My Certified Organic, Biodynamic, Wildcrafted, Cold-pressed, Unfiltered Novel
Oscar Schwartz investigates a website that proves your art is human-made.
At the beginning of Michel Houellebecq’s 2015 novel, Submission, the narrator, François, a solitary, alcoholic bachelor who teaches literature at the Sorbonne, declares that the best books bear the human imprint of their author. “In the same way,” he says, “to love a book is, above all, to love its author: we wish to meet him again, we wish to spend our days with him.” Michel himself seems to be the abject exception to this rule, but in any case, whether the men at 23 First Corporation have ever read Submission, or consider themselves Houellebecqians, I cannot say. But I suspect they would agree with François’ argument, because it is precisely this type of author-reader romance that the Corporation is working hard to preserve.
To that end, the Corporation offers a service called ProudlyHuman™, which functions as a Voight-Kampf test for books. Send them something you’ve written, and they will certify, for a fee of $99, that it emerged from flesh and not silicon. This makes Alan Finkel and Trevor Woods, the Corporation’s founder and founding CEO respectively, something like self-proclaimed Rick Deckards—the bounty hunter in Blade Runner, whose job is to detect and “retire” androids masquerading as human—if Rick Deckard wore beige chinos, maintained an active LinkedIn account, and rented a modest, unadorned office in a co-working space on St Kilda Road.
When I arrived at their office last week, Alan Finkel, former Chief Scientist of Australia and Chancellor of Monash University, was browsing an article on The Conversation. He greeted me warmly, offered me a hot chocolate, which I declined, and then Scotch whiskey, presumably as a joke—it was 11am and Finkel, with his avuncular demeanour, didn’t strike me as a Don Draper type—before stepping out to get his own drink. Trevor Woods, a former university executive who had the chipper, well-shorn gleam of a corporate dad in a 1990s family sitcom, told me that we should wait for Finkel to return before beginning the conversation proper, so we gazed out the window and made small talk about the view over Melbourne Grammar. Finkel arrived minutes later clasping a paper cup of steaming hot chocolate. Then, we all sat down on ergonomic desk chairs, and Finkel began recounting how ProudlyHuman™ came to be with the meticulous detail of someone who relishes precise explanation.
The idea for human certification came to him organically, he said, in the first half of 2023, when he was just about to publish a book on Australia’s transition to renewable energy. ChatGPT had gone public six months earlier and a troubling thought popped into his head: what if people thought he used ChatGPT to help him write it? He asked his publisher if he could add an inscription at the front of the book declaring that he wrote it unassisted. Woods, who was listening with rapt attention despite probably having heard this origin story more than a few times, passed me a copy of the book in question, open to the publisher’s page. There, underneath the copyright attribution, was the following sentence: “Other than its behind-the-scenes involvement in web searching, Alan Finkel declares that artificial intelligence software had no role in the authorship of this book.”
The inscription, the first of its kind according to Finkel, offered him some peace of mind, but he knew that self-report ultimately wouldn’t cut it. What he needed was an objective system as trustworthy as Certified Organic for tomatoes or Appellation d’Origine Contrôlée for champagne. He enlisted Woods and a few part-time engineers, and together they developed a certification process that can now be accessed via the ProudlyHuman™ website, which features images of fedora-clad men playing in jam bands, painting palm-tree lined beaches, and other scenes one might expect to encounter after prompting ChatGPT with the sentence: “generate me an image of a creative man vibing out during the artistic process.”
The website outlines what ProudlyHuman™ deems to be acceptable use of AI. Research, spelling and grammar checking, editorial prompting: all ok. Generating paragraphs and chapters: verboten. This is called the de minimis standard, Finkel told me, and it was established by the U.S. Copyright Office, who will not copyright AI-generated text because under the American Constitution, copyright and patents are for people and people only. “We ask the author to sign a declaration in metaphoric blood that they have only used AI within these acceptable limits,” Finkel said. If the author agrees to these terms, they upload the manuscript, which is then processed with a “very sophisticated” system. If the system finds the text to be of human origin, the text gets certification. If it doesn’t, they can appeal, though Finkel and Woods are confident that their method is robust enough to get it right 99% of the time.
When I probed as to how the system works, Finkel said that they don’t disclose to avoid attempts at gaming it. Recently, an acclaimed philosopher uploaded two texts on the same topic to ProudlyHuman™—one that he wrote and another by Claude, after having arduously trained the model to imitate him. They caught him out. “Afterwards, this philosopher fed this back to Claude and asked how we did it,” Finkel said, “and Claude wrote back saying, ‘I don’t know. My congratulations to your friend. I feel humbled.’”
“Incredible,” Woods interjected.
“Staggering,” Finkel added. “That particular exchange convinced me that artificial general intelligence has arrived. This was not Claude just parroting some combination of words that led to that statement. This was an entity, a being, thinking an original thing.”
