Judging an A.I. Book by Its Cover

Emerging Civil War is pleased to welcome guest Nick Sacco. Nick is a St. Louis-based historian and owner of History Beyond Black and White.

For those of us who are enthusiastic readers, the idea of not judging a book by its cover is a mantra. The philosophy is simple. Our personal biases, first impressions, and expectations can lead us to rush to judgement too quickly. We run the risk of making premature conclusions about an author’s writings without giving a fair review of his or her arguments, often based on a book cover the author did not design.

While I have always tried to not judge a book by its cover, the infiltration of AI into the book publishing industry is testing my patience and has me questioning this philosophy.

As a student of Ulysses S. Grant, I like to keep an eye on the latest scholarship around his life. I periodically check university and popular presses to see if any relevant titles have been released. While there has been no major Grant biography published since Ron Chernow’s book nearly ten years ago, there remains a nice collection of scholarship by the likes of Charles Calhoun, Chris Mackowski and Frank Scaturro, John Reeves, Ryan Semmes, and Louis Picone that keeps the field moving forward. However, the number one source of new “scholarship” on General Grant in recent years is AI slop.

All you have to do is go to Amazon and search “Ulysses S. Grant” to see how bad it’s gotten. Filter your search by publication date and you’ll be inundated with recent AI book covers. A Paws in Time; The Siege of Petersburg; Ulysses S. Grant: The Drunk They Wrote Off Who Won the War. All of these books and more feature AI book covers.

One book particularly irritated me. Humble Hero: The Story of Ulysses S. Grant for Kids by Brian Thomas features a cartoonish Grant that is very obviously AI. Grant sports a fishing hat and a modern belt buckle, no doubt infuriating uniform purists nationwide. After looking at the cover, I decided to dig deeper. Who is Brian Thomas, and what are his credentials to be considered an expert on General Grant?I clicked on Brian Thomas’s Amazon page and immediately became suspicious. There was nothing there. No picture, no bio, no information whatsoever. And yet, Thomas has more than 70 publications to his name that are available for purchase. Even more amazing is that all of the titles I clicked on were published within the past 18 months. All had AI slop covers. All were about 150 pages in length. That comes out to about 10,500 pages of text, or roughly 584 pages of text per month written and published by Brian Thomas over the past 18 months. One would think that someone producing at such a prolific rate would have a significant digital footprint.

I then turned to Google’s AI search engine, Gemini, and asked for more information on Brian Thomas. Unsurprisingly, Gemini got confused and offered contradictory information. It initially pointed me to a Dr. Brian Thomas who had no publications on Grant and was not a historian. Then it pointed me to a motivational speaker who appears to have passed away several years ago before AI became mainstream. I told Gemini it had the wrong person and it then admitted that it could find no information on Brian Thomas. After digging around on my own, I found that all of Thomas’s books were published via Lulu, a self-publishing platform that requires no author verification of any sort. Although I did not buy Humble Hero, I am very confident in asserting that “Brian Thomas” is an AI author. Publishing more than 70 books in about 18 months’ time indicates to me that this is not a real person.

AI slop books are spreading online because various entities, from self-publishing platforms like Lulu to sellers like Amazon and Barnes & Noble allow these books to publish and are actively profiting from them. A recent feature by writer Kashmir Hill explored one AI “author” who is even more prolific than “Brian Thomas.” The “author” in Hill’s essay has written 445 books through ChatGPT and uploads ten books per week on Amazon (the maximum allowed). He makes $7 per book and made nearly $6,000 selling 821 books during the holiday season last year.

These developments should be worrisome for historians and readers of history books. Generative AI might be able to lay out some very foundational insights and parameters for a given historical topic. However, AI cannot produce original arguments or formulate interpretations based on research of primary and secondary sources. Equally important is that when I purchase a book, I expect the words, thinking, interpretations, and creativity of the work to be that of the author. AI books rely on stolen content, often from scholars who have spent years researching their topic. AI books mash up previous works, often fabricate sources (if they cite sources at all), and are really a form of mass plagiarism. The intense media scrutiny towards various academics accused of plagiarism in recent years has overshadowed the mass plagiarism of texts by AI tech companies.

I acknowledge that some people writing genuine works of original historical scholarship may use an AI generated cover for their book. My response is threefold. One is that these slop covers take away work from professional graphic designers and artists. As we see with Humble Hero, these covers often get fine details wrong and can be historically inaccurate. More importantly, they look sloppy, soulless, and unappealing. Second is that, at least within the context of the Civil War, there are literally millions of historical images, often public domain, that can be used for a book cover. For Ulysses S. Grant specifically, there are more than 300 confirmed photos of the man, again most being public domain. The Civil War is arguably the most photographed event of the 19th century. That source base is rich and doesn’t need an AI generated image to “bring history to life.” Finally, an AI book cover erodes my trust in the author’s work and their legitimacy as a scholar.

AI book covers create a slippery slope that I cannot escape. For me, if an author uses an AI book cover, then they may have used AI to index the book. They may have used AI to edit the book. They may have used AI for the author’s headshot photograph. They may have used AI to generate maps, graphs, and other visualizations. They may have used AI to structure the book’s content, cite sources, and make interpretive arguments. I don’t know any of these details for certain, and that’s the problem.

Moving forward, the big question for history readers is if they will care about whether the books they read are original works or AI generated. If we do care and the distinction matters, then we should work together to ensure we do not purchase AI books. Check information on authors and their publishers before purchasing. Do they have an online presence? What are their credentials, and what makes them qualified to write on this subject? But if the status quo holds, AI “authors” will continue flooding the market with crap. Civil War historiography will grind to a screeching halt as we descend into a constant AI generated regurgitation of scholarship published years ago without any new arguments to speak of.

A body builder does not create a muscular physique in one day or one week. An artist does not fully realize their artistic vision in one day or one week. A historian cannot produce original scholarship in one day or one week, much less ten books a week. We cannot AI our way to good history or, more specifically, we cannot AI our way to new understandings of the past that advance historiography. We must work to preserve a thriving history, artistic, and literary culture in an environment shaped by the creative talents of humans, not bots. If the final product is not worth the effort to create original scholarship, literature, or art, then it’s not worth my effort to invest time in that product.

Part of a series.