A.I. and the Mysterious Mountain at Spotsylvania Court House

An image came up on my Apple News feed this week that immediately caught my eye and simultaneously surprised me.

The image advertised an article in History of War titled “The Bloody Angle: Inside the Civil War’s most brutal 20-hour hand-to-hand combat.” (In the interest of full disclosure, the article was part of the Apple News+ feature, which I don’t subscribe to, so I haven’t read the actual article.) Here’s the image someone used as a thumbnail to help advertise the article:

The bottom of the image features combat depicted in a widely circulated 1887 painting by Thure de Thulstrup, The Battle of Spotsylvania Court House. A lithograph of the image hangs in my home, and for years, a version of the image appeared on the brochure for Fredericksburg and Spotsylvania National Military Park. It’s one of the few paintings in existence that depicts the battle of Spotsylvania Court House.

What struck me, though, was the juxtaposition of part of that painting against a backdrop that seems to suggest a tall mountain in the background.

There are no mountains on the Spotsylvania battlefield.

One part of the battlefield is known as Laurel Hill, but that’s little more than a gently sloping field with lots of tumultuous hills and swales around it. In fact, that’s the best way to describe nearly the entire Spotsylvania battlefield: tumultuous hills and swales.

There was an actual Laurel Hill about a mile or so to the east of the battlefield, but the part of the field we today call Laurel Hill was actually part of Sarah Spindle’s farm and is more appropriately called Spindle Field. It sits to the east of the Bloody Angle, depicted in the painting.

To the west of the Bloody Angle stands Myers Hill, shrouded in woods today but mostly open at the time—but in the completely opposite direction of the perspective shown in the painting.

In neither case are any of the hills large enough to loom menacingly in the background.

I might seem nit-picky here but for two reasons. First, when I give tours of the battlefield, I concentrate a lot on topography. It’s a key part of nearly any battlefield, but it really dictates the flow of events at Spotsy. Tying the battle narrative to the ground makes the whole thing story into place for a visitor.

Second—and, really, the much larger point—is that so much AI-generated history material is showing up online these days that the line between real and fabricated content is getting blurrier and blurrier. This image might not even be A.I., to be honest, but the fact that I can’t even tell is part of the problem.

My friend Nick Sacco, a historian in St. Louis, made an interesting video about this very topic earlier this week. Nick did a little digging about an “author” who’s written more than 70 history books, including one on Ulysses S. Grant. Nick’s assessment: “A lot of new history ‘scholarship’ is AI slop.”

If you follow Nick’s work at “History Beyond Black and White” (and if you don’t, you should—fascinating stuff!), you know he uses digital technology to colorize historical black-and-white photos—a process he does by hand using a digital pen and pad connected to his laptop. “I do everything from researching the colors (to the best of my ability), to editing the photo, to applying color,” he told me. He’s been doing it for five years, but lately, “AI colorizations have flooded the internet and really degraded what to me is an interesting art form.”

While he sees a potential use for AI as a tool for processing large data sets, he’s “generally opposed to AI usage in the humanities, liberal arts, and creative arts more broadly.” He used the example of the AI “author” in his video to make his point.

My friend and longtime collaborator Kris White offered me another example this week. He asked AI to generate a map of the “theaters of the Civil War.” Here’s what AI gave him:

For anyone curious about the prompt Kris used, here’s what he input:

“Create a slide for a presentation. It’s a map of the United States in 1861. It should have an old-timey feel to it. The map is showing the theaters of the American Civil War. Different colors should be used for each theater. Theater 1 is the eastern theater, which is from the North Carolina/Virginia boarder to Harrisburg, Pennsylvania to the Atlantic Ocean to the Appalachian mountains. Theater 2 is the western theater, which runs from the Mississippi River to the Atlantic Ocean and Gulf of Mexico and south of the Ohio River. Theater 3 is the the United states west of the Mississippi River.”

Notice that A.I. didn’t even get basic things like the names of the states correct.

Kris tried the same exercise with several different AI generators and came up with different results from each one. Kris saw the red flags immediately, but a lazy person or, more likely, someone who doesn’t know any better, might be tempted to say “good enough” and use the first thing an AI generator spits out. It’s little wonder that so much “AI slop” is showing up online.

Nick isn’t the first person to use that phrase (although I laud him for using it!). I hear an increasing chorus of credible historians expressing the same frustration: AI-generated content is becoming more and more common, which is problematic because so much of the audience is not discerning or discriminating.

This is, of course, a problem that extends beyond the bounds of just history. Social media feeds are full of AI–generated content, and unfortunately, much of it has been designed to mislead or even misinform. Algorithm-curated click-bait is everywhere, specifically designed to generate instant emotional responses that amuse us or soothe us or make us afraid or angry. Those emotional appeals are specifically designed to end-run around the critical thinking parts of our brain. We react emotionally before we even have time to think—and of course, once we’ve decided how we feel about something, it becomes a lot harder to change our minds about what we think about it. The use of AI to create even more-sophisticated content makes that problem exponentially worse.

As someone who teaches in a school of communication, I spend a lot of my professional time teaching information literacy. I often feel underwater from the get-go. There’s so much bad content out there because it’s easy to create and distribute and, unfortunately, so many “creators” are perfectly content to be lazy and/or unethical and use AI to generate material.

The key is to be on the lookout for it, to be aware that it’s trying to provoke and titillate. The algorithms are intended to give you content you want, that reinforces what you believe, and so it’s easy to get trapped in a feedback loop.

And this is the most dangerous (and frustrating) part: so many people are convinced they’re immune to that sort of manipulation. “I never fall for that stuff,” they say. They know better. They are discerning enough. Maybe that’s true, but in my experience, a fish doesn’t know it’s submerged in water, either.

I’m not opposed to AI, by the way. In the communications field, it’s a fact of life. I can’t tell my students “don’t use it,” because when they get into the professional world, they’ll have to use it. The key is to teach them how to use it effectively and responsibly.

That said, I’m still skeptical of the technology (for a whole host of reasons beyond the scope of this piece). When AI-generate content is presented as factual or actual—or once people start believing it to be so—serious issues arise. People can be too easily fooled by AI-generated content, particularly because so many of them confidently believe they can’t or won’t be fooled by it. We become complicit in the problem.

Have I made too much of a virtual mountain out of an AI molehill? Maybe. But until mountains appear on the Spotsylvania battlefield, I’m going to keep a skeptical eye on the digital landscape—and I encourage you to, as well.

Part of a series.