A former fact checker on Meta’s big changes

Summary

POLITICO Tech Podcast Summary
Episode: A Former Fact Checker on Meta’s Big Changes
Release Date: January 13, 2025
Host: Stephen Overle
Guest: Alexios Mansarlis, Director of the Security, Trust and Safety Initiative at Cornell Tech

---

Introduction to Meta's Content Moderation Overhaul

In the January 13th episode of POLITICO Tech, host Stephen Overle delves into the significant shifts in Meta's content moderation policies following comments made by CEO Mark Zuckerberg. The primary focus is on Zuckerberg’s recent decision to dismantle the fact-checking system previously supported by the International Fact Checking Network (IFCN), an initiative co-founded by guest Alexios Mansarlis.

Zuckerberg's Critique of Fact Checkers

Stephen Overle opens the discussion by highlighting Zuckerberg's sharp criticism of Meta's former fact-checking team. Zuckerberg labeled the fact checkers as "too politically biased" and accused them of eroding trust rather than building it, particularly in the U.S.

Mark Zuckerberg (quoted by Overle, [00:57]): "The fact checkers have just been too politically biased and have destroyed more trust than they've created, especially in the U.S."

Zuckerberg announced the replacement of fact checkers with a community-driven model, similar to Elon Musk’s approach on the social media platform X. This move has been perceived as an attempt to align Meta more closely with the incoming administration of President Elect Donald Trump and other Republican figures, sparking diverse reactions across the political spectrum.

Guest Insights: Alexios Mansarlis on Meta's Policy Shift

Stephen Overle introduces Alexios Mansarlis, who brings extensive experience in trust and safety from his tenure at Google and as the founding director of the IFCN. Mansarlis provides a critical perspective on Meta's abrupt policy changes.

Alexios Mansarlis ([02:52]): "Zuckerberg took a hyper partisan, politically charged, termite filled language to take it down, rather than a more measured approach that would be appropriate for the CEO of one of the largest online platforms."

Mansarlis expresses surprise not at the decision itself, which he anticipated, but at the manner in which Zuckerberg executed it—characterizing the language used as "hyper partisan" and "politically charged."

Political Motives Behind the Changes

Overle probes whether Zuckerberg's actions were intended to appease President Elect Trump. Mansarlis concurs, pointing out the immediate references to Elon Musk and Donald Trump in Zuckerberg's termination statement.

Alexios Mansarlis ([03:22]): "He mentions Elon Musk's acts immediately... and then he mentions Trump's election immediately. So there were many ways to terminate this program."

He further notes that Zuckerberg framed this move as part of a broader five-step plan to "restore free expression to the social network," signaling alignment with Trump’s anticipated policies.

Debunking Claims of Bias and Censorship

Overle challenges Zuckerberg’s assertion about bias and censorship, asking Mansarlis for his disagreement. Mansarlis counters by highlighting the lack of transparency from Meta regarding the data that could substantiate Zuckerberg’s claims.

Alexios Mansarlis ([04:25]): "Zuckerberg sits on eight years worth of data that could very much prove or disprove his assertion. He chose not to share anything, not even a headline statistic."

He references a study by Dave Rand and Gordon Pennycook which found that "US Conservatives shared more false news on Twitter than the US liberals," suggesting that perceived bias might stem from the distribution of misinformation rather than the fact-checkers themselves.

Effectiveness of Community Notes

The conversation shifts to the efficacy of the new community notes model. Mansarlis expresses skepticism, drawing from his experience with crowdsourced fact-checking.

Alexios Mansarlis ([07:17]): "Most people are motivated by counterpartisan reasons. So they go and they fact check people they disagree with. There's a big delta. So this is bias."

He also points out that only about 10% of community notes actually get affixed to a tweet on X, indicating limited visibility and impact.

Scalability and Future of Fact Checking

Addressing scalability concerns, Mansarlis acknowledges the challenges but counters Zuckerberg’s criticisms by highlighting Meta's combination of human and algorithmic fact-checking efforts.

Alexios Mansarlis ([08:59]): "The error rate was about 3%, which was much, much, much lower than the error rate on all the other abuse vectors. So this combination of like human plus machine can work."

However, he remains doubtful about the long-term viability of fact-checking given the current political climate in the U.S., contrasting it with other regions like Brazil and the European Union where legal frameworks support misinformation mitigation.

Alexios Mansarlis ([11:36]): "The pendulum right now in the US has swung against misinformation interventions... So the big question here will be to what extent."

Implications for Trust and Safety

Mansarlis expresses concern over Zuckerberg’s public denouncement of fact-checkers, suggesting it undermines trust in Meta’s platforms.

Alexios Mansarlis ([13:08]): "He has said it is fine for me for more meta users to get harassed... he was full on cosplaying as Elon Musk."

He emphasizes the importance of rebuilding trust incrementally, especially in the face of increasing AI-generated content and impersonation threats.

Conclusion: The Road Ahead for Fact Checking on Social Media

As the discussion wraps up, Mansarlis remains pessimistic about the future of fact-checking on social media platforms, highlighting the challenges posed by political polarization and technological advancements.

Alexios Mansarlis ([13:56]): "We need to rebuild the trust one by one. But when you mix that with kind of this avalanche of AI generated content and impersonation, it's going to be a rough few years."

Host Stephen Overle thanks Mansarlis for his insights, concluding the episode with reflections on the critical state of content moderation and trust in social media.

---

Key Takeaways

• Meta's Policy Shift: Mark Zuckerberg has replaced Meta's fact-checking system with a community notes model, prompting debates on bias, effectiveness, and political motivations.

• Critique of Zuckerberg's Approach: Alexios Mansarlis criticizes the partisan language used in terminating the fact-checking program and questions the transparency of Meta’s decision-making process.

• Effectiveness of Community-Based Fact Checking: While community notes offer transparency, their low visibility and inherent biases due to partisanship pose significant challenges.

• Scalability Concerns: Despite Meta's integration of human and algorithmic fact-checking, scalability remains a hurdle, exacerbated by the vast volume of content on platforms.

• Future of Trust on Social Media: Rebuilding trust in social media platforms requires incremental efforts, increased transparency, and robust defenses against emerging threats like AI-generated misinformation.

---

Notable Quotes

• Mark Zuckerberg: "The fact checkers have just been too politically biased and have destroyed more trust than they've created, especially in the U.S." ([00:57])

• Alexios Mansarlis: "Zuckerberg took a hyper partisan, politically charged, termite filled language to take it down, rather than a more measured approach that would be appropriate for the CEO of one of the largest online platforms." ([02:52])

• Alexios Mansarlis: "Most people are motivated by counterpartisan reasons. So they go and they fact check people they disagree with." ([07:17])

• Alexios Mansarlis: "We need to rebuild the trust one by one. But when you mix that with kind of this avalanche of AI generated content and impersonation, it's going to be a rough few years." ([13:56])

---

This comprehensive summary encapsulates the critical discussions and insights from the POLITICO Tech episode, providing readers with a clear understanding of the complexities surrounding Meta's recent changes to content moderation and the broader implications for fact-checking in the digital age.

Transcript

[Eduardo]

Imagine what's possible when learning doesn't get in the way of life. At Capella University, our game changing flexpath learning format lets you set your own deadline so you can learn at a time and pace that works for you. It's an education you can tailor to your schedule. That means you don't have to put your life on hold to pursue your professional goals. Instead, enjoy learning your way and earn your degree without missing a beat. A different future is closer than you think with Capella University. Learn more at Capella. Eduardo.

[Stephen Overle]

Hey, welcome back to Politico tech. Today's Monday, January 13th. I'm Stephen Overle. When Meta CEO Mark Zuckerberg rolled out changes last week to how the company will moderate content, he had a sharp critique of the people who had been fact checking misinformation on his platforms.

[Alexios Mansarlis]

The fact checkers have just been too politically biased and have destroyed more trust than they've created, especially in the U.S.

[Stephen Overle]

Zuckerberg went on to say that he was replacing those fact checkers with a community notes model that allows users to flag content they see as misleading, the same system used by Elon Musk over on his social media platform X. It's one of several changes Zuckerberg announced that put him in the good graces of President Elect Donald Trump and other Republicans, and that has stirred up a lot of opinions. Last week I spoke with John Samples, a First Amendment scholar at the Cato Institute who also serves on Meta's oversight board, and he was a fan of Meta's new rules. But today I wanted to hear from someone who has experience with social media fact checking. So I called up Alexios Mansarlis. He's the director of the Security, Trust and Safety Initiative at Cornell Tech, and before that he worked on trust and safety issues at Google. But back before even that, from 2015 to 2019, he was the founding director of the International Fact Checking Network. That's the consortium of fact checkers that partnered with Meta to flag misinformation. I called up Alexios to get his reaction to Meta's policy changes and to Zuckerberg's assertion that fact checkers are politically biased. Here's our conversation. Alexios, welcome to Politico Tech.

[Alexios Mansarlis]

Thanks for having me.

[Stephen Overle]

You know, Meta announced sweeping changes to its content moderation practices, including eliminating a fact checking system that at one point you were involved with. You heard that and thought, what?

[Alexios Mansarlis]

Part of me was like, oh, this was a long time coming surprise. It lasted eight years. The part of me that was shocked, however, was the way it happened. Zuckerberg took a hyper partisan, politically charged, termite filled language to take it down, rather than a more measured approach that would be appropriate for the CEO of one of the largest online platforms.

[Stephen Overle]

Well, that's. I was going to ask this and I'll just get into it now. I mean, do you think Zuckerberg did this to try to appease incoming President Trump?

[Alexios Mansarlis]

I think if you look at the 96 words that he used to terminate the fact checking program, it's hard to think anything other than that. He mentions Elon Musk's acts immediately. Unusual for a company to refer to a competitor's product, especially if that competitor has threatened to wrestle you in the coliseum. And then he mentions Trump's election immediately after. So there were many ways to terminate this program. It's also, incidentally, the first thing he mentioned in his five step plan to restore free expression to the social network, which was again, a signal because he was expecting, craving, and ultimately got a positive reaction from the President elect.

[Stephen Overle]

You know, when he was rolling out these changes, Zuckerberg said that the fact checking system that was in place was too politically biased and censored too much speech, which is a contrast from what Metta has said in the past and sort of defended itself against accusations of conservative censorship. You disagreed with his take? I think it's fair to say. Why is that?

[Alexios Mansarlis]

Yeah, I mean, let's unpack that. Right. There's the bias part and there's a censoring part. And at the top of it all is Zuckerberg sits on eight years worth of data that could very much prove or disprove his assertion. He chose not to share anything, not even a headline statistic. Right. So that already tells you something. But unbiased. There was recently a study by Dave Rand, Gordon, Pennycook, et cetera, that found that US Conservatives shared more false news on Twitter than the US liberals. And to be clear, false news here wasn't defined by fact checkers, wasn't defined by academics. It was defined by a politically balanced group of laypeople. So equal numbers of Democrats and Republicans. And so if Zuckerberg is seeing an imbalance in the fact checking on his platform, it may well be to to an imbalance in the sharing of some of the false loops. And he could have opened up the data for ulterior study rather than trashing the program altogether. The approach here was not to remove any content that fact checkers were labeling. Right. It was to add an interstitial big label saying, this has been fact checked. Here's a Link. If you look at, you know, Justice Brandeis, I'm not a First Amendment scholar, but I know enough. The counterspeech argument, I.e. you address falsehoods with more speech. This was a counter speech approach, not a censorious one.

[Stephen Overle]

Got it. So I guess one way I might play devil's advocate is that fact checkers are people and all people have a point of view. Right. As someone who's close to this process, does that not creep into it?

[Alexios Mansarlis]

Yeah. Look, obviously there are issues with how you might select the claims you fact check. There are issues with what salience you give to different topics in a limited world. But generally speaking, these fact checkers abide by a code of principles that requires them to fact check everyone equally to have corrections to accept submissions. So if there was a systemic issue, I think Zuckerberg would have seen it in his data.

[Stephen Overle]

I guess the statement that you heard from Zuckerberg make, I'm curious, what about that stood out to you most as someone who follows this so closely?

[Alexios Mansarlis]

I mean the whole statement is worrisome. Right? The idea that they are rolling back some of the trust and safety protections altogether is going to lead, and he said so, to more harassment and hate to meta users. This is to me quite shocking. Right. These are always hard trade offs. Right. To be clear. But he has said it is fine for me for more meta users to get harassed, but generally I think it's the bad faith argument that he used across the board. If he thinks that excess committee notes or some form of beacon of bipartisanship and censorship, free discussion, he is deeply mistaken.

[Stephen Overle]

I was going to ask you how well you think that model works because he did call out X specifically and that is what he's now going to try to emulate on meta platforms.

[Alexios Mansarlis]

Yeah, and look at the top. I should know. I'm a big fan of the idea of crowdsourced fact checking. I ran a crowdsourced fact checking project into the ground 11 years ago and still wish it had lasted longer than a year and a half. But look, access community notes. Here are things. The one good thing I'll say is that it's transparent. All of its data is available. That's great. And that's great because we can assess it. Two things we can assess of it. One, and this has been in a study either in nature or science. Most people are motivated by counterpartisan reasons. So they go and they fact check people they disagree with. Right. There's a big delta. So this is bias. It's another form of bias, but it's bias. And the other thing which is just amusing to me is that on any given day only about 10% of community notes actually gets affixed to a tweet on X. That means that the vast majority of output of Community Notes users, and there are tens of thousands of them, never sees the light of day. How this is somehow a non censorious approach to speech and moderation, I just don't know.

[Stephen Overle]

That's an interesting point you make because I recently spoke to John Samples for the podcast. He's a First Amendment scholar at the Cato Institute and a member of Meta's oversight board. One point he made to me was that fact checking ultimately is not scalable in his view, especially when you consider the volume of content on platforms like Facebook or Instagram. And there is, I think, some truth to that in a way that, you know, there is so much content for human beings, at least actively fact check all of it would be impossible. Is that a fair criticism of the fact checking model? And how do you overcome that 100% right?

[Alexios Mansarlis]

Even if there wasn't a technology mismatch, it is much harder to fact check something than to lie. It just takes more time to do the former. But Meta ironically had a solution to this, right? It was already fuzzy matching with algorithmic tools, claims that had been fact checked by human raiders, by the fact checkers with other claims in the wild. And in a report it had to provide to the eu again, transparency obligations work. It showed that the error rate was about 3%, which was much, much, much lower than the error rate on all the other abuse vectors. So this combination of like human plus machine can work.

[Stephen Overle]

I was going to ask you if curtailing that algorithmic filtering is really the bigger deal here, because that is one thing Zuckerberg also said was like they were taking down too much content and now at the risk of bad stuff still getting through, they're going to wind that back.

[Alexios Mansarlis]

Yeah, and some of this, you know, he hinted at which topics are still going to get algorithmic treatment. And I think you mentioned drugs, terrorist content, kid safety, incidentally. All stuff that has legal repercussions. Right. If you get it wrong. So we should assume that hate and harassment are going to be things where there's going to be a lot looser approach and this will require users to report more. Right. And it's not always evident if I am targeted by non consensual intimate imagery, deep fake nudes, where and how and what I report to whom and why. And so yeah, we're just going to See plenty more harm on Meta's platforms.

[Stephen Overle]

Oh, such a clutch off season pickup, Dave.

[Dave]

I was worried we'd bring back the same team.

[Stephen Overle]

I meant those blackout motorized shades.

[Dave]

Lines.com made it crazy affordable to replace our old blinds.

[Alexios Mansarlis]

Hard to install.

[Dave]

No, it's easy. I installed these and then got some from my mom. She talked to a design consultant for free and scheduled a professional measure and.

[Alexios Mansarlis]

Install hall of fame.

[Dave]

They're the number one online retailer of custom window coverings in the world.

[Alexios Mansarlis]

Blinds.com is the goat shop blinds.com right now. And get up to 45% off select.

[Stephen Overle]

Styles, rules and restrictions may apply. I'm curious what you think the future is of fact checking on social media, if there even is a future for fact checking on social media. I mean, obviously, you know, you were involved in the international fact checking network. You were the founding director of it. Is there a future for this on social media platforms?

[Alexios Mansarlis]

Yeah, I mean, look, the future of fact checking, specifically we should ask fact checkers because it's been a while for me at this point. But I think the pendulum right now in the US has swung against misinformation interventions. But the US despite the way it is treated by the big platforms, is not the world. There are countries where there are legal obligations to take mitigation against misinformation. Think Brazil, think the European Union under the dsa. And so the big question here will be to what extent? And it's no surprise, it's no surprise that Zuckerberg's five minute message ends with a plea to Trump to help him fight the censorious euro.

[Stephen Overle]

Was that worrying to you?

[Alexios Mansarlis]

I mean, I don't know if worrying is the right word, but it was a shockingly political message for five minutes, kind of delivered with this seeming monotone in his posh house, wearing a $900,000 watch. Right. It was just a blunt and hyper partisan take that really has me thinking he was full on cosplaying as Elon Musk.

[Stephen Overle]

One of the comments that Zuckerberg made, which I do think there's some truth to, is about a lack of trust in these platforms. And that's just not unique to social media. I mean, the news media are dealing with a lack of trust. Institutions in general are dealing with a lack of trust. I'm just curious for your opinion, like, what's the solution to that?

[Alexios Mansarlis]

Yeah, million dollar question, right? Frankly, I don't know. I mean, in general, I do think opening up processes to everyday users is a good step. Greater transparency is a good step. But you know you have fact checkers who put up their bank statements and full funders lists and still it doesn't help when there is a weaponization of this sense of trust when people like Zuckerberg just throw an entire industry under the bus in a flip and one sentence. So I don't know, I don't have a ton of hope frankly. I think we need to rebuild the trust one by one. But when you mix that with kind of this avalanche of AI generated content and impersonation, it's going to be a rough few years.

[Stephen Overle]

Well, Alexios, I appreciate you being here on Politico Tech.

[Alexios Mansarlis]

Thanks for having me again, Steven.

[Stephen Overle]

That's all for today's Politico Tech. If you enjoy Politico Tech, be sure to subscribe. And for more tech news, subscribe to our newsletters, Digital Future Daily and Morning Tech. Our managing producer is Annie Reiss. Our producer is Afra Abdullah. I'm Stephen Overle. See you back here tomorrow.