AI detector Pangram (which has some false positives) and the additional check of looking for “rare expressions” that supposedly are not common in human published works: “We call an expression rare when it appears in at most five Google Books volumes and is absent from a 4.7-trillion-token snapshot of the internet.”
Although Chakrabarty’s pre-print study doesn’t identify any author or title of the books analyzed as far as I can tell, Oremus apparently learned from his interview that one of the books that Chakrabarty’s study flagged as partly AI-generated was H.M. Wolfe’s Daggermouth. And Oremus’s article is almost entirely about the Daggermouth aspect instead of the entire study.
Although Oremus concedes that Pangram has false positives and that “[t]he study has yet to be peer-reviewed, and to be clear, there is no definitive proof that Wolfe used AI,” Oremus also quotes the opinion of Mohit Iyyer, a professor of computer science at the University of Maryland: “it’s ‘almost statistically impossible’ for an entirely human-written book to register as 60 percent AI” as Daggemouth’s book reportedly did.
Indeed, The Atlantic’s subtitle reads like clickbait: “Daggermouth could be the first best-selling novel partly written by a chatbot,” insinuating that the book may well be partly AI-generated.
Oremus’s article on Chakrabarty AI study enrages BookTok
Well, I thought it would be a lazy Sunday but my TikTok feed lit up.
Many BookTok influencers were engaged by Oremus’s article on Chakrabarty’s AI book study. I don’t think I’m exaggerating in the use of the word “enraged.” Take a listen to some of the BookTokkers. Warning: the language may contain expletives.
Some BookTokker criticized Chakrabarty’s study’s use of an AI detector Pangram:
Will AI Defamation Lawsuits Be Next?
One BookTokker predicts AI defamation lawsuits may become a thing.
There already have been several defamation lawsuits filed by students who faced disciplinary actions by universities for alleged AI work product.
I did buy a copy of the book
Maybe the controversy worked. I did buy a copy of the book. And, yes, the book is on top of the latest issue of The Atlantic. Is that a sign of how this controversy will turn out?