In a world where state policy is shaped by A.I. hallucination, it’s cute to remember that consulting Wikipedia was once seen as irresponsible. Schools banned its use, and journalists were shamed for credulously repeating its hoaxes. “Reads like Wikipedia” was a literary insult. As a millennial, I had teachers who spoke of the encyclopedia in terms of epistemological stranger danger: You don’t know who wrote those articles and what they want you to believe! And, indeed, we do not. But we do know, a quarter century on, that Wikipedia—a text of thirty billion words, in three hundred languages—is the greatest compendium of human knowledge ever assembled. It is also, as Richard Cooke writes, “an escapee from a parallel universe, a refugee of the internet as it should have been.”

Discover notable new fiction, nonfiction, and poetry.

Cooke’s new book, “The Last Best Place on the Internet: A Human History of Wikipedia,” takes on this colossus, a relic of early-aughts idealism on a web that tech giants long ago carved into algorithmically barbarized fiefdoms. Like some crumbling Roman library in the Dark Ages, the encyclopedia is, in many ways, “both too big to understand and somehow too ephemeral to bother trying,” he writes. (The English-language Wikipedia alone would take forty sleepless years to read.) But he’s less interested in its content than he is in those who built it. The book’s driving insight is that Wikipedia is best studied as a subculture. It was neither the first online encyclopedia—Britannica had a web presence by 1999—nor even the first that users could edit. Rather, its vitality arose from the community of obsessives whom it attracted, and its decline has coincided with the market’s capture of their creativity.

Wikipedia has lost more than a billion monthly visits in the last four years—a twenty-five-per-cent drop that, because the site recruits its editors from its audience, also threatens its growth and quality. Much of its traffic has been intercepted by A.I., as ChatGPT, Claude, and Gemini overtake traditional search. And these chatbots were weaned on Wikipedia, “the single most important training corpus in the history of artificial intelligence,” Cooke informs us. In short, the encyclopedia is being destroyed by its biggest beneficiaries, like a golden goose slaughtered for slop-pâté. It’s a cruelly ironic fate for a project that began as a failed startup’s castoff. “Had anyone predicted Wikipedia’s true value during its formative phase,” Cooke writes, “it would never have survived.”

Invisible hands built Wikipedia, but those of the free market hardly figured. It wasn’t for lack of trying: Jimmy Wales, one of the project’s co-founders, was an Ayn Rand fan who worked in finance. He launched Wikipedia’s for-profit predecessor, Nupedia, at a company called Bomis, which owed most of its advertising revenue to a softcore-porn directory called the Babe Engine. (Denis Diderot, too, sold smut, Cooke tells us, in the form of an erotic novel that supported his work on the Encyclopédie.) Nupedia was written and peer-reviewed by scholars, and was eventually supposed to make money from advertising. But the editorial process was painfully slow, and the business model was doomed by the dot-com crash. Wikipedia was a desperate attempt to salvage the project through crowdsourcing. It went online in January, 2001, against the objection of Nupedia’s academic advisory board.

Wikis—or modular, collectively edited websites—preceded Wikipedia. They were invented by Ward Cunningham, a software engineer in Portland, Oregon, who wanted to assemble a community history of programming ideas. His project touched off a “Jurassic wiki period” of similar experiments. It was Larry Sanger, whom Wales had hired to oversee Nupedia, who hit on the idea of building a wiki encyclopedia, though the two men dispute the exact details. (Sanger later left Wikipedia, and has criticized it in conservative media outlets.) Cooke does his best to dramatize these origins, often through dry humor, but the personalities involved don’t quite rise to the level of an Aaron Sorkin drama. Happily, Cooke devotes the bulk of his narrative to everyday Wikipedians and their collaboration’s emergent properties.

Wikipedia began life “as an unapologetic first draft, full of experiments, doodles, and marginalia,” Cooke writes. One of its decisive early advantages was a low barrier to entry. Because articles didn’t require anyone’s approval to write or edit, they accumulated quickly—and, with them, reader-editors, who flocked to the site from Google’s search engine, which at the time was still new. Much of what they wrote was junk. Early Wikipedia was full of short articles, called “stubs,” and unnecessarily detailed information about fictional universes; the Washington, D.C., of “Atlas Shrugged,” for instance, had a page before the real city. Yet the site’s pop-friendliness drew traffic, and its poor quality lured editors; then as now, people loved to correct strangers online.

The site’s breakout moment was 9/11. In an era before social media, Wikipedians’ real-time aggregation of news about the attacks established it as a serious informational resource. This also demonstrated that volunteers who’d come to the site for their own niche reasons could redirect their efforts en masse. (Cooke cites the Wikipedia researcher Brian Keegan, who compares this “migration of expertise” to the decentralized coördination of corals and insect swarms.) Bomis briefly considered trying to monetize the site’s runaway success. But the rumor of ad sales led to an insurrection on Spanish-language Wikipedia, which threatened to break away. In 2002, Wales transferred control of the site to his new nonprofit, the Wikimedia Foundation: an admission that Wikipedia wouldn’t work if it wasn’t free.

The encyclopedia became a mecca for outsiders. Cooke notes that its main period of growth, between 2001 and 2008, coincided with the coming of age of a generation “denied the hallmarks of adulthood by a conspiracy of circumstance.” This included jobless graduates, millennial “failsons,” layoff victims, hardcore fans, and the neurodivergent—just about everyone with “abilities that were hidden, overlooked, undervalued, or simply too weird to express elsewhere.” One “very senior Wikimedian,” we are told, “came to think of the project as a ‘city built by Asperger’s.’ ”

Cooke’s narrative is studded with mini-portraits of pioneering “super editors.” We meet Ruth Ifcher, an ex-coder and Ph.D. so agoraphobic that she rarely left the house, and Einar Kvaran, an unemployed and uncredentialled art historian editing from his monograph-crammed trailer home in New Mexico. Derek Ramsey, a former lay preacher from the edge of Amish country, believed that “everybody should have their own hometown” in the encyclopedia—and wrote the site’s first bot, which scraped census data, to make it so. A more controversial figure was Ryan Jordan, or Essjay, who presented himself as a Catholic theologian, and who featured in a 2006 article about Wikipedia for this magazine. Alas, it had to be corrected: Essjay was, in fact, a mere ex-legal secretary, who, according to Cooke, had been fired for using his firm’s computers to hoard porn.

His unmasking fed an emerging panic over Wikipedia’s accuracy. Cooke samples an absorbing range of hoaxes, from the college students who convinced the world that an imaginary Scotsman had invented the electric toaster to the Chinese housewife who fabricated an entire war out of boredom. Other examples are more sinister—for instance, the far-right takeover of Croatian Wikipedia, or efforts by corporations to launder their images. (An Exxon employee doctored an article about an oil spill.)

Wikipedia began tightening its policies in the wake of a 2005 scandal involving the journalist John Seigenthaler, Sr., whose page falsely implicated him in the assassinations of John and Robert F. Kennedy. Unregistered editors lost the ability to create new articles, and sensitive categories, such as “Biography of living persons,” implemented stricter sourcing requirements. The Wikimedia Foundation persuaded cultural institutions to hire “Wikipedians in residence,” and Wales became the toast of TED. The professionalization worked, but at the expense of the site’s internal culture, Cooke argues, as generous eccentrics were displaced by neurotic lovers of rules.

Wikipedia has two main factions—the “Inclusionists” and the “Deletionists”—and their schism centers on how big the encyclopedia ought to be. The former held sway in the early days, but, around 2007, a year after English-language Wikipedia hit its millionth article, “a ‘Soup Nazi’ culture” set in “almost overnight.” Even Wales—affectionately known as the WikiPope, and studiously neutral about the schism—ran afoul of these “hanging judges,” who deleted his stub about a popular Cape Town restaurant. Nowadays, articles “nominated for deletion” have a sixty-five-per-cent conviction rate. The encyclopedia continued to grow, but its number of active editors hit a ceiling in 2007, and hasn’t recovered since.

Cooke makes no effort to hide his sympathies; at one point, he compares the Deletionists to bullies knocking down sandcastles at the beach. But are they really as responsible for the encyclopedia’s decline as their opponents pretend? After all, Wikipedia’s stagnation also coincided with the rise of social media, which Cooke hardly mentions. Could it be that the site’s lost generation wasn’t pushed away by Soup Nazi editors but pulled in by posting, podcasting, and short-form video? Annie Rauwerda, arguably the encyclopedia’s No. 1 booster, started out as an editor but made her name on Instagram, where she runs the wildly popular account DepthsofWikipedia.

Cooke’s more convincing charge is that Deletionism has kept Wikipedia parochial. Its editors are overwhelmingly male and Western, and their outlook is reflected in what’s deemed “notable” enough for inclusion. For many years, Pokémon were more thoroughly covered than most of Africa; the page for Makmende, a Kenyan superhero, was deleted over and over again for years, despite being the most popular meme in one of the Anglophone world’s most populous countries. Gender bias is similarly prevalent; in 2018, the physicist Donna Strickland’s article was culled, just a few months before she won the Nobel Prize. Wikipedians use edit-a-thons and affinity-based campaigns to address these disparities; a group called WikiProject Women Scientists actually inverted the “quality gap” in its category. But there are also deeper epistemological issues: What’s to be done about oral traditions, for instance? Or minority languages? (Greenlandic Wikipedia’s sole active admin had it shut down, arguing that its poor quality made it a threat to their culture.) Wikipedia has yet to overcome the blind spots of its Enlightenment predecessors, Cooke suggests: “Not only is most human knowledge not Western, most of it has never been written down.”

A.I. promises to exacerbate these problems. Wikipedia has already banned content generated by large language models, but it can’t stop people from outsourcing their inquiries to chatbots, Cooke writes, “breaking the virtuous chain of reading and writing that has sustained it so far.” L.L.M.s amplify the encyclopedia’s biases, which are repackaged as oracular truth, and shrink the community tasked with fixing them. For some A.I. boosters, that’s precisely the point. Last year, Elon Musk launched Grokipedia, an L.L.M.-written encyclopedia billed as a Wikipedia-killer. Humans can only suggest changes to articles, and there’s no forum for debating them. Grokipedia’s fate is uncertain—until recently, it had been “paused” for five months—but similar projects are surely in the offing. Cooke speculates that in the future, encyclopedists may be reduced to tending “the frost-bitten harvests of machines.”

Meanwhile, the same academia that once deemed Wikipedia dangerous is throwing open its doors to artificial intelligence. Chatbots, we are told, will empower students to engage with difficult primary sources—for example, by letting them debate theology with a virtual St. Ignatius de Loyola—or at least give professors a break from the thankless enforcement of A.I. bans. (A recent, widely criticized report on the state of the humanities even used L.L.M.s to identify politicized scholarship.) This embrace of the “inevitable” styles itself as a kind of humility: If the ivory tower misjudged Wikipedia, couldn’t the same be true of L.L.M.s? But there’s a difference between widening the conversation and sacrificing it to a simulacrum. Wikipedia’s “crime” was to democratize the encyclopedia, allowing anyone with an internet connection to participate in an Enlightenment project previously monopolized by experts. A.I., by contrast, threatens to expel us from the writers’ room of consensus reality, leaving to oligarchs what was once hashed out on the Talk page.

For all that Cooke celebrates Wikipedia, he writes as though it were already dead. He opens with an image of the encyclopedia literally lost on the moon—as part of the Lunar Library, a nano-sized repository of human knowledge that crashed in 2019—and repeatedly suggests that it may be the last such project that humans undertake. He may be right. But there’s still time to stand athwart history and shout, “Citation needed.” ♦