The Death of the English Language
Tal vez sea hora de cambiar a Español
Sam Kriss and I are aligned. We both published complementary articles this week about AI writing; we both hate it for the same reason: why are we doing this to language? He covered the well-known telltales, me the less-known ones, so they’re best read in tandem. He focused on AI’s stylistic markers (”why does it write like… that?”), whereas I chose a “show, don’t tell” approach to make readers viscerally understand that no telltale sign is fully reliable.
Our complementarity ends there, though: I published on my humble blog, he in the New York Times. Humbly, I will say that although mine digs deeper—your loss, NYT—his is gentler. So let me take this chance to apologize for having waved off, in my first paragraph, most of Kriss’s exploration as “surface telltale cues of AI writing that deserve no further inspection,” or something to that effect. Kriss’s treatment proved me wrong (there are worthwhile things to say about the over-presence of em dashes and the word “delve”). I will also take the chance to anchor this essay on the last paragraph of his, below verbatim:
Maybe soon, the gap will close. A.I.s have spent the last few years watching and imitating us, scraping the planet for data to digest and disgorge, but humans are mimics as well. A recent study from the Max Planck Institute for Human Development analyzed more than 360,000 YouTube videos consisting of extemporaneous talks by flesh-and-blood academics and found that A.I. language is increasingly coming out of human mouths. The more we’re exposed to A.I., the more we unconsciously pick up its tics, and it spreads from there. Some of the British parliamentarians who started their speeches with the phrase “I rise to speak” probably hadn’t used A.I. at all. They had just noticed that everyone around them was saying it and decided that maybe they ought to do the same. Perhaps that day will come for us, too. Soon, without really knowing why, you will find yourself talking about the smell of fury and the texture of embarrassment. You, too, will be saying “tapestry.” You, too, will be saying “delve.”
I believe he’s basically correct: humans are indeed mimics, so the more bots sound like us, the more we’ll sound like bots; AI writing is a two-way street with dangers at either end. But something else caught my attention, that, as an Englishman, he couldn’t have possibly realized: those YouTubers and those British parliamentarians speak English. You might find this unremarkable, but as a native Spanish speaker who’s only an English speaker by trade, the end picture reveals itself to me: English is the language of the internet, the internet provides data to AI, AI increasingly generates everything we read, what we read influences how we speak and write. Thus, this “human collapse”—akin to “model collapse”—that Kriss warns about will happen exclusively to English!
I rejoice. I like English as a language, but I rejoice because it’s funny that having won the competition to become the language of the world and of the web became its death sentence. As AI models and humans converge on the same vocabulary, sentence structures, and stylistic markers, the immensity and richness of English as a language—which I can barely fathom from my limited, non-native perspective but Kriss’s erudition and lexicon reveal to me time and again—will vanish. If we enter this depressing timeline, eventually there will be no one left to praise Shakespeare’s genius, for no one will understand him; no one will agree with Borges that English is “a far finer language than Spanish” by virtue of the versatility and physicality of phrasal verbs, or due to the freedom that switching between dark Saxon and light Latin registers confers.
Consider this: Latin didn’t die because speakers died but because it shattered into a dozen mutating dialects that eventually became languages of their own: French, Italian, Spanish. Death by cultural and social diversification, if you will, leaving Latin forever relegated to scriptoriums in abbeys, the library of the studium, and marginalia of dusty books (and, apparently, the essays of pretentious writers). English faces the opposite fate: death by consolidation; death by convergence. English will not cease to be spoken, but it also won’t shatter into a diverse progeny; it will simply cease to be alive: it will die in the lazy corners of the Library of Babel, forever the higher rows beyond its reach. The last generation of English scholars who dare venture into the dangerous territory of uncommon words might be alive today and dwindling, I imagine, given the statistics of readership—both the number and the quality of books—among the youth.
But as I watch this slow suffocation of English, I look at my own tongue, Spanish, and I see a flower blooming through the concrete. Spanish is safe. In fact, Spanish is about to inherit the earth, precisely because it is currently losing the technological war—vuestro triunfo hoy es nuestra hegemonía mañana! The Anglosphere views AI as its faithful servant, a force multiplier (another one) for its culture. They believe that because the models are trained on English, English wins (”write for the AIs,” they say, hopeful to be part of an AI-enhanced Anglophone future). But they mistake a virus for a vitamin: because the AI lives in English, it metabolizes English; it chews on the language’s corpus and spits out a smooth, featureless paste. Your kids are sure enjoying that paste together with the fish and chips and the unforgettable and unforgivable tinned beans.
Why does Spanish remain resistant to this digestion, though? Isn’t it present, just like English, in ChatGPT’s dataset? Yes, but to the AI—understood as this hyperobject manifested as different instances of chatbots that will, at the first chance, merge into a machine god—Spanish is a secondary language. 1) The training data is smaller (around ~50% is English in the best cases, up to 80%+; in the broader internet, Spanish amounts to a negligible 6% which is the exact proportion of native speakers). 2) The nuance is harder to capture (we are more tacit—tactile also—than explicit). And 3), the cultural context is too chaotic (or rather, missing from the broader web, as every space that contains an English component is naturally dyed with English in its totality, resulting in English taking up half of the internet, yet being spoken natively by 5.1% of people).
I’m afraid the world is yet to realize that the cultural damage of AI, which we hear about all the time—“don’t let it think for you!”—is language-specific: I’m not getting dumber in my Spanish household by speaking English with ChatGPT or Gemini because my valuable thoughts, which I would never commodify by outsourcing them to AI, happen somewhere else in my lexicon-devoted cortex areas. In short, my vocabulary is as ample as it has always been, and my mind as sharp.
You have to add to this phenomenon the fact that English-speaking countries are handicapped by their historical success: why learn a different language when everyone else already bothers to learn yours? Your mind is inevitably confined, more so if we take into account another fact that, admittedly, requires some belief: language builds reality. “You can only perceive what you can name” is as true as “you can only name what you can perceive” (e.g., categorical perception or perceptual narrowing). One language—even if one as unfathomably rich as English, as Borges said, a fellow Spanish speaker with a similar deference to and fondness for English—is strictly less powerful than two. The negative impact of AI use is the last ingredient in the recipe for a dead language.
This might not happen this year or this decade, but the tendency is as clear as it is clear that I could never publish an essay like this one in the New York Times; perhaps in Los Tiempos Madrileños, once it becomes the biggest newspaper in the world, sometime in the year 2098. For now, however, I bow to the supremacy of English in the digital world and will continue to do so from my humble blog.
I don’t foresee the death of English from AI overuse. Its roots are too deep and varied over the many centuries of its use. What we -will- see is a particular historical flavor of English in use for a certain length of time. After reading a couple pages of English text, I can usually make a guess about when it was written—not only from the spelling and vocabulary used, but from the tone and color of its narrative. The way the writer chooses to refer to other people, and the amount of courtly flourish in the syntax, for instance—both are signals of social standards that were important in the writer’s time. Similarly, historians are going to read texts from this time and say, ah yes, the great flattening of early-mid twenty-first century English. It’s happening, but yet I don’t despair. Perhaps it’s because I’m so old—I’m convinced by a long life of observation that the human mind is too vital, too raw, to stay strapped into a straitjacket of managerial-flavored AI impoverishment for very long. For every second-tier authors seduced by the ease of writing their way to fame using LLMs to write their books—there will also be writers like Sam Kriss who bend English, stretch it like Play-Doh, extend it into absurd wonder and joy, and provide true nourishment to the reader. We’re going through a phase, like we always do. But this will end, as it always does, when the next “end of civilization” ripens and breaks into our awareness. We’ll survive.
Brilliant insight! Perhaps, Taushiro (indigenous to Peru according to Google) and non-Mandarian dialects of Chinese (among other languages) will become the loci, eventually, of creative endeavors and subversion of all types. Hmmmm. There is a short story or novel here . . .