A.I. Policy Non Grata #1

A guide for contributors and readers.

Before ayahuasca ceremonies—in the United States, that is—the shaman will comment on the liquid’s taste. If it is your first time trying the stuff you will, upon hearing the words, begin readying yourself like you haven’t since childhood. Big bad Jim is back except now he’s five-eight—not four-four—and he’s daring you to drink psychedelic juice instead of your own piss. With everyone’s eyes on you, you take a deep breath and then the plunge. But what the newcomer finds, and often tells anyone who will listen afterwards, is that the shot tasted quite good. To that, anyone with any experience will respond, “Just wait for the next time.”

The next time ayahuasca tastes terrible—so terrible that I can’t describe it because my mind has blocked out the memory. It is not uncommon for the advanced psychedelic hitchhiker to grab his bucket and puke as soon as he gets a waft of the smell. For many this is the most unpleasant part of the journey; it’s more difficult to bear than when the ego starts drifting out into the Andromeda three hours later. Though we’re always forgetting it—such is the strength of our subjectivity—taste is mere perception. There’s nothing objective about it. In the case of ayahuasca, five hours of throwing up will change what was a pleasant taste into a horrific one.

Proofreading is like ayahuasca. Unlike editing—the creative act of smashing two minds together to excavate treasures that were only poking through—proofreading is rote, boring, brainless, mundane, and maddening. It is like asking a golfer to put down his clubs and crawl through the dirt to find every brown blade of grass on the course. No one can enjoy searching through the same words 300,000 times, hunting only for typos, missing punctuation, and erroneous formatting. Words once adored start to taste like shit. At the end of his line, the proofreader cannot bring himself to read any more. It would take the effort of climbing a steep rocky hill covered cow dung. There isn’t even the glory or difficulty of conquering a mountain.

In the arts, I cannot think of many better functions for A.I. to serve. The algorithm would not be displacing a creative job, like novel writing, but one that would allow humans to focus on more interesting pursuits. A press that once needed to pay a proofreader could instead train and, hopefully, promote that person to writing or editing. Personally, instead of spending 47,000 hours trying to find missing italics in Coby’s piece, I could have worked on essays and fiction for readers to enjoy. While I wrote computer servers would, in this hypothetical, hum in the background, cleaning up pieces based on a style guide of my direction. Once the algorithm finished, the readers would have a more polished product.

There is not a strong moral case to be made against using A.I. for proofreading. One wouldn’t even be funding the development of LLMs because many platforms offer a free tier without any advertising. Even if magazines and presses were to resist the technology, there are thousands, if not millions, of less principled businesses that would continue to pay for them. Barring restrictive regulation or mass revolution, these companies will keep building whether you use their products or not. The best a widespread boycott, at the individual level, could do is slow their speed down. The cat is, frankly, out of the bag.

Some people may refuse to use A.I. for dogmatic reasons; others may refuse for reasons of identity (e.g. baristas in Portland). It’s easier to write the whole thing off or stick to a previously held opinion than it is to think critically or admit an error in judgement. Though I am against A.I. for many use cases, it is a mistake to not take advantage of the good parts while discarding the bad. Refusing to use Claude to better understand a cancer diagnosis would be asinine. Why wouldn’t you make use of an always-on, private tutor of the sort our favorite geniuses worked with during their aristocratic upbringings? Aren’t readers better off if A.I. helps writers embody new words? Doesn’t everyone benefit if A.I. can help people more deeply understand The Brothers Karamazov?

There is only one good reason not to use an LLM to proofread: Reader preference. Using ChatGPT to find formatting errors assumes that readers prefer a greater volume of perfectly polished products to a lesser volume of imperfect ones. At first glance, this assumption appears correct. But in a world where everything is moving towards computer generation, I am betting that there will be robust demand for publications that are entirely human-created. There is already so much content out there that volume will become less important than craft. Typos and missing punctuation will become valuable stamps of human authenticity.

And how much do small errors really matter, anyway? The Village Voice, one of the most influential newspapers of the twentieth century, was marked by a multitude mistakes twenty years after its circulation commenced. Would one prefer to see the David, with its various crumbles and cracks, or 5,000 robot-built statues without a single marble speck out of place? The computers may be playing yet another trick on us, making us believe that if “perfection” is on offer we should grab it. Yet, in reality, it may turn out that no one ever wanted perfection as much as they wanted to see how close humans could get to it. If Grok can start pounding out 1,900 “perfect” novels an hour, then maybe the mistakes in human-novels become valuable.

Realizing this, if it is indeed true, unscrupulous publications could start having their A.I. manufacture faults. That’s why building and maintaining trust is so important. The reason I’m writing my entire thought process out—on a typewriter, no less—is because I want our readers to know exactly what we at Non Grata believe. Without further adieu, these are our A.I. guidelines:

1. Contributors cannot use computers to generate words. Every word must originate in a human mind. Writers cannot use LLMs to pen or suggest them.

2. Contributors cannot use A.I. to edit. Every piece Non Grata publishes receives in-depth editing from a human. It is a sacred, creative, collaborative process that we will not bequeath to the machines.

3. Non Grata will not use A.I. to proofread. As horrible a job as it is, we would rather publish imperfect, human-crafted pieces than “computer-perfect” essays and stories.

4. Non Grata will not use A.I. to generate images. Illustration and photography are the domains of human beings.

5. Contributors can use software to point out structural patterns. Websites, like Grammarly or the Hemingway app, are fine to use as long as they only reveal existing patterns (e.g. run-on sentences, filler words, etc.). It is up to the writer and the editor to determine what to do about them.

6. Contributors can use A.I. for research. Questions like “What’s the vacancy rate in Austin, TX?” are akin to Google searches; they do not detract from the creative process.

7. Contributors can use software for spelling and grammar. This functionality has been around since the advent of the word processor.

8. Non Grata reserves the right to update these rules based on feedback and learnings. We’re a new publication, and this is a rapidly evolving space. We will publish new guidelines as necessary.

The goal of these guidelines is to ensure the reader knows exactly how Non Grata and its contributors use A.I., if we decide to use it at all. Please leave us any feedback you have; we are always open to new approaches and compelling arguments.

---