The Century-Old Workhorse That Just Might Save Writers From AI

As someone who has been publishing books for twenty-five years, I naturally worry about the threat to my livelihood from AI. I’m a slow writer. It takes me many years to write a book, whereas a writer using AI can toss dozens off at light speed (and brazenly brag about it in the New York Times).

Writers who are just starting out, who are honing their craft in the same way writers have done for centuries—by reading and writing and reading some more—have good reason to worry that AI will diminish their chances of ever getting liftoff. Not only must writers compete against books “written” by generative AI; now we must also prove that we wrote our own books.

The latest “did a human really write it?” fiasco clearly highlights the importance of something that writers have relied on for more than a century: copyright.

I’ll get deep into how copyright may be our best bulwark against literary annihilation later in the post. But first…

There’s a good piece in Publishers Weekly this week by Edward Nawotka about the debut book by Jerry Falade that was pulled by the author’s agents due to allegations that the author used AI to write it. Falade has vehemently denied the allegations.

Looking ahead, as these cancellations continue to pile up, the industry may soon be forced to codify policies to address similar situations as they arise.

Sooner rather than later, please. No author should lose a contract or have their reputation damaged based on unproved allegations. And no agent should go out with a book, putting their reputations on the line, only to discover that it was written with AI. Literary agents work for years and decades to build and maintain trust with acquisitions editors and publishers.

Obviously, I have no idea what happened in this case and am not making a judgement about the agency or the author. It would be reasonable for an agency forego a hefty commission if they had clear reason to believe that the book they had sold on good faith as an original work was not fully human-authored.

That said, authors also need to know—and give consent—if their manuscripts are going to be fed into AI-detection software by agents or publishers. There is a legitimate concern that authors’ original, unpublished works will be saved on insecure servers and used in nefarious ways, eroding authors’ ownership over their own work.

There’s also a legitimate concern that the detection software just might get it wrong, very wrong. Dr Sam Illingworth, who writes the Substack Slow AI, “a newsletter for people who refuse to outsource their judgement to a machine,” shares an alarming study in this post:

As I have written about numerous times now, researchers at Stanford, writing in 2023, ran seven widely used detectors over 91 essays written by non-native English speakers. Across the seven, the average false-positive rate on that human writing was 61.3%, and 97.8% of the essays were flagged as AI by at least one detector.

That was three years ago, so it’s possible that detection software has gotten better since then, but there’s still a lot of wonky software floating around out there. Do we really want to invest these tech companies with the power to ruin the lives of writers?

There needs to be an industry standard about what constitutes acceptable use of AI. I think if a group of publishing professionals sat around a table trying to come to a consensus on what constitutes acceptable use, there would be rigorous debate, but there would also be some points of clear agreement: yes to AI-aided research, perhaps, and a hard no to AI-generated sentences. But until there can be a clear, unified standard among publishers, each publisher needs to create individual standards, clearly stated on their websites, about what they expect from authors who submit to them.

Just in the past week, Open AI revealed numerous instances where the AI “escaped” the testing environment. What’s to prevent an original, unpublished novel from being made public, without the author’s consent, before the author even has a book deal, because the entire manuscript was fed into AI software? The tech companies don’t care about writers. They care about profit.

There is a legitimate concern that authors’ original, unpublished works will be saved on insecure servers and used in nefarious ways, eroding authors’ ownership over their own work.

After all, long before the public had heard about AI, Google was scanning copyrighted books without the consent of authors or publishers and using those books to serve ads, at profit to Google. This from the corporation whose motto was, hilariously, “Do no evil.” And we now know that Google was also using those scans to train LLMs.