Authors and Artificial Intelligence

There is a certain irony, perhaps even hypocrisy, in the publishing industry’s increasingly absolutist attitude toward authors and the use of artificial intelligence. In most industries, no one particularly cares whether a spreadsheet formula, grammar checker, Photoshop filter, digital audio plug-in or sophisticated piece of software helped create the finished product. We judge the result.

Books, however, occupy a different cultural space because the author is understood to be part of what the reader is purchasing. The author’s voice, imagination, judgment, memories, research, experience and choices are inseparable from the work itself. That makes artificial intelligence especially troublesome. If readers believe the machine has supplied those essential elements, they may reasonably feel that the implied contract between author and reader has been violated by dishonesty.

For the publishing industry, the issue is further complicated by the origins of artificial intelligence itself. Publishers and writers argue that many generative AI systems were developed, in part, by training on vast collections of copyrighted material without permission or compensation. Whether one accepts every legal argument surrounding that dispute or not, it explains why the subject has become unusually emotional. Writers are not simply confronting another technological innovation. Many believe they are confronting a technology that learned its craft by consuming theirs.

Copyright presents another legitimate concern. If substantial portions of a book are generated by artificial intelligence, questions arise about what portions of the finished work are actually protected by copyright and precisely who—or what—can claim authorship. Publishers dislike uncertainty, especially when it involves ownership.

Traditionally, a commercially published manuscript will pass through many hands before reaching the reader. A developmental editor can dramatically restructure a book, recommend the elimination of characters, move chapters, alter pacing and suggest major changes in narrative direction. A line editor may work sentence by sentence to improve style, clarity, rhythm and flow. Then come copy editors, proofreaders, fact-checkers and, increasingly, sensitivity readers. Yet, after all of that intervention, one author’s name remains on the cover.

Apparently, assistance itself has never been the real problem.

Consider ghostwriting. Entire books have been written, substantially rewritten or heavily shaped by professional writers under the name of a celebrity, politician, executive or public figure. Ghostwriting is a substantial business and accomplished ghostwriters may command tens of thousands of dollars, or considerably more, for a single project.

Prince Harry’s memoir Spare, written with the acknowledged assistance of J. R. Moehringer, is a modern example of collaborative authorship. John F. Kennedy’s Pulitzer Prize-winning Profiles in Courage has likewise been the subject of decades of discussion concerning the contributions of Ted Sorensen and others. Publishing has somehow survived these arrangements without concluding that the existence of “assistance” automatically invalidates authorship.

The intellectually serious question is not, Did the author receive assistance? The real question becomes, Where does assistance end and authorship begin?

That distinction matters. The offense, if there is one, occurs when technology moves from helping an author express something to creating the thing being expressed. There is an enormous difference between asking artificial intelligence to generate a 90,000-word novel and publishing the result largely unchanged, and asking AI to perform functions comparable to those of an editor: identifying repetitive passages, awkward sentences, continuity errors, grammatical problems or unclear transitions.

One replaces the writer. The other assists the writer.

A reasonable use of AI might include discussing historical questions and then independently verifying the research; brainstorming alternative wording; organizing notes; identifying inconsistencies in a complicated manuscript; testing the clarity of an argument; or asking for criticism of a chapter that the author subsequently rewrites in his or her own voice. None of those activities surrender authorship any more than consulting an editor, librarian, thesaurus, style manual or research assistant.

The publishing industry will eventually have to develop a more sophisticated vocabulary for this distinction because artificial intelligence is not going away. The useful dividing line will not be between authors who use AI and authors who do not. It will be between those who use technology as a tool and those who use it as a substitute for imagination, judgment and creative responsibility.

For the serious independent author, therefore, the strongest position may not be: “I never use artificial intelligence.”

A far more defensible statement is: “The ideas, characters, research judgments, narrative decisions and final language are mine. Artificial intelligence is one of the tools I use during research, editing or revision but I remain responsible for every word published under my name.”

That definition acknowledges both technological reality and artistic responsibility. More importantly, it preserves the thing readers have always had the right to expect when they open a book and see a person’s name on the cover: that the author actually wrote it.

In the end, as always, honesty matters. The harmony between technology and art will come, by and by.