[Originally published in Quillette]
The year is 1983. At 8pm sharp on the 28th of February, 106 million viewers tune in to watch the M*A*S*H finale. The next day, everyone from truck drivers to corporate attorneys had the same reference points. For years, M*A*S*H functioned as a national ritual, allowing families across the country to laugh at the same jokes and debate the same plot twists. Of course, it wasn’t just M*A*S*H. Everyone could assume that nearly everyone else had seen shows like Seinfeld or Dallas. And it wasn’t just the US either. In France, Bernard Pivot’s Apostrophes, a literary talk-show, regularly reached six million viewers, or 10 percent of the population. This anchored social life in common references, fostered a sense of collective identity. The logic carried to cinema (the Oscar ceremony attracted 85 million viewers in 1973), music (Michael Jackson, Madonna, Prince, Springsteen, or Queen appealed across demographics), video games (entire generations grew up with Tetris, Zelda, Pokemon or Super Mario) or even software (everyone used the few available corporate tools).
Fast-forward to today, and that monoculture has eroded, replaced by a landscape of personalized content appealing to different niches. In the US, the number of channels grew from three in the 1950s to hundreds today, with dozens of specialized channels (think MTV for music enthusiasts or ESPN for sports fans). In 1998, the Seinfeld finale reached 76 million viewers; seven years later, the CSI finale drew 27 million; by 2018, the top-rated show (The Big Bang Theory finale) managed just 18 million. Over these decades, the Oscars lost three quarters of their audience. Internet culture and streaming services obviously accelerated the fragmentation, with millions of viewing hours catered to sociological subgroups on Netflix or Disney+, and various segments of Gen Z doomscrolling on very different TikTok or Instagram feeds. The same applies to music (184 million tracks streamed in 2023, with 86% of them recording 1,000 plays or fewer), video games (players are clustered across Twitch communities, with little cultural overlap), and software (thousands of verticalized SaaS have replaced the old universal desktop suite).
Many essays have been written to analyse the fragmentation of culture, with explanations ranging from the primacy of individualism to the rise of identity politics. The culprit, however, is probably technology. When the costs of producing and distributing content are high, it is only worthwhile to produce the said content if it can appeal to a mass audience. When technology reduces the costs, creators can serve niches. You’ll only greenlight the production of a high budget movie if you hope to attract millions to the cinema, but you’ll be happy to produce a TikTok video if it can reach a hundred people. You need, say, 1 million clients to breakeven from a software that costs ten million to develop, but with costs reduced tenfold, you can address a use case ten times narrower. You’ll have no incentive to appeal to the long tail of music amateurs if distributing a soundtrack implies costly physical manufacturing, but once it becomes possible to upload a song to Spotify, you can earn a living by satisfying a micro-audience craving a micro-genre.
In other words, when creation is expensive, it is scarce and audiences converge on the few things that do exist. Common culture emerges. As creation becomes cheaper, audiences fragment. Now what if creation wasn’t cheap, but free? Of course, you can guess where this is going. AI enters and suddenly, creation does become free. Now, for a content to be worth producing, it must appeal to one person. AI gives us cultural niches of one. Culture was fragmented; it will become atomized.
Take software. Despite the trend towards personalization, many humans still rely on shared tools (often suboptimal for their exact needs) simply because, as Anish Acharya of A16z recently put it, « the preferences of 20 million developers dictate the software we all use. ». Tomorrow, vibe-coding will make it possible for anyone to build software tailored to their own constraints, without relying on off-the-shelf solutions used by at least a few fellow citizens. Exit Excel, enter custom tools for ultra-specific enterprise processes, equipped with a dashboard tuned to a given employee’s reasoning style and UI/UX preferences. Exit the App store with solutions available to the masses, enter personal apps generated by yourself for your idiosyncratic routines. A journalist at the New York Times documented building a fridge-photo analyser suggesting school lunches for his kids, a custom podcast transcriber, and a furniture-fit calculator calibrated to his exact car trunk. No other person on the planet will use these apps.
Take cinema. When there was such a thing as actors whose salaries had to be paid, you needed to breakeven by attracting at least a few spectators to your screenings (except if you were lucky enough to be part of the fully subsidized French film industry). Tomorrow, if (when) it becomes possible to generate a high-quality, full-length film by entering a custom prompt (“Gladiator-like but set in my hometown with a hint of science-fiction and a strong-female lead, lasting only one hour because I want to sleep early”), the calculus changes. Tailored to the whims of one spectator and discarded after each viewing, these movies won’t endure to bind people together (when things are free and available in unlimited quantity, they don’t last – among other reasons because the cost of storing them becomes higher than the cost of regenerating them).
Or take video games. For better or for worse, millennials grew up playing GTA, World of Warcraft or Call of Duty, trading tips about maps, gameplay or tactics. In the future, however, video games, augmented by AI world models, could offer custom narratives for each player, evolving maps, and difficulty levels optimized to keep users addicted. No two players would ever experience the same storyline. Extend this to music, novels, or art, which AIs could soon calibrate to your mood, personality or life story. A third of content on Deezer is already-AI generated (with most users failing to tell the difference), and in the future, algorithms could push to each user AI generated content that fits him – and only him – exactly right.
Consider what is already happening to non-fiction : when people wish to learn about a new topic, they increasingly turn to one-on-one conversations with LLMs rather than to essays or articles they could later recommend to friends. “Speak to Grok”, is an answer I often get when asking for reading recommendations. (The habit of seeking advice from others via Reddit or such forums - which produced pockets of shared experience - will fade as people default to dialogues with LLMs. The number of posts on Stack Overflow decreased significantly after the release of ChatGPT.) Even schooling, where in the past millions experienced the same rituals, textbooks and exams (which forged generational frames of references) could become individualized. Some Silicon Valley investors predict that AIs will soon provide material tailored to each student’s cognitive abilities and learning curve. Google recently released “Learn Your Way”, an app that rewrites textbooks based on each student’s quirks, creating personalized audio lessons, mind maps or quizzes. The world is atomizing.
My colleagues think I am too pessimistic. They point to the many forces that could work against such a future. Some for example argue that by lowering the barriers to creation, AI will break the monopoly previously enjoyed by a few cultural gatekeepers, giving every creative genius the means to express his talent. This could raise the ceiling of what humanity produces. Although slop for micro-niches could multiply, exceptional art could also emerge, compelling millions to converge on them.
Others observe that the human need to bond with others won’t disappear overnight. Technology changes, culture evolves, but the biological predispositions encoded in our brains through millions of years of evolution do not. The forces of mimetic desires, the impulse to watch what others watch and play what others play, will limit fragmentation. To which I reply that technology can provide virtualized emotional rewards. The craving for adventure and adrenaline has been virtualized through video games. The craving for sex has been virtualized through pornography. The need for social belonging could become virtualized too, with AI mimicking the cues of human connections and “tricking” our evolutionary wiring.
Still, I tend to agree with them : humans will always find a way to connect. What forms our shared experiences will take, however, is an open question. Those who manage to answer it may find themselves building something important.