AI is Useful, but it Will Not Be Brilliant
AI is here to stay. Those who don’t explore its use and capabilities may soon find themselves left at the dock.
At our company, we treat AI like any other engineering tool — no different than finite element analysis or computerized performance prediction. When used well, it’s extremely useful. When used poorly, it’s useless. At that level, we don’t worry about “intelligence.” It’s more like a fluffy Wikipedia, which, by the way, remains a surprisingly solid source of hard technical information.
Early on, I asked AI to list the causes of slip and fall incidents on stairs. (I purposely avoided the marine term “ladders” to keep it simple.) It produced a neat list which was not particularly innovative, with a few items slightly off, but with one entry that made me pause and think, “Hmm, I ought to remember that one.” In that sense, AI works as a kind of global bookkeeper, keeping score on the world’s collective knowledge.
Recently, I wrote an article for the OPA90 Forum newsletter. It turned out to be too long, and instead of bothering me to shorten it, the editor asked ChatGPT to cut it from 1,000 to 500 words. He touched it up and sent it back. To my surprise, the message was still mostly intact, but it certainly wasn’t in my voice.
So, I decided to ask ChatGPT to rewrite the edited version in the style of Rik van Hemmen. Because I’ve littered the internet with enough of my writing, it recognized my style and came back with something that was 99% accurate. It looked and felt like me, which was both impressive and a little unsettling.
Over the past few years, I’ve also been writing historical fiction about the settlement of eastern Monmouth County, New Jersey. The stories are posted slowly on www.vanhemmen.us. They’re written in the James Michener style, historically grounded but populated with fictional characters. Fiction allows an author to explore the human dynamics behind the history, the context that pure historical analysis often misses simply because no one wrote it down at the time.
The early chapters explore the encounter between European settlers and the Lenape people. The historical record, written by Europeans, naturally carries a European bias. Native Americans often appear “unsophisticated,” but when you write historical fiction, certain truths become self-evident. The Lenape were not backward; they were highly sophisticated, but were forced to play with a bad hand. Ignoring that is as misguided as claiming that Jews were “unsophisticated” during the Holocaust if the Nazis had controlled the narrative.
In my stories, I try to show Lenape sophistication through their ability to resolve complex issues in trade negotiations, language, mathematics, and their appreciation for new technologies. One chapter, for instance, includes a Lenape man learning to sail a European-style boat.
After my success with ChatGPT on the OPA90 article, I decided to run one of these chapters through it for a light readability edit. It was an unmitigated disaster. ChatGPT filled the text with florid, emotional language and stripped out the long trade discussions. My careful account of a Lenape learning to sail was reduced to: “After some instruction, he quickly learned to sail.”
In other words, it deleted the entire purpose of the story.
I tried to coax it back — locking sections, issuing strict instructions — but the story kept collapsing. This was my first real encounter with AI’s limits. ChatGPT recognized the text as “historical fiction” and pulled sentence patterns from thousands of similar works online. It learned that most historical fiction avoids lengthy technical or trade discussions — so it deleted them, chasing the average reader’s attention span.
That may be fine for producing more readable stories, but it completely destroyed my intent to shift the way we think about Native American history.
If I had accepted ChatGPT’s edits, I would have joined the mass of generic historical fiction out there. Instead, I chose to stay the course, trusting my judgment, and maybe leaving behind enough of my own writing that, someday, AI will learn not to delete the trade and sailing discussions.
In the end, AI is very good at producing average noise. It’s far weaker at developing novel ideas without human guidance. It reflects the old joke about democracy: the flawed assumption that more than 50% of the time, more than 50% of people are right.
Innovation, truth, and insight still demand persistent human effort — and I’ll bet they always will.
For every column I write, Maritime Reporter & Engineering News makes a small contribution to an organization of my choice. I nominate the Nanticoke Lenni-Lenape Tribal Nation, which works tirelessly to preserve its culture.
