[music]

Brian Lehrer: It's The Brian Lehrer Show on WNYC. Good morning again, everyone. Next time you're scrolling on your Facebook newsfeed, consider this, why am I seeing or not seeing what I'm seeing or not seeing? A recent study published in an Atlantic piece titled Facebook Has a Superuser-Supremacy Problem found that a very small minority of hyperactive often abusive superusers are largely responsible for much of the content catapulted to the top of everyone else's newsfeed. This is thanks to Facebook's algorithm, which elevates certain posts depending on the different ways that users engage with them.

It causes the top 1%, listen to the stat, causes the top 1% of Facebook's 230 million users in the US to be responsible for about 45% of all the content reaching the eyes and ears of other Facebook users and, big surprise, much of the content being spread far and wide because of the superusers is misinformation, racism, homophobia, antisemitism, and other forms of hate and verbal violence. In fact, in analyzing Facebook's superusers, one thing that came up in the datasets were many of the individuals indicted for their roles in the January 6th insurrection.

Facebook, of course, has publicly condemned hate, misinformation, and certainly insurrection and violence, and vowed to fight it with some 15,000 moderators. Privately, Facebook rarely disciplines its most abusive and more important superusers. Maybe because they have a vested interest financially not to. You might recall in October that whistleblower Frances Haugen testified before the Senate about how Facebook's algorithms put their bottom line above all else. Here's a clip.

Frances Haugen: I saw Facebook repeatedly encounter conflicts between its own profits and our safety. Facebook consistently resolved these conflicts in favor of its own profits. The result has been more division, more harm, more lies, more threads, and more combat. In some cases, this dangerous online talk has led to actual violence that harms and even kills people. This is not simply a matter of certain social media users being angry or unstable, or about one side being radicalized against the other. It is about Facebook choosing to grow at all costs, becoming an almost trillion-dollar company by buying its profits with our safety.

Brian Lehrer: Frances Haugen before the Senate in the fall. Joining us now to talk about Facebook and superusers is Nathaniel Lubin, fellow at the Digital Life Initiative at Cornell Tech. Some of you know he was former director of the Office of Digital Strategy at the White House under President Barack Obama. He's one of the authors, along with Matthew Hinman and Trevor Davis, of this study in The Atlantic article called Facebook Has a Superuser-Supremacy Problem. For full disclosure, though I'm not sure we've ever met as adults, we are related. If I think I've got this right, Nathaniel is my father's brother's wife's nephew's son, I think. Nathaniel, hello. Welcome to WNYC.

Nathaniel Lubin: Hi, thanks so much for having me. I hope you have that right on that relationship.

Brian Lehrer: We'll talk after this segment and try to figure it out. By way of background, what exactly was your role for President Obama?

Nathaniel Lubin: In the second term between 2013 and 2015, I was helping run an office called the Office of Digital Strategy, which was the team that put out information online and tried to communicate with folks using digital tools and tactics to get his message out.

Brian Lehrer: Was Facebook on your radar screen back then as a purveyor of hate or a problem in any way or is this mostly a development since 2016?

Nathaniel Lubin: I think these problems have been around for a long time. We definitely experienced that. I worked on his campaigns before that and was running the digital marketing team for a second campaign and we were experiencing a lot of the same questions. Certainly, it's become a much bigger focus and much more attention, particularly since 2016.

Brian Lehrer: The opening of your article, and we'll get to the data because you did a study. You didn't just write a magazine article. You did an actual data crunch and came up with some new numbers. The opening of the piece characterizes Facebook as a cesspool of hate and disinformation and highlights three Facebook superusers in particular; John, Michelle, and Calvin. You want to tell us about one of those folks? Pick anyone.

Nathaniel Lubin: The idea of this work was to look at misinformation and abuse that was happening on the platform generally. We all have an idea about the experience of this. Most people we know are on these platforms. We wanted to figure out a way to see how some of the bad users were actually acting out. We conducted a study, as you mentioned, that enabled us to look at who the most frequently engaged users on these platforms were. What we found, perhaps unsurprisingly, is that those people tended to be some of the folks that we've all seen likely in our feeds, in the comments, and elsewhere who are very often committing threatening violence, insulting other people using slurs of various kinds.

You referenced three of them. They tend to be-- One of them is an upstate person who is the worst of the worst, is a good way to characterize it right there. They're very often taking the most strident language, oftentimes in racist terms and sexist terms, and homophobic terms. When we were doing this work, a lot of it was the early stages of COVID, so a lot of it relates to claims around the pandemic, that kind of thing that was misinformation around public health as well.

Brian Lehrer: What's the definition of a superuser? Is that a term that you coined, a term that Facebook uses officially? What's a Superuser? Of course, the premise of your piece is the problem of superuser supremacy.

Nathaniel Lubin: That's our language here with The Atlantic. The idea of a superuser in technology terms is not new by any means, but in this context, what we found is that around 1% of users are accounting for about half of all the public activity on these public pages and groups. That means the comments, the likes, the shares, that kind of material, and so they are driving a lot of the activity on these different groups and pages that are getting so much attention on our public discourse. Then when we actually look at who those users are in terms of what they're saying, that's how we get the assessment of their levels of abuse.

Brian Lehrer: In terms of their levels of abuse, I'm just going to put a little more color on your answer about those three particular superusers who you start out with from the beginning of your article.

It says, "John, a caps-lock devotee from upstate New York, calls House Speaker Nancy Pelosi PIGLOSI', uses a certain N-word to refer to Black people, and says that the right response to Democrats with whom they disagree is to shoot all of them. Michelle rails against the pandemic. Calvin uses gay as a slur and declares that Black neighborhoods are always assholes." Then you say you've almost certainly encountered people like these on the internet. What you might not realize, though, is just how powerful they are. How powerful are they on Facebook?

Nathaniel Lubin: The question we were asking was about abuse. What we found is that because of their share of activity is so high, we can look at what that means in the context of the various disclosures Facebook has put out over the last few years, particularly the Frances Haugen disclosures that you mentioned in the intro to this. We don't know exactly how the algorithms work. Facebook hasn't disclosed that, but we do know that these various terms prize engagement overall. That Facebook has, in various ways over the years, prioritized engagement over everything else. That's what their business is organized around.

What we find is that, at least on these public pages and groups, these users are the primary drivers of that engagement. The specifics of the mechanism would imply that those folks are having a huge share of the say over who is actually seeing what in these algorithms. These mechanisms are for Facebook to explain, but that's the clear pattern that we're seeing.

One thing that we see about that is because they're so skewed in terms of the amount of engagement, oftentimes the descriptions of this talk about changes that they've made, small tweaks to the algorithms that change the ordering of certain types of engagement, or changing certain things. What we find is that because it's so pervasive across all these different types of engagement, that's effectively reordering among the same set of users.

Brian Lehrer: You have the stat that I cited in the intro, the top 1% of Facebook's users in the US tend to be responsible for about 45% of all the content reaching the eyes and ears of other Facebook users. You take a deep dive. As I said, this is a data analysis. This is a study that you did, you didn't just interview people for an article. You took a deep dive into Facebook's algorithm called MSI, or Meaningful Social Interaction. Tell us how they define Meaningful Social Interaction and how that winds up elevating superusers and disinformation and hate.

Nathaniel Lubin: Yes. Facebook's talked about MSI in various different ways, and it's the primary driver. They have said to us, since the article came out, that we've mischaracterized it, in their words, in certain ways. The key part of it is that it is driving the ranking systems in some ways and that they're trying to predict what users are going to do in response to what they see in their feed. All the disclosures in the Haugen testimony and the rest of the disclosures that come out in New York Times, The Washington Post, Atlantic previously have illustrated that they are trying to prioritize how these engagement metrics contribute to that.

The primary drivers of these algorithms are things like comments, things like likes, things like shares that users are putting in. When you look at the actual behavior of these superusers, you see that on public pages and groups they are accounting for the lion's share of the high types of activity, particularly comments and shares that in the Haugen disclosures are described as the primary drivers of how these algorithms are working.

Brian Lehrer: Why does so much hate and misinformation rise to the top in terms of how widely disseminated it is through this algorithm? Does that say something more about human nature than about Facebook?

Nathaniel Lubin: Absolutely. I think there's a reality here that is the people who are most active on these platforms really are a concentrated minority. It's certainly not everybody. We look at these users. It's not the case that every single one of these superusers is abusive or giving bad things, but what we found is that about two-thirds of them are. That is a reflection of the actual behavior of people. It's not a decision that a company or anyone else made. The loudest voices tend to be the ones who are the most extreme.

Brian Lehrer: Listeners, we want to invite you to help us report from the front lines of Facebook world. 212-433-WNYC. You're hearing Nathaniel Lubin from the article that maybe you've even seen in The Atlantic. 212-433-9692. Try to put some of your stories, or we're inviting you to put your stories on this data. Who keeps coming up for you on your newsfeed? What are they posting about? Do you notice the same users, so-called superusers appearing over and over again, or certain trends that you see? 212-433-WNYC, 212-433-9692, or tweet @BrianLehrer.

Have you noticed that the algorithm changed over time, and how that affected what's in your newsfeed, or maybe it's changed how you interact with content on Facebook, or if you've left Facebook, what compelled you to leave? 212-433-WNYC for Nathaniel Lubin, fellow at the Digital Life Initiative at Cornell Tech and former director of the Office of Digital Strategy at the White House under President Barack Obama and, again, one of the authors of the study published in The Atlanta called Facebook Has a Superuser-Supremacy Problem.

Also, listeners, I'm interested to hear if you think in your even passive use of Facebook, you have a lot of hate speech, a lot of disinformation coming your way, and you're like, "Why is this in my newsfeed?" 212-433-WNYC. We'll start taking your calls in a minute, and we'll also talk about the connection that Nathaniel and his co-authors found between the superusers on Facebook and people allegedly involved in the January 6th insurrection so stay tuned.

[music]

Brian Lehrer: Brian Lehrer on WNYC as we continue with Nathaniel Lubin, co-author of Facebook Has a Superuser-Supremacy Problem in The Atlantic. What was the correlation in the data that you found between those involved in the January 6th insurrection and superusers?

Nathaniel Lubin: It's a great question. The data we're looking at was from 2020, but we looked back after that at how these users relate to that and we talked about it in the piece. The pattern of behavior of those users who show up in the January 6th events mirrors pretty closely what the superusers that we're talking about in this piece are saying and doing. It is a subset of, it is a way to think about it.

It raises questions about what should Facebook be doing around prioritizing accountability to some of these users who are the frequent posters of abuse. Certainly, not everyone is going to be doing the kinds of things that happened on January 6th, but enough of them seem to be participating in that it does raise questions about that.

Brian Lehrer: The piece says that the superusers tend to skew older, white, and male. What else do we know about the demographics of the users who engage the most on Facebook? It's funny because it started once upon a time as a thing for college students. Now it seems like the big influences on Facebook are old and white and male. Tell us about that if there's any more to put on that, and who Facebook's algorithm is elevating.

Nathaniel Lubin: That's a good characterization of it. When I first got on Facebook and I was of the age where that was one of the main interests in going to college was you could get a college email address and log on to Facebook for the first time, definitely that was the center of this course. The kinds of users we're seeing here are not the unique users who are on the platform. Generally, I think most everyone we know has an account on these platforms, spends time on these accounts, in these platforms.

The ones who are showing up overwhelmingly in these superuser categories, the ones who are the most frequent commenters on these public pages and groups definitely are skewing older, definitely skewing whiter more conservative. Those are the folks who are right there. We're not looking at anything that's information that you couldn't see directly, so the stuff that they have described about themselves, we're not doing individualized investigations or anything here. The idea was to look at aggregate behavior and aggregate groups, but that definitely is the pattern that we see.

Brian Lehrer: Then Facebook's algorithm, I see from your piece certain weights are given to certain types of engagement in the algorithm. A like is worth one point if somebody likes a post, while reaction emojis are worth more than a like. Do these rates being public allow purposeful manipulation of the algorithm in which case users or even bots can then deliberately boost certain content over other content?

Nathaniel Lubin: Yes. Our research here was on the public activity side of this. We were looking at what behavior we actually saw. The numbers that you're referring to came out in the disclosures, and we're interpreting this through the lens of what we've seen. I don't think Facebook has deliberately revealed that kind of information before, but they have talked about in certain ways that they've changed it over time. The Atlantic had a story last year, for example, that talked about how they reduced the effect of the angry emoji and how that had effects on the kinds of content that was pushed into the algorithm more or less frequently based on some reporting inside of Facebook. That's one example of that.

Absolutely it is the case that various folks who are just trying to promote their own material or if they have nefarious bills, either way, are trying to leverage knowledge of these algorithms to make their material show up more frequently. It's certainly more complicated than just the raw number of counts that these different engagement post trigger, but we definitely see over and over again that those are the determinants. Those are factors that cause different things to rank higher and the more engagement you get, the likely you're to show up in feeds.

Brian Lehrer: Evanin Redhook, you're on WNYC. Hi, Evan.

Evanin Redhook: Hi, Brian. I'm a long-time listener. I work in-- I started my own social media marketing company a while ago. I just wanted to call out that it feels like really implicitly, these social media companies want everybody to become a social media user. On some basic level, a superuser. Basically, by posting more, that's really what we do with all of our clients is you just have to keep posting more and more.

I just think the demand on communication is so high. We deal with this with brands all the time where just the onus on communicating constantly with the public on all of these issues, it just becomes unclear why people need to comment on anything. I guess it's just that is really the formula that it's the more you post, it will eventually work. A lot of these people, even small accounts, their initial strategy is just post as much as possible, comment as much as possible. There really is no more a strategy than that.

Brian Lehrer: That's a marketing strategy, but then with your experience, Evan, do you have any theory as to why hate and disinformation get elevated as much as they do in Facebook and maybe other platforms?

Evan: I think we can all experience this just on a personal level when you see something that enrages you. I found myself doing it even not fully aware of it unnecessarily. Just you see something that annoys you, and then suddenly you're typing into the thing. It should really be that you write a letter and then throw it out, it doesn't need to necessarily get posted. There's some cathartic thing and you feel heard when you make that post, I guess.

Brian Lehrer: Evan, thank you very much. Somebody said, probably 15 years ago, maybe you know this quote, Nathaniel, the central organizing principle of the internet is why wasn't I consulted? As information democratized through various things online, if anybody expected calm, rational, civil discourse that's going to include more people than work at the major news networks, it has turned into such a prevalence of grievance. Did you expect something different when you started using the web when it was new?

Nathaniel Lubin: I got into this personally through I mentioned the Obama campaigns. I think, particularly at that point, we had a sense, and I think it was true, that these kinds of tools could be used to broaden the number of people participating, engage more people, get more people involved in activism. I think that's definitely a good thing and has borne out. The flip side of that, though, is without having restrictions on bad behavior that can keep up with the amount of bad behavior, we've definitely lifted the blocks that are out there. We expected to see problems on these networks when we looked at this kind of work, but this scale was a different pattern and different amount than we expected to see.

Brian Lehrer: You list a few of the top pages on Facebook; Ben Shapiro, right-wing commentator, Fox News, and the group Occupy Democrats. What does that tell you? Is the right just more prevalent on Facebook? That is, are there more people who are politically to the right on Facebook, or do they just get their stuff out more effectively?

Nathaniel Lubin: I think what we're seeing is that most users exist on these platforms, but the ones who are the loudest voices as users tend to skew more conservative. That's what we're seeing in this dataset in this analysis. We don't know exactly how that connects to the reasons why which pages are the largest to do the best, but it does beg the question that maybe there's a connection there between why some of these pages get so much attention and have grown much. It's not the case that all the top pages in this analysis were political pages, but they're definitely over-represented. That's a reality of the kinds of public discourse that you see on these platforms.

Brian Lehrer: Ben in Valley Stream, you're on WNYC. Hi, there.

Ben: Hi, how you doing? I'm calling because back when the pandemic began, I was posting just keeping my friends up to date with what was going on with the former governor and the state of affairs, and it became a centerpiece for a lot of dissent among my friends. When that happened, even I was like the previous caller just said, you just type something out, you don't even realize it, and then next thing you know you've posted it. I was engaging in that as well. When that happened, I just started to see how toxic Facebook had become and the type of people that subscribe to that. I think that it's become the type of place that has become--

It's very easy to offend people and people misinterpret things more often than they would in reality, or maybe on another social media platform. Because of that, you end up with, I think you were talking about this yesterday, Brian, or maybe last week, lots of people who have the same types of thoughts grouping together, and then you're all just bouncing the same exact ideas off of one another. That's basically what Facebook has become. I still use Facebook Messenger because it's super convenient. It's basically like texting. I have friends around the world and I can communicate with them still, but I've deactivated my account and I don't log on anymore. I feel better [crosstalk].

Brian Lehrer: Did you feel like you were getting political content that served as an echo chamber that just reinforce what you already thought, or were you also getting bombarded with things really different from you either in a good way or misinformation?

Ben: See, I want the alternate perspectives-

Brian Lehrer: Sure.

Ben: -because that's what's opinion and that's how I form my own opinions. I'm not forming my opinions based on public affairs or the state of the world just from NPR. Even though I don't agree with it, I'll check out Fox News because they are doing some form of reporting. I found that what I was getting bombarded with was just Democratic media, Democratic media. I may be a registered Democrat, and I don't know how Facebook knows that, but I don't necessarily agree with the Democratic Party. In my opinion, they basically do nothing.

Then for Facebook to see that and then bombard me with that to try to get me involved in those programs that they're running and try to get me involved in those communities, it was very forceful. You also start to see ad campaigns from people running for a local office and a state office of the party that you are a member of, or even maybe not a member of the party that they think you follow. It just became very decentralizing. It made me not even want to engage because I'm being told to do it, or I'm being told to engage in it.

Brian Lehrer: Right. I hear you.

Ben: Especially when they're not even following through.

Brian Lehrer: Ben, thank you much for your call. I really appreciate it. Nathaniel, lift the veil a little bit for a lot of our listeners. How did Facebook know Ben's politics?

Nathaniel Lubin: You'd obviously have to ask Facebook to get into the details for any particular user. Each user they say has a different response, which I'm sure is true. They're looking at the pattern of activity that he has done and users like him have done. If he's getting lots of political content about Democrats in his feet, it's likely because they think people like him are going to respond to that. On average, they probably are. That's the underlying mechanism.

I don't think it's true that they're trying to surface political messaging or misinformation. Certainly, deliberately, it's just that's a consequence of trying to maximize the amount of engagement that an individual user is looking at. If users are responding disproportionately to the more extreme voices, that is what drives more engagement, drives more content. That's the effect of it. In this case, he's probably either himself or the users like him responding to this material even if he doesn't like it.

Brian Lehrer: To wrap up, I'm going to play a clip of Mark Zuckerberg, and then I'm going to ask you if having done this data analysis of Facebook's superusers and how influential they are, and how much hate and misinformation they spread around the site, and around the world therefore-- I'm going to ask you if your article comes to any conclusions about what Facebook could or should do. Here's Zuckerberg talking with tech reporter Casey Newton last April talking about individuals getting power over ideas as a net positive.

Mark Zuckerberg: I think if you look at the grand arc here, what's really happening is individuals are getting more power, and more opportunity to create the lives and the jobs that they want and to connect with people they want, and to connect with the ideas that they want, to share the ideas that they want. I just think that that will lead to a better world. It will be different from the world that we had before.

I think it will be more diverse, I think more different ideas and models will be able to exist, and I think it inevitably means that some of the people who had control over that world in the past will lose it. I can see why those folks will lament the direction that it's going in. My concern is that we're too frequently telling the negative sides of it from the perspective of the institutions that may be not on the winning side of these changes, where I think the people who are on the winning side of these changes are individuals.

Brian Lehrer: Mark Zuckerberg last year. Obviously, that's going to strike most of the people listening right now as insanely idealistic about how these algorithms turn out or make discourse turnout. They did hire 15,000 moderators, at least said they were going to do that, to try to moderate content. Do you make policy recommendations for Facebook or for government with respect to Facebook in your article?

Nathaniel Lubin: In the article, we're less going to policy than description of what we're seeing. I think we do see that given how much this concentration is of abuse, Facebook hasn't described exactly how they triage the 15,000 plus reviewers that they have, but one recommendation that comes up directly from this work is that they very likely could focus more attention towards the top users from an engagement perspective and try to crack down abuse specifically among those more frequent.

We don't know exactly what they're doing or why, but one thing that we do see as we look back at the sample of superusers that the study was based on about 18 months later, and only about 5% of the worst abusers had been removed from the platform. I'm not saying that every single one of them should have been removed, but that number presumably should have been higher than 5%. Reprioritization towards the folks who are the most engaged and also the most abusive does seem like a thing that would be useful.

I think more generally, the future policy thoughts and work that's coming out of this is we're thinking about what are ways that you could restrict pure engagement as the prioritization of these algorithms. If they want to do that, that's fine, but there have to be counterweights built into restrictions that tighten more pro-social outcomes, and at least restrictions on things that reduce those [unintelligible 00:30:57].

Brian Lehrer: Nathaniel Lubin, fellow at the Digital Life Initiative at Cornell Tech and former Director of the Office of Digital Strategy at the White House under President Barack Obama, and now a co-author of The Atlantic Magazine article, Facebook Has a Superuser Supremacy Problem. Thank you so much for joining us. Really informative.

Nathaniel Lubin: Thanks so much.

Copyright © 2022 New York Public Radio. All rights reserved. Visit our website terms of use at www.wnyc.org for further information.

New York Public Radio transcripts are created on a rush deadline, often by contractors. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of New York Public Radio’s programming is the audio record.