WIKI FAKES
Online encyclopedia wages battle with hoaxes, from an Australian deity to a Mexican hipster guru,
Jar’Edo Wens is an Australian aboriginal deity, the god of “physical might” and “earthly knowledge.” He’s been name-dropped in books. Carved into rocks. And, as of March, conclusively debunked.
There is no such figure, it turns out, in aboriginal mythology; instead, Jar’Edo Wens was a blatant prank, a bald invention, dropped into Wikipedia nine years ago by some unknown and anonymous Australian. By the time editors found Jar’Edo Wens, he had leaked off Wikipedia and onto the wider Internet.
He had also broken every other Wikipedia hoaxing record. At nine years, nine months and three days, Jar’Edo Wens is the longest-lived hoax found on the free encyclopedia yet.
Ask any diehard Wikipedian about hoaxes, of course, and they’ll call them a natural byproduct of the Wikipedia project: since the day the open-sourced encyclopedia opened for business in 2001, pranksters, vandals and other saboteurs have done their best to disrupt it.
But in the past year, Wikipedia hoaxes appear to have grown far more frequent — or at least far more visible. Editors have uncovered 33 major hoaxes since January, including several about fake bands and fake political parties. Of Wikipedia’s 16 most egregious hoaxes, 15 were discovered in the past six months. There’s no telling, of course, exactly how many hoaxes we simply haven’t yet dug up.
“There’s a lot of nonsense on Wikipedia that gets papered over,” says Gregory Kohs, a former editor and prominent Wikipedia critic. “Wikipedia is very good at catching obvious vandalism, like swearing and caps-lock. But non-obvious vandalism?” Not so much, he says.
To understand how misinformation spreads on Wikipedia, you must first understand how the site works. Anyone can edit Wikipedia, of course: more than 130,000 readers have done so in the past month. But because wide-open editing is an obvious recipe for disaster, the site is undergirded by a vast volunteer bureaucracy. These editors and administrators aren’t paid, and they aren’t technically affiliated with Wikipedia or Wikimedia, the aloof non-profit that oversees the site. But whether because they believe in Wikipedia’s mission, they like the power or they’re bored, they spend hours po- licing the site’s new changes, checking links, tweaking grammar and arguing on internal message boards.
Their success rate, by all accounts, is a pretty high one; in a recent interview with 60 Minutes, Wikipedia co-founder Jimmy Wales boasted that he no longer saw vandalism as much of a problem. And yet, critics like Kohs and his colleagues at the Wikipedia watchdog Wikipediocracy maintain that there are untold errors that editors don’t even know about, let alone fix.
On Monday night, Kohs wrapped up an experiment in which he inserted outlandish errors into 31 articles and tracked whether editors ever found them. After more than two months, half of his hoaxes still had not been found — and those included errors on high-profile pages, like “Mediterranean climate” and “inflammation.” (By his estimate, more than 100,000 people have now seen the claim that volcanic rock produced by the human body causes inflammation pain.) And there are more unchecked hoaxes where those came from. Editors only recently caught a six-yearold article about the “Pax Romana,” an entirely fictitious Nazi program. Likewise “Elaine de Francias,” the invented illegitimate daughter of Henry II of France. And the obvious eight-year-old hoax of “Don Meme,” a Mexican guru who materializes at parties and mentors hipster bands.
“I think this has proved, beyond a reasonable doubt, that it’s not fair to say Wikipedia is ‘self-correcting,’ ” Kohs said.
There is, surprisingly, not much data to conclusively confirm or deny this. While Wikipedia’s accuracy has been a favourite subject of study for Internet-minded academics, the usual methodology compares articles from an authoritative reference work with their Wikipedia equivalents. Since the Encyclopaedia Britannica doesn’t have articles on Jar’Edo Wens or Don Meme or Elaine de Francias, most studies that have trumpeted Wikipedia’s accuracy haven’t accounted for intentional hoaxes.
According to Wikipedia’s own, self-admittedly spotty records, once a hoax crosses the one-year mark, it can be expected to survive for another three years.
Though no one knows who started the Jar’Edo Wens hoax, it’s easy to track the history of the page. Wikipedia logs every change to every article page on a tab called “history,” just as it logs discussion of every article on the so-called “talk” page. Editors didn’t stop by Jar’Edo Wens’s page too frequently — it was, after all, pretty obscure — but someone did make a quick grammar fix in 2006, and someone flagged the page for lacking sources three years later.
In November 2014, an anonymous user tagged the page as a possible hoax: “Not found in several (reliable sources) on Aboriginal religion,” the user noted. The page still wasn’t immediately taken down; it’s Wikipedia policy to debate articles, sometimes at great length, before deleting them.
“Lacked sources for almost a decade,” one editor argued.
“The letters D,J, O and S are not used in the Arrente (Aboriginal) language,” another said.
On March 3, prompted by a Wikipediocracy post that made fun of the long deletion process, veteran Wikipedia administrator Ira Matetsky deleted the “blatant and indisputable hoax,” calling it an “embarrassment.” And yet Matetsky, like many of his fellow Wikipedians, counts the incident less as a loss than as a win. In the end, the system worked; the hoax was deleted. The growing number of hoaxes could suggest that Wikipedians are getting better at uncovering them.
“Wikipedia is uniquely vulnerable to deliberate mistakes,” Matetsky said. “But Wikipedia is also uniquely gifted at its ability to fix misinformation.”
Dozens of sophisticated, automated programs crawl the encyclopedia and delete vandalism, according to an evolving (and overwhelmingly accurate) algorithm. On top of those bots are the editors themselves, many of whom keep an eye on controversial articles or watch for suspicious additions to the “new page” feed.
As of this writing, there were 5,476 unreviewed pages in the English Wikipedia, the oldest of which had been around 111 days. I spotted a probable hoax — an unsourced article on an otherwise un- Googleable “new world religion” — within minutes of loading the page.
It’s not perfect, exactly. But Matetsky points out that newspapers and books and GPS systems also make minor errors every day.
“The question is not whether Wikipedia is more or less reliable than a day at the New York Public Library,” Matetsky said. “The question is whether Wikipedia is more or less reliable than whatever other results top Google search.”
But even given the growing awareness of hoaxes, what’s a Wikipedian to do? There are 4.8 million pages on the site’s English version, but only 12,000 veteran editors. That works out to roughly 400 pages per volunteer — far more than at any other time in the site’s history.
“Wikipedia could acknowledge that by now, it contains hundreds of thousands of articles on marginal topics that its volunteer system is simply unable to curate responsibly,” said Andreas Kolbe, another contributor to Wikipediocracy. Instead, he says, the Wikimedia Foundation has taken an alternative approach: dismiss each hoax as a one-off deception, and “lament how terrible it is that someone abused their trust.”
Reformers, both within and outside the site, insist there are other ways. You hear frequent references to a feature called “pending changes,” which was promised by Wikimedia in 2010 and again in 2012. The feature would hold new edits in a queue until an experienced editor could review them. On German Wikipedia, where “pending changes” has long been the norm, that little speed bump seems to work quite well.
Still, none of this changes the numbers problem at the core of Wikipedia. The site’s editor base has atrophied since 2007, and today’s editors are largely young, white, western men. It’s no coincidence that, in Kohs’s vandalism experiment, an error on an obscure New York canal was corrected, while lies about Ecuadorian customs, Indian legends and Japanese history were not. Likewise the Wiki-troll Jagged85, who meddled with articles about Islamic history for years; it was only when he messed with a video game page that he finally got kicked off.
“If the Jar’Edo Wens hoax had been about Greek or Roman or Norse mythology, it would’ve been found faster,” Matetsky admits. And yet, there’s some suggestion that even that wouldn’t have helped, that even snazzy new initiatives and more moderators couldn’t save the Internet from itself. For years, a group of interested editors waged an organized campaign to improve articles about indigenous Australians, including a page on “Aboriginal deities” that listed Jar’Edo Wens.
They added a photo and changed the page title. They grouped the deities under regional headings. Gradually, the campaign broke up, all without anyone noticing the invented aboriginal spirit: god of earthly knowledge — and its inevitable limits.