My introduction to film theory was the Dogme 95 manifesto, which I read in high school. A funny place to begin, I know, since the Danish directors Lars von Trier and Thomas Vinterberg were writing in reaction to thirty-five years of accreted custom and habit in European filmmaking. But so it goes. Trier and Vinterberg argued that the artistic sensibilities expressed by the heroes of the New Wave—which they described via a spoof of François Truffaut’s journalistic style—were no longer compelling. By 1995, they wrote, the “anti-bourgeois cinema” of the 1960s was commonplace; with the arrival of digital filmmaking, “for the first time, anyone can make movies.”
This “ultimate democratization of the cinema” was a catastrophe, and extreme measures had to be taken to correct it. “The more accessible the medium becomes, the more important the avant-garde,” the two directors wrote. But rather than initiate another New Wave, they proposed a technical solution. To preserve the integrity of filmmaking Trier and Vinterberg formulated “an indisputable set of rules” designed to set a high barrier against amateurs. They then invited anyone who dared to join them in swearing the “Vow of Chastity.”
The rules are extremely strict. All shooting must be done on location, and all props must be local to the shoot. The camera must be handheld; all sounds, including any music, must be diegetic; no special lighting. No optical filters, no black-and-white film, no period settings (also no murder mysteries or other genre pictures). Every film must be shot in Academy 35 mm. Finally, the director, who is not permitted to credit himself for the work, must declare: “My supreme goal is to force the truth out of my characters and settings. I swear to do so by all the means available and at the cost of any good taste and any aesthetic considerations.”
Who could possibly keep all those rules? Who would want to? Not even the writers of the manifesto tried. The first film Trier made after co-writing Dogme 95 was Breaking the Waves, a strange, weirdly affecting movie about adultery and misplaced religious fervor. It violates the Vow of Chastity: the film was not shot on location, it has a soundtrack, and Trier even used C.G.I. in some scenes. No one minded. Breaking the Waves was praised for its Dogme 95 “aesthetic,” and it won the Grand Prix at Cannes. It is part of the reason why anyone outside the insular world of auteur filmmaking even knows about Dogme 95. Consequently, in the years since, the manifesto has alternately been praised and derided, tributed and parodied, but never forgotten. Even if its own creators could not abide by its strictures, the rules remain worth falling short of. There is something tragically noble about swearing vows that cannot be kept.
I have gone on at such length about Dogme 95 because I see something similar in the attitudes of some writers on the subject of artificial intelligence. I need hardly rehearse the facts: the same conditions apply. My field is also in the middle of a “technological storm.” Everyone knows how easy it is to use an L.L.M. to generate text that to the untrained eye could pass as a novel or a short story or a magazine feature. Even trained eyes have been fooled. For a couple years now, awards, prizes, and other honors have been bestowed on work later revealed to have been machine-generated. Some writers have straightforwardly incorporated A.I. into their work, reasoning that, like the stylus or the word processor, the L.L.M. is just another technology in the writer’s toolbox. But most know that there is a stigma attached to using A.I. in writing and try to hide its contributions, leaving editors like me to catch and censure them.
But the group in this storm that interests me most is those who have sworn a new version of the Vow of Chastity. These are the writers who have publicly stated that they will never use A.I. in their work. They are emphatic in their refusal and they often savagely attack those other writers who do use A.I. In practical terms, they seek to ban A.I. in their workplaces and to discourage its use in creative endeavors. If you work or have contact with anyone in certain corners of academia or law or the world of literary journalism, you probably know some of these people. If you are reading this, you may well be one yourself. Well, I am here to say that your efforts are doomed, beautifully doomed, just as Dogme 95 was doomed.
Allow me to argue by way of anecdote. Earlier this summer a fairly well known novelist sent this magazine an essay. Almost as soon as we opened the document, it became clear that he had made Claude or ChatGPT do the work for him. We didn’t even bother with running it through Pangram, ZeroGPT, or one of the dodgier A.I. detectors. It was so obviously bot work. Later in the summer, we received another essay. This one we liked. But just to see what would happen, I ran it through a detector. One hundred percent A.I. That seemed wrong: first, I knew he wrote it himself, and second, I had edited it, so my work was intermixed with his. So I ran it again. Eighty percent A.I. That was even stranger. I ran it a third time. One hundred percent human-written. Bizarre. I kept running the work on different A.I. detectors and received wildly varied results. In the end, I distrusted them all.
I learned from these experiences that you cannot enlist a machine to do a human being’s work, even if that work is the unmasking of another machine. And I see now that I should not have even bothered with the A.I. detectors. In both cases, I was able to tell A.I. from human writing with my eyes alone. The first submission was written in non-idiomatic English with impossible metaphors and contentless phrases that gave the work an unsettling surface-level smoothness. No human mind could have produced it. The second submission was rife with spelling errors, mistaken capitalizations, and syntactical infelicities that no machine would commit. But it also was entertaining and made a coherent argument. I could tell human from machine because my eyes have been trained to distinguish good writing from bad on a purely technical level. On top of that, I have developed in myself some measure of taste that allows me to identify the merits of a given work in a manner particular to me alone.
Now, wait, wait, you might say: isn’t this a case for the neo-Dogme 95 movement? I wish! But what I do when deciding whether or not a piece of writing is A.I.-written is uncommon, especially in my field. Best practice these days is for a news organization or publishing house or literary journal to have a subscription to Pangram which is meant to ferret out A.I. compositions. If someone suspects that a text is the product of A.I., he runs it through a detector and lets the machine deliver its verdict. And that opinion is unquestionable. Of course, this system just extends the dominance of A.I. even deeper into the field. The chaste may feel that Pangram protects their writing from intermingling with that of the defiled, but it is ultimately an A.I. making decisions about A.I. It precludes all possibility of human taste and prejudice. It does not protect human writing. It only cedes more ground to the machine.
The most delicate question in all this is accessibility. The Internet has been around for a long time now—much longer than I have been alive—and it has become commonplace to assume that most information is accessible within a few clicks. (Never mind that it isn’t, actually. The amount of useful information freely available on the Internet is rather paltry: the overwhelming majority of books written in languages other than English, or in any language before 1750 and after 1930, especially scholarly or critical editions; essays and articles published in printed newspapers, magazines, and academic journals; and quality reference works such as the Oxford English Dictionary, are either locked behind paywalls which require expensive institutional subscriptions or else gathering dust on library shelves. What was supposed to be an enormous opening of the range of material available to writers has actually shrunk our horizons, enabling a surface-level knowledge of anything, while discouraging deeper understanding—but that is another essay.)
For years before ChatGPT came online, many writers had gone to the Internet for sourcing, ideas, inspiration. If you think you are pure, you almost certainly are not. Who among us hasn’t cribbed from Wikipedia before? Or from Google Translate or an online dictionary or the Internet Archive? How often do we source our work from a disorganized repository of online published texts that we happen upon dilettantishly, rather than the old-fashioned way, through years of suffering and study? And how far removed is this process, really, from asking a chatbot for the same material? When my fellow writers, who built careers working from material they found for free online—just as I did—complain that others are unfairly deploying an advanced search engine to gather and compile the same information, I cannot join in the jeering. I shudder resignedly.
For that reason, sometime last year I swore a vow of “absolute chastity,” to borrow the phrase of another eccentric director, Éric Rohmer. That meant swearing off the Internet entirely. I decided that when writing magazine essays, I would only use material gathered from my own life or from paper archives in university libraries. (This rule is roughly equivalent to the one about location shooting in Dogme 95.) The idea was that only in the stacks could I do honest, satisfying work. Only in rigorously examining personal correspondence, draft manuscripts, faxes, etc., all yellowing with age, could I discover stories uninfluenced by the vulgarity of my own time.
Very quickly did I discover that I was mistaken. The only way to publish anything now is online (even if, like in this magazine, there is also a print product). And in publishing online you necessarily participate in the Internet’s ecosystem. A few weeks ago I wrote about a telegram from Rebecca West I found tucked inside my copy of Black Lamb and Grey Falcon; already the piece has wormed its way into the Internet’s mainframe and retrained Google Gemini to spit out new facts about the grande dame of English travel writing. I could detail many other such cases. That which was kept sacrosanct in a cardboard box is now splayed out promiscuously all over the Internet. Despite my best efforts to remain pure, when it comes time to publish, I have to give in to the machine.
What am I getting at here? I suppose you could say that there is no way out of this mess. “To resist was fatal, and it was impossible to fly,” Gibbon writes of the fugitive at the height of the Roman Empire’s power. And so it is with those who have sworn the new Vow of Chastity. A.I. is everywhere, and it is not going away. Even those of us who think its effects are crude and mentally degrading will find ourselves using it if we wish to continue to live reasonably within the bounds of American society.
Yet there is a way through, if not out. Trier and Vinterberg wrote that general accessibility makes the avant-garde more important than ever. Theirs is a common expression, one that comes tumbling out of some dissatisfied artist’s mouth every time a new fad or technology threatens his craft. My favorite version of it comes from the young Joyce, in a story told by his friend Padraig Colum. In 1902 Joyce was coming out of the National Library in Dublin when his friend stopped him in the stairwell. They walked down into the street and around the neighborhood, arguing for hours about Yeats, the Gaelic League, and Lord knows what else. At last Joyce stopped on a corner and turned to Colum. “I distrust all enthusiasms,” he said. He spoke not “with any youthful bravado,” Colum recalled decades later. Rather, Joyce stated his dissent as fact, “like one giving a single veto after a tiring argument.”
I distrust all enthusiasms—how many times have I repeated those words this year? And what argument could be more tiring than the one over A.I.?