Against AI Slop. For Feelable Thought

“This isn’t X; it’s Y.” Bland, overlapping triads. Em-dashes that elaborate—but not really. The uncanny sense that what appears meaningful lacks actual intention. The stench of AI slop is becoming nearly unavoidable. Here’s an example from a recent viral essay on how amazing AI is: “It wasn’t just executing my instructions. It was making intelligent decisions. It had something that felt, for the first time, like judgment. Like taste. . . . These new AI models aren’t incremental improvements. This is a different thing entirely.” Indeed. A thing that apparently destroys whatever capacity this “author” once had for intelligent judgment.

It’s not a surprise to encounter the rhythms of AI-generated prose in a social media post “written” by a tech CEO. It’s a bit more jarring to find them in a Washington Post op-ed about “guardrails” for AI or a book extolling AI education by Sal Khan (a book that also includes fake quotations attributed to Mother Teresa and Pablo Picasso, among others). It’s particularly jarring to see them in more niche publications I respect—and have myself published in—or in essays submitted to Front Porch Republic. (Why would you submit an essay generated by AI, which makes an argument critical of AI, to a website that doesn’t remunerate authors? No idea, but it happens consistently now.)

I rarely know for sure whether a suspicious-feeling essay is in fact AI generated. My AI radar is flawed, as are the various AI-powered detectors (what strange tail-chasing-dog situations we find ourselves in these days). Many of these essays are likely written through some sort of human-AI collaboration by people who think they are centaurs but who are more likely becoming reverse centaurs. And many may simply be written by humans who have read so much machine-generated prose that they themselves now write and think in machine rhythms. Nic Rowan calls this strange phenomenon the AI “mind meld.” We learn how to think and speak from interacting with our communities, and if we inhabit intellectual ecosystems blasted by industrial-scale token generators, our ability to think well will suffer.

OpenAI’s recent decision to sunset Sora, its AI-powered video producer, is likely motivated by economics rather than any twinge of Sam Altman’s vestigial conscience. And on its own this decision will do little to slow the torrent of manufactured images, videos, and words that have little to no connection with any reality outside the algorithmic functions that whir in data centers across the world. We now dwell among simulacra, and vertigo is endemic. Its most common manifestation is a kind of ennui. The world is a fascinating place, but when we marinate our minds in robotic tokens, boredom ensues.

I need to acknowledge that machine-generated words and images are not the first threat to our ability to seek truth together. Plato complained about the sophists quite some time ago. George Orwell and Thomas Merton warn about the political and moral dangers of doublespeak. Business leaders have long been double-clicking on problems and leveraging best practices to implement impactful solutions. Academese, or what Hannah Coulter calls “the Unknown Tongue,” excels at using many words to say very little. Mass media has eroded local accents and regional vocabulary, imposing a homogenized, bland, and formulaic discourse that is, if we’re honest, little better than its machine imitations, which only further this convergence on a generic mean. As anyone who has been paying attention can attest, the internet was not exactly a thriving verbal ecosystem pre-AI.

We like language that enables us to evade the rigors of thought and its often uncomfortable consequences. In one of his many marvelous asides in Orthodoxy, G.K. Chesterton observes that “most of the machinery of modern language is labour-saving machinery; and it saves mental labour very much more than it ought. Scientific phrases are used like scientific wheels and piston-rods to make swifter and smoother yet the path of the comfortable. Long words go rattling by us like long railway trains. We know they are carrying thousands who are too tired or too indolent to walk and think for themselves.” If this was true in 1908, we shouldn’t be too quick to blame AI for our own moral and intellectual vices.

That said, people immersed in AI-generated text will lack opportunities to develop a sense for sense, a taste for meaning. Good speech or writing makes sense palpable. I have in mind Wendell Berry’s meditation, in “Standing by Words,” on the etymology of “sentence”:

When we reflect that “sentence” means, literally, “a way of thinking” (Latin: sententia) and that it comes from the Latin sentire, to feel, we realize that the concepts of sentence and sentence structure are not merely grammatical or merely academic—not negligible in any sense. A sentence is both the opportunity and the limit of thought—what we have to think with, and what we have to think in. It is, moreover, feelable thought, a thought that impresses its sense not just on our understanding, but on our hearing, our sense of rhythm and proportion. It is a pattern of felt sense.

If we hope to think well, we’ll need to exercise our minds and experience these patterns of felt sense. That’s the only way to develop a taste for truth. Machines—it should go without saying—cannot feel sense, no matter how well they may be able to imitate the verbal behavior of humans who can. Thus, if good writing makes sense palpable, machine text induces a kind of neuropathy, disrupting our ability to intuit truth. Given this danger, we desperately need conversation partners who will shake us out of our mental ruts and provoke—rather than save—intellectual labor.

In an earlier essay, “In Defense of Literacy,” Berry directly addresses the danger that “prepared, public language,” language intended to make us “buy or believe somebody else’s line of goods,” poses to people who have not learned to work carefully with words. As he poses the question, “What is our defense against this sort of language—this language-as-weapon?” His answer is simple yet demanding:

We must speak, and teach our children to speak, a language precise and articulate and lively enough to tell the truth about the world as we know it. And to do this we must know something of the roots and resources of our language; we must know its literature. The only defense against the worst is a knowledge of the best.

There are no shortcuts. We have to consistently exercise our minds by savoring and producing verbal efforts to make sense feelable. The only defense against AI slop is to give more time and effort to carefully crafted language. (For a wonderful book with practical advice on how we might care for language, see Marilyn McEntyre’s Caring for Words in a Culture of Lies.)

One way to do this, of course, is to read books written before November 2022. I certainly want to be formed by Shakespeare, Milton, Thoreau, and Dickinson. I’ll never exhaust the riches of English-language writing (much less good writing in other languages!) pre-generative AI—I could spend the rest of my days savoring Annie Dillard, David James Duncan, G.K. Chesterton, Flannery O’Conner, Robert Farrar Capon, and many, many more. But I also want my word-hoard enriched by people responding with care and wisdom to the events of today. I need poets like Seth Wieck, essayists like Teddy Macker, and story-tellers like Kazuo Ishiguro.

What will it take to sustain the remnants of a contemporary republic of letters on the margins of a public square blasted by machine-speak? I don’t know. And I don’t think anyone does. I am heartened, though, to know of many people who want to think and learn with other humans who are striving to muster a language precise and articulate and lively enough to tell the truth about the world as we know it. In responding to this desire, many readers and writers are looking for ways to signal their commitment to human words. Paul Kingsnorth started an effort called Writers Against AI, and Justin Clark created a set of logos that participating writers and readers can display. (FPR has one of these on our submissions page.) The UK’s Society of Authors has launched a Human Authored program where authors can register their books and get authorization to use their trademarked seal. This program has now expanded to the US Authors Guild as well. Oddly, to my mind, a work can be certified as Human Authored even if authors use AI for “research, brainstorming, or outlining.” Created by Humans allows authors to register their books, attest human authorship, and then license these to companies looking to train AI models. (Though if you value human verbal exchanges, I’m not sure why you’d want to help train machine word generators.) These all rely on “self-certification”; as an author, you give your word that you wrote the words that appear under your name. This approach is prone to abuse.

Other approaches aim to increase reliability. Books by People certifies publishers and vets their editorial processes. Human Creative has taken another route to establish trust; it monitors a writer’s process. This is much more invasive than AI detectors that scan the finished product, and it’s also much more reliable. I’m drafting this essay in Human Creative’s web application, and when I’ve finished revising and am ready to publish it, I’ll receive a shareable certificate (viewable here) that gives readers a relatively high level of confidence that they’re reading something a person composed.

There are real drawbacks to this approach: First, even these guardrails are not insurmountable. I imagine I could generate an essay in some AI platform, print it out, and then type it up in Human Creative (though the application does have some capacity to catch AI-sounding prose). And, of course, if I’ve read so much robot prose that I think and write in machine cadences, typing out such pablum won’t make it worth reading. Second, I’m used to drafting essays in Scrivener, and the various AI deterrents that Human Creative relies on introduce friction into the writing process. Given the widespread distrust generated by machine text, though, process verification will prove essential in many academic, business, and literary contexts. As an editor, I’d welcome essay submissions accompanied by a Certified Human Content seal, particularly from authors I haven’t worked with before. (Human Creative is offering a free trial to anyone who signs up right now; if you give it a try, I’d be interested in hearing what you think.) If the tide of AI slop rises much further, FPR may have to require this seal from new authors. Already, nonexistent citations litter peer-reviewed articles, and some who manage code repositories have had to close pull requests to avoid being inundated with buggy code.

It may not seem fair that the hubris of AI developers inflicts costs on those who want to know whether they are listening to a person or a machine. Polluters always harm not only themselves but many others, and these harms can ripple outward for generations. It’s always easier to damage an ecosystem than to restore it to health.

Unlike the tech-bro CEOs who promise that their products will usher in a utopia where we won’t need to work or think but can merely enjoy the fabulous fruits of machine work and thought, I have no technological panacea to sell. There are no machines that can magically give us the intellectual and moral virtues required to think well and experience truth—or, in Stanley Hauerwas’s resonant phrase, enjoy “the relation of an adequated mind to its object.” These pleasures are the hard-won fruits of caring for our places and for the words we need to know and treat them justly. Those of us who value verbal virtues must commit to preserving spheres where they and their joys are practiced and shared—in our homes, our churches, our schools, and also, if at all possible, on the margins of a digital and print ecosystem being destroyed by machine-generated simulacra of sense.

Image credit: Albrecht Dürer, Of the Just Shaping of Letters