Facebook Shuts Down Machine Learning Systems for Identifying Viral Hoaxes

*Audible groaning*

Hoax on You

In response to the rising tide of social media-spread misinformation in the 2010s, the social media behemoth Meta-formerly-Facebook built machine learning systems that proactively identified and limited the spread of viral fake narratives. Now, as reporting from Platformer reveals, it's switched those hoax-tracking systems off.

The automated systems have targeted bonkers — and sometimes physically dangerous — narratives, like the eternally stupid Pizzagate conspiracy theory. By Meta's own admittance, the effort has been extremely effective: according to internal documents reviewed by Platformer, the machine learning tools have limited the spread of viral misinformation with an astonishing over 90 percent effectiveness.

Meta hasn't publicly addressed the choice to deactivate its hoax trackers. But last week, as flagged by Platformer, the company's recently-promoted new Head of Global Policy claimed in a blog post that Meta's torpedoed safety practices have unjustly demoted "too much content that our systems predict might violate our standards."

"We are in the process of getting rid of most of these demotions," the blog adds, seemingly alluding to Meta's automated misinformation-limiting programs, "and requiring greater confidence that the content violates for the rest."

Chain Letter Office

The decision to axe the hoax-trackers was made amid broader, deeply controversial changes to Meta's overall trust and safety practices across its many platforms, which CEO Mark Zuckerberg — the Silicon Valley giant's "unimpeachable king," as The Atlantic's Charlie Warzel recently put it, given the billionaire's iron-clad corporate control over the company — characterized in a video as a company effort to "get back to our roots around free expression."

These allegedly free speech-oriented changes include the sunsetting of its third-party fact-checking program, as well as a renewed allowance of dehumanizing language on the platform. Users across Meta platforms are now allowed to misgender trans people, refer to women as "household objects or property" or declare themselves to be "proud racists."

Efforts to combat bad or harmful information, meanwhile, will be regulated mostly to a user-centered model reminiscent of X-formerly-Twitter's "Community Notes" system.

Meta's move to phase out its automated hoax-tracking systems fits into this wider tapestry of Zuckerberg-led policy changes at the company. As it stands, it's unclear how these many changes will ultimately impact the day-to-day experience of Meta users, not to mention our broader information ecosystem. But they've certainly curried favor with incoming president Donald Trump, who previously — in a coffee table book, no less — threatened to jail Zuck over false allegations of election interference. Allegations, by the way, that Zuckerberg will presumably have to rely on random Meta users to correct.

More on Meta's current vibe: Facebook Is Creating Fake AI-Powered Black Women While Changing Its Rules So It’s Okay to Harass Real Ones