Zuck chucks fact-checkers to cosplay as Elon Musk
🤥 Faked Up #31: Meta terminates fact-checking program, Matryoshka migrates to BlueSky, and Spiegel identifies MrDeepFakes
This newsletter is an ~8 minute read and includes 74 links. Consider sharing it with your friends (or don’t, you do you).
Headlines
Telegram has blocked access to channels of Russian state-controlled media in Europe. Fact-checkers are calling out Bing for misrepresenting its partnerships in order to appear in compliance with EU commitments. NASA had to fact-check a post on X claiming the ISS was about to crash. Didier Raoult’s study on hydroxychloroquine as a COVID-19 treatment was retracted. Apple is under growing pressure to drop its garbled AI summaries of news alerts. The British government plans to criminalize creating and disseminating deepfake porn. Facebook tops the list of platforms used by scammers in Taiwan. Elon Musk’s opinion piece endorsing the AfD may have been AI-generated.
Top Stories
Meta terminates Third-Party Fact-Checking Program
On Tuesday, the CEO of a social network born to rate the hotness of college students announced he was taking it back to its roots by “restoring free expression.”
Mark Zuckerberg’s Free Speech Plan™ includes forcibly relocating his Trust & Safety employees to Texas1 and allowing a little bit more misogyny and homophobia. It also unwinds algorithmic scanning for harmful content that will result in more Meta users getting harassed.2
But his marquee announcement, the one Zuckerberg opened with because he knew it would get him the response he craved from the president-elect of the United States, was that he will terminate a long-standing partnership to fight misinformation.
The program pays fact-checking projects to review viral and potentially false claims through a dedicated tool. When they label something as false, that content sees its future reach reduced. In addition, a fact-checking label is added to the post that gives users the context with a link to the fact check (see below).
First launched in 2016 in response to criticism about rampant fake news, the program since grew into a multi-million dollar affaire involving 90 partners covering 130 countries. (This is a good moment to flag that as the director of the fact-checkers’ association from 2015 to 2019, I advocated for Meta to fight misinformation and was closely involved in some key discussions that shaped the program.)
Facebook has over the years repeatedly trumpeted this feature as a sign that it is a Very Responsible Social Network. Testifying to a group of US members of Congress in March 2021, Zuckerberg called the fact-checking program “unprecedented.” His written submission branded it “industry-leading.”
No longer. As of yesterday, fact-checkers are bad and they are censorious. Here’s Zuck’s unhinged statement on the topic in its entirety:
First, we’re going to get rid of fact-checkers and replace them with Community Notes similar to X, starting in the US. After Trump first got elected in 2016, the legacy media wrote nonstop about how misinformation was a threat to democracy. We tried in good faith to address those concerns without becoming the arbiters of truth. But the fact-checkers have just been too politically biased and have destroyed more trust than they have created, especially in the US. So over the next couple of months we’re going to phase in a more comprehensive Community Notes system.
There is so much bad faith reasoning in 96 words that it’s hard to know where to start. But let’s go in order.
Meta has 8 years’ worth of data to prove that the fact-checking program was biased. Zuckerberg shared none. Instead, he chose to ignore research that shows that politically asymmetric interventions against misinformation can result from politically asymmetric sharing of misinformation. As you can see in the fun little chart below, American conservatives tended to share more urls from false news websites on Twitter even when the definition of “false news” was left to a vote of a bipartisan group of laypeople rather than professional fact-checkers.
Back in 2016, Zuckerberg’s staff desperately sought an independent verification system for potential partners of the fact-checking program that couldn’t be gamed by blatantly bad actors like Alex Jones. The International Fact-Checking Network (IFCN)’s code of principles provided it; flawed as it is, it has stringent transparency requirements that are reviewed by an external assessor annually.
That code proved nonpartisan enough to allow for the certification of the conservative magazine The Weekly Standard, which was far from uncontroversial at the time. Meta’s US fact-checking partners also include Check Your Fact, a project tied to the Daily Caller, a website co-founded by the well-known liberal icon … *checks notes* … Tucker Carlson?
This is all stuff Meta’s CEO knows personally, by the way. When asked by Alexandria Ocasio-Cortez about Check Your Fact in another congressional hearing, Zuck said the IFCN had “a rigorous standard for who they allow to serve as a fact-checker.“
Beyond adopting a hyper partisan stance on media bias, Zuckerberg also didn’t mention that a big chunk of the content fact-checkers have been flagging is not political speech. Instead, it is the low-quality spammy clickbait that Meta platforms have commodified turning him into a gajillionaire who wears $900,000 watches.
PolitiFact, one of Meta’s US-based fact-checking partners, collects all the falsehoods it has labeled as part of the program in one place. I reviewed more than 100 of them and found that only about 21% were about politically sensitive posts about social security cuts or Trump’s impeachment. A similar share was spammy political hoaxes about Mitch McConnell’s health condition or Kash Patel calling the FBI ‘gay’. The relative majority (45%) was not political at all, including extremely sensitive speech about suspended NFL referees or a ship that’s pretending to be the Titanic.3
Zuckerberg justifying the termination of this program as a defense of free speech is particularly galling given that fact-checking labels did not lead to the underlying posts getting removed. Meta (rightly) took the approach of fact-checking as a contextual intervention which reduced a post’s reach but didn’t prevent users from continuing to access it. I am no First Amendment expert but this seems like a decent application of Justice Louis Brandeis’ exhortation to fight falsehoods with “more speech.”
The fact-checking program was not perfect and fact-checkers have no doubt erred in some percentage of their labels. Meta’s transparency report suggests this error rate in the EU may be as low as ~3%, orders of magnitude lower than the error rate for other demoted content (h/t EFCSN).
Meta had every right to eventually terminate its contract. But getting rid of fact-checkers in this manner was politics, not policy.
Typically for Meta, it is also a choice of American politics. In 2016, Facebook launched the fact-checking program once it was in hot waters with domestic stakeholders; earlier warnings from the Philippines had been ignored. Today, Meta’s non-US fact-checking partners are left assuming but uncertain that the program is terminated globally. The requirements under Article 34 and 35 of the Digital Services Act may mean Meta holds on to or tapers off its partnership in the EU more gradually.
And now for Zuckerberg’s proposed alternative to censorious fact-checkers.
It is unusual to see a CEO say he will emulate another platform’s product, especially after having threatened to wrestle that company’s owner.
But let’s assume that Zuckerberg is genuinely committed to a crowdsourced effort to combat misinformation in an unbiased and pro-speech manner.
He should probably read the research that suggests Community Notes users are motivated by partisanship and tend to over-rate their political adversaries. He should also probably be aware that as many as 90% of Community Notes never get displayed on X. How’s that for free speech!
I was running crowdsourced fact-checking projects into the ground long before Zuckerberg entered his bling-heavy middle age crisis. So I’m not in principle opposed to user-driven fact checks (more on that next week). But the quality of a crowdsourced project hinges on the incentive structures of the underlying crowd. There is very little in Meta’s history to suggest those incentives will be enlightened.
Matryoshka migrates to BlueSky
BlueSky’s Trust & Safety team did not have a very quiet holiday season. In mid-December, the platform blocked an extortion scheme demanding payment from prominent bloggers and journalists to hand over accounts and domains that were impersonating them. A few days ago, it blocked a small network of accounts trying to build a liberal following with AI-generated reply spam.
But the biggest headache was probably the fact that the pro-Russian influence operation known as Matryoshka has entered the fray. The network posted videos like this one where a clip of a real person introducing themselves is followed by unrelated footage voiced over by an AI-generated version of the speaker. The posts tag media outlets and fact-checkers, likely to bait them into engaging with the content.
As of Jan. 3, all accounts detected by the antibot4navalny collective had been suspended, and the company is promising more action in this space.
Meta is for AI slop
Last year, I semi-seriously predicted that the majority of new users on Instagram will post primarily AI-generated content. Apparently eager to prove me right, Meta announced on Dec. 27 a new push to let users create AI accounts that “exist on our platforms, kind of in the same way that accounts do.”
Users weren’t too happy about this, resurfacing and lampooning a dozen such accounts the company itself had created in late 2023. The AI accounts, bland stereotypes by definition, were deactivated after Meta discovered “a bug” that was making it impossible for users to block them (and not because they were an international embarrassment, of course).
But worry not, fellow travelers. There is plenty of AI slop still to be had. Maybe you will enjoy this fake talent show contestant that turns into a watermelon?
Behind Mr Deepfakes
Der Spiegel claims to have uncovered the identity of at least one of the operators of notorious deepfake porn website MrDeepFakes:
David D. [is] a 36-year-old who lives near the Canadian city of Toronto, where he has been working for several years in a hospital. Among several other indicators, DER SPIEGEL was able to identify him with the help of an email address that was temporarily used as a contact address on the MrDeepFakes platform. He apparently used the same email address privately for a running app. In the last 15 years, D. has registered an astonishing number of websites, many of them apparently rather dubious, as our reporting has found – including a platform for pirating music and software. David D. did not respond to numerous queries about his role with MrDeepFakes.
The German magazine also claims that the top 300 most active users on the website have been paid at least 100,000 euros in bitcoin over the past seven years. That feels low, but then maybe the driving motivation here isn’t primarily financial.
Noted
Critical Reasoning with AI: Initial Analysis of a Conspiracy Claim (The Ends of Argument)
Nothing Is Sacred: AI Generated Slop Has Come for Christmas Music (404 Media)
The Biggest Stories Indian News Outlets Misreported In 2024 (BOOM)
This company rates news sites’ credibility. The right wants it stopped (WaPo)
Why Misinformation Must Not Be Ignored (APA PsycNet)
The Year of the AI Election Wasn’t Quite What Everyone Expected (Wired)
Kash Patel Believes the FBI Planned Jan. 6th (The Bulwark)
Why YouTube Should Cut Off Its Thumbnails (New York Magazine)
Manchester Arena attack survivor calls for protection from conspiracy theorists (The Guardian) with Celebrity pastor plagued by Diddy misinformation on YouTube asks court to intervene (NBC Philadelphia)
Flat Earthers Befuddled as They Visit Antarctica and Earth Appears to Be Round (Futurism)
Police couldn’t catch my deepfake porn tormentor, so I did (The Times)
Scottish Parliament TV at Risk of Deepfake Attacks (Infosecurity Magazine)
You should watch all five minutes of his video because it is hard for me do his cringe justice.
Even around Election Day, posts being labeled were as likely to be about voting irregularities as they were to be spammy fakes about Trump saying he hated SNL or Melania Trump endorsing Kamala Harris.