In March 2024, I published an essay titled Humane Content in World of Language Models. While I have since taken a stronger stance about describing meaningful writing and media as content, it fits here. This essay was prompted by an incisive essay by Maggie Appleton titled The Expanding Dark Forest of Generative AI. I agreed with much of her essay, but disagreed with framing the challenge that humans writing on the internet face in an internet flooded with generative AI as a sort of “reverse Turing test” where humans should think about proving that they are not bots.
My first argument against the reverse Turing test framing is that neither I nor anyone else writing on the internet can control what other people think. I know how I produce my articles, essays, and photographs and images, all without generative AI. I have promoted my projects as being humane in the sense that I write as a human being for other human beings who may happen to come across my projects. I subsequently went so far as to apply these principles to coding when I created a visual novel by “reading the docs” instead of relying on AI, despite the fact that I am not a developer by training or trade.
I run my projects in good faith for human readers who want to read them in good faith. John Ruskin described good-faith reading well in his Sesame essay:
[B]e sure that you go to the author and get at HIS meaning, not to find yours. Judge it afterwards if you think yourself qualified to do so; but ascertain it first.
John Ruskin (Sesame)
I am sure someone, somewhere, could read one of the human-generated essays I publish for no compensation and shout “J’accuse, AI!” Do I have anything to prove to this person? I think not. I take the time to craft my articles and publish them, with all my human thoughts and typos, for free. Someone who decides it is AI most likely opened the link with his or her “meaning” firmly in mind. That hypothetical person is not my audience.
The second reason I rejected the reverse Turing test idea is because there are many problems with internet writing separate and apart from LLM-generated text. To quote myself:
Human beings writing original content or producing media should not think of needing to pass a reverse turing test. They should instead think about how to produce distinctly humane writing (or media) on the internet. By humane, I mean by humans, for humans. You, a human being, are creating something from your own perspective for other human beings. This concept is more useful than focusing solely on language models because it is entirely possible for human beings to create inhumane content for consumption by search engines (e.g., articles full of SEO keywords for the primary purpose of ranking on Google) or to prompt readers to action instead of to inform (e.g., review articles that exist primarily to convince readers to buy products through affiliate links instead of providing informed first-hand opinions of the quality of the products).
Nicholas A. Ferrell
Or to quote Jim Nielsen:
You don’t need AI to produce slop because slop isn’t made by AI. It’s made by humans — AI is just the popular tool of choice for making it right now.
Jim Nielsen
Well-put, as I opined on Emu Café Social.
While I did not like one aspect Ms. Appleton’s framing, I did appreciate her recommendations following from that framing. People who want to write good articles and produce good media on the internet should focus on just that. Good articles and media are not defined by how they differ from AI slop, SEO spam, or affiliate marketing. An essay which flows from clear thinking is good in and of itself, without making reference to the many brain numbing things on the internet. The same principle goes for good videos, discussions, photographs, and the like. The arguments I presented in The Positive Case for Good Tech could easily be re-fashioned into The Positive Case for Good Online Writing and Media.
I agree with my arguments I presented in Humane Content in World of Language Models, but I was prompted to write this new essay because there was an angle that I neglected back in 2024. I came across two articles which inspired my new angle: First, a Daily Dot report by Katka Lapelosa titled ‘I use like 30 em dashes a day’: Someone said the em dash is an AI tell—and writers aren’t having it, and second, an opinion piece for The Ringer by Brian Phillips titled Stop AI-Shaming Our Precious, Kindly Em Dashes—Please.
Both articles deal broadly with the “online” perception that the em dash is a mark indicator of LLM-generated text. This belief—which is evidently held by some people on the internet—had come to my attention before reading this articles. I will submit for the record that I take no position on whether large language models spit out an abnormally high number of em dashes. I have not spent enough time conversing with the chatbots or “consuming” chatbot-generated text to study the em dash usage or lack thereof. I will note that from the LLM-generated text that I have seen and know for certain to have been LLM-generated, I have not noticed high em dash usage. Moreover, I have seen a decent amount of internet writing that I suspect to have been LLM-generated, but never on account of the presence of em dashes.
I will not discuss either of the em dash focused articles in greater detail because the question of whether LLMs favor em dashes in their text generation is beside the point here—let us instead proceed assuming arguendo that there is a broad perception, regardless of whether it is well-founded, that em dashes come with a whiff of LLM. Assuming this is the case, it would seem to follow that a human being who publishes articles with em dashes may be suspected by other human beings of publishing text that is at least partially generated by a chatbot. Assuming this is the case, should a writer who would otherwise use an em dash—either in the exercise of stylistic discretion or because he or she believes that it is the best punctuation to use in the moment—not use the em dash to avoid AI allegations?
Those of you who have read this far should be unsurprised that I would not recommend that writers avoid using em dashes to avoid unfounded allegations of AI-usage. I incorporate by reference all of the arguments I presented in my 2024 essay relating to writing in good faith for good faith readers and focusing on producing the best written product for those readers. Now I present a different way to think about the issue—inspired by the em dash articles: If my goal is to write my human thoughts for human readers, should I change how I express my entirely human thoughts based on how AI “writes”?
To be sure, a human being making punctuation choices to hide from AI allegations is not relying on AI to think and write on his or her behalf. I will venture that this human being’s efforts are most likely for naught—the internet people who suspect the em dash this week may suspect the hapless human’s replacement punctuation next week (not least because the LLMs may have trained on it mid-week)—but the human is still producing entirely human-conceived and “generated” writing. But something feels off. This human author is changing how he or she writes based on the perceptions of internet people—most likely percolating on the bot playgrounds of algorithmic social media. This author is making choices about how to present his or her thoughts that are necessarily tied to his or her perception and the perceptions of the “crowd” of the large language models. The author’s text is not AI-generated, but it is AI-influenced in that the author is making choices that he or she would not make independent of AI and “perceptions” thereof.
Human writers who care about publishing their own thoughts in their own words should not allow their word and punctuation choices to be modified, much less watered down, by what the unmoored blob thinks AI is producing (unless, of course, they are actually writing about AI). People have written many words for more than 2,000 years without adjusting their prose to accommodate internet opinions about AI-generated text, and nothing about AI makes it impossible to continue this great tradition. It is better to have the courage to write well and in one’s own voice than to change register in the hope of forestalling AI allegations from people who most likely have no interest in reading and engaging with the author’s work in good faith. The way to convince people that human writing is worth reading is to write well and in one’s own voice—a principle which applies to other forms of art. Or to re-phrase it—”the best way out is always through.” If the onslaught of AI and LLMs inspires one to refine his or her “human” craft, think clearly, and write better, that is well and good—so long as the goal is to write well and not with reference to tools he or she does not use.
