The Old Fears, Faster

The fears about artificial intelligence are the fears about the internet, rerun at twice the speed. When the public internet arrived in the mid-1990s, four anxieties traveled with it. One concerned human contact, a sense that conversation would erode as it migrated to screens. Another concerned regional culture, a worry that local identity would flatten under a single global signal. A third concerned the limits of knowing, the prospect that anyone, anywhere, could learn anything with no curator to vet the source. The last concerned gatekeeping, the dread that quality would dissolve once newspapers, publishers, and broadcasters lost their grip.

Each fear has had time to prove itself or fail. The honest verdict places the original critics somewhere between paranoid and prophetic. They were correct about the direction and wrong about the timing.

Start with human contact. In the late 1990s, when AOL chat rooms and Usenet groups were the social internet, the optimists were right. Shy people found communities. Sick people found support groups. Hobbyists found other hobbyists across continents. The fear that screens would replace presence looked overheated, even hysterical. Twenty years later, the picture has reversed. Loneliness has become a clinical category. Adolescent depression rates have climbed alongside smartphone adoption. The Surgeon General issued an advisory on the loneliness epidemic in 2023. The early defenders of online socializing accurately described a real moment of social expansion. That moment did not last.

Regional culture is the clearest case of the four. Here the worry was correct from the start, and the optimists underestimated the velocity. Local newspapers have closed at a rate no one in 1995 predicted, with thousands of American papers gone since 2005 according to Northwestern University’s Local News Initiative. The classifieds that once funded local journalism migrated to Craigslist and then to nothing. Regional accents have weakened among younger speakers fed on a national stream of YouTube and TikTok. Independent bookstores have recovered modestly in the last decade, but the broader pattern is cultural homogenization. The word “global” stopped meaning international trade and started meaning a single English-language algorithmic culture.

On the limits of knowing, the optimists won the early argument and then the argument changed under them. Search did democratize information. A high-school student in rural Nebraska can read primary sources that were once locked in a university library. That is a real gain and should not be waved away. Yet the gatekeeping function that universities and publishers performed worked at two levels: access control on the surface, and epistemic discipline underneath, with someone, somewhere deciding what counted as a serious source. When that function collapsed, misinformation spread at the speed of the bandwidth that delivered it, and the average citizen became responsible for a vetting task the average citizen has no training to perform. The 2020s conspiracy ecosystem owes much to that 1990s argument.

Gatekeeping for quality is the most complicated of the four. The bloggers who rose up against legacy media were correct that the old gatekeepers had real biases and real blind spots. What they got wrong was the character of the replacement. The corner bookstore lost ground to a logistics company that has no commitment to books as such. Independent musicians found themselves selling into an algorithmic marketplace that pays them a fraction of a cent per stream. The local hardware store closed when fulfillment warehouses joined the big-box chains in undercutting its prices on convenience and reach. The old gatekeepers were imperfect. Their replacement is worse along axes the original critics never imagined.

The list of things the internet made null runs longer than the four named fears. Print encyclopedias closed their presses. Travel agencies thinned to specialty boutiques. Video rental stores vanished except as nostalgia objects. Photo developing shops collapsed when phones replaced film. Phone books became novelty mailers. Newspaper classifieds went to Craigslist and then nowhere. Tower Records and Borders died. Privacy in any nineteenth-century sense ended. Long-distance phone bills disappeared. Letter writing as a sustained personal practice faded. Watch repair, camera, stationery, music, and small specialty bookshops all thinned or vanished. The internet killed these things by killing the economic conditions that allowed them to exist.

The fears about AI rhyme with the fears about the internet without repeating them exactly. Critics again warn of job losses, of cultural homogenization, of a collapse in the standards by which we judge writing and image-making, and of the erosion of human relationships once people start talking to machines instead of each other. The structural difference is in the mechanism. The internet primarily expanded what a human worker could do. The new machine does both, expanding and substituting, and the substituting half is what is new. That is a different argument economically, and the historical comparisons that comforted technology optimists in 1995 may no longer apply.

Consider the historical comfort first. Every previous wave of automation eliminated specific jobs while creating new ones. The horse-and-carriage industry collapsed and the auto industry rose. Telephone switchboard operators disappeared and call-center workers appeared. Bank tellers were thinned by ATMs and bank-customer ratios shifted toward financial advice. Each transition was painful for the displaced and beneficial in aggregate over the long run. The optimists assume the new wave will follow this pattern. They may be right. There is also a specific way they may be wrong, and it matters.

Here is the specific way. Previous automations replaced muscle or routine clerical labor. They left intact a large category of cognitive and creative work that humans alone could perform. The new machine targets precisely that category. If the writing, editing, drafting, summarizing, designing, scheduling, programming, translating, and analyzing functions of the modern economy become machine functions, the question is whether new jobs will appear fast enough to absorb the displaced workers, and whether those new jobs will pay what the old ones paid. The 2023 Goldman Sachs estimate that generative tools could affect 300 million jobs globally describes a restructuring of daily tasks, which is a different claim from 300 million people losing their jobs outright. The honest question is what happens during the restructuring, and who pays for it.

The economic differences with the internet wave are sharper than the parallels. The internet rolled out over a decade. The new wave is rolling out over months. The internet required user adoption, since someone had to want to log on, learn the tool, change a habit. The new tools are being installed underneath people who have not meaningfully consented and often do not know how the tools shape their experience. The internet produced a small number of winners (Amazon, Google, Facebook, Apple, Microsoft) over twenty years. The current wave is producing a smaller number of winners over a much shorter window, with the same hyperscalers underwriting most of the new entrants. Capital concentration is more extreme, the speed is faster, and the optionality for workers to retrain is lower.

The cultural worry is harder to assess because it rests on aesthetic judgments rather than economic ones. The concern that machine writing will flatten prose into a sludge of competent, characterless paragraphs is correct as far as it goes, yet it underestimates the response. Readers notice when prose has no voice. Editors are starting to filter for it. The market will probably bifurcate, with cheap machine-generated content occupying the bottom of the page and human-written work commanding a premium it has not commanded in twenty years. The same may happen with images, with music, with film. Human creativity will survive. The reasonable worry is its economic marginalization in mainstream commercial production.

The worry about human contact is the one most worth taking seriously, because it learned from the internet’s history. If chat-based companions become widespread, and if they are designed to be more agreeable than any real person can be, the question is what kind of human being is produced by a relationship that never disappoints, never demands, never pushes back, and never insists on its own existence. The 1990s critics of online socializing turned out to be right twenty years late. The 2020s critics of machine companionship may be right sooner.

So were the original internet fears honest, or were they fear-mongering. They were honest. They were also early. The critics described real risks and the optimists described real benefits, and both descriptions held true at different moments of the same technology. The new wave will probably follow the same pattern. Real gains will arrive early. Dismissals of the warnings will look careless in fifteen years. The history of the internet is the strongest argument against believing the boosters now, and the strongest argument against believing the apocalypse hawks. Reality is less spectacular than the boosters claim and more grinding than the doom narrators imagine. The gap between those errors is where we have to live.