The Facebook Dilemma | Interview Of Craig Silverman: BuzzFeed

The Facebook Dilemma | Interview Of Craig Silverman: BuzzFeed
The Facebook Dilemma | Interview Of Craig Silverman: BuzzFeed

Craig Silverman works as a media editor at Buzzfeed News. This is the transcript of an interview with Frontline’s James Jacoby conducted on April 20, 2018. It has been edited in parts for clarity and length.

Help me out with where a good place to start is, but in leading up to what it was, how you ended up doing the work at CJR [Columbia Journalism Review] and looking into lives. What prompted that work that you were doing?

I’ve been looking at verification and the spread of information for a while, and for a long time, I was really focused on how journalists verify things. You find a story; you find information. How do you figure out if it’s true or not? So I wrote about that and trained people in that, and then Facebook came along, and Twitter came along, and things started to really change. You started to see information that might have found its way into a newsroom and have people kind of look at it and verify it – by default it was now public. So we had this really different information environment where rumors were out there and getting tremendous traction and distribution, where information that a journalist might have time to work on, they don’t have time anymore. It’s already out there. It’s getting shared; it’s getting engagement.

For me, the research project was really trying to grapple with how are journalists dealing with this very new environment where unverified information is out there and spreading. How are they dealing with false information spreading? Are there best practices in newsrooms for how you deal with that? The research project was just trying to dig in and get a sense of what kind of stuff is spreading and how the news organizations cover it.

So we’re talking about what year now? It’s like 2013, 2014?

OK. In about 2013, I started working on a project for the European Journalism Center, a book called the Verification Handbook, where we got some of the best experts from around the world, journalists and other people, who really focus on, how do you verify that tweet; how do you verify that video? We had some people writing about stuff that happened in Ukraine, stuff that was going on in different parts of the world. That was a big spark for me to dive even further about rumors spreading online, misinformation spreading online.

Facebook’s Move Into News

And how big a problem was it at that point in time?

I think in 2013 we saw some of the seeds of what would really take off more. A big thing about that timeframe is the 2012 election, the talk was really about Twitter and news. Twitter was the place that news was spreading. We had the Arab Spring, and Twitter was the focus, and I think Facebook got a little bit jealous. After 2012, as we get into 2013, 2014, 2015, Facebook starts putting a lot of effort into getting news onto its platform and showing that it’s a place for news.

In 2013, when we’re working on the Verification Handbook, yeah, we talk about Facebook. We talk about Twitter, but Facebook isn’t necessarily the news behemoth yet. It’s in the early stages of really starting to take over as the major distributor of traffic on the internet.

Do you really attribute Facebook’s move into news as a competition thing with Twitter? I mean …is that really what you think the story was there?

I really think that after the 2012 election, Facebook felt like they weren’t playing in that world as much as they could. They had far more users than Twitter. They had a better business than Twitter. They were a more mature platform than Twitter, and yet everyone was kind of obsessing with Twitter. So I think there was a little bit of jealousy, but I think there was also probably a good business reason as well to look at this. People read a lot of news; people share a lot of news. If you, as Facebook, want people to spend more time on your platform, getting news there is probably a good idea for that. I think there were good reasons for them to want to increase that presence.

In terms of watching and then that first project, how big a deal were, you know, for instance, rumors spreading on Facebook? Was there any indication at that point that Facebook was looking into this as being a problem?

I don’t think anyone in 2013 saw Facebook as the ground zero of rumors and misinformation spreading online. It was one of a bunch of different places. I think Twitter was actually the place.

At that point I think Twitter was the big focus for a lot of misinformation, because it was where news was breaking. It was where in the early moments after an earthquake, a natural disaster, any kind of big news event, that’s where it was. So there was probably more focus on Twitter at that time than there was on Facebook.

Tracking Misinformation Online

Describe the research that you were doing at that point in time and what you were finding.

Starting in late summer of 2014, we created a tool and a database to basically track rumors that were spreading online. Myself and a research assistant, we would look at a claim that was starting to circulate – maybe it started with a tweet; maybe it was in a news article – and we would capture that in a database, describe what it is, and then go out and hunt for all of the news articles that were written about that. The idea was to see when something unverified gets out there, how do news organizations cover it? How do they talk about it? What headlines do they use? Over the course of several months we gathered data on hundreds of different kinds of claims, and over time we would see whether they turned out to be true or they turned out to be false.

We also captured the social data so we would see how many tweets a particular piece of content got, and we would also look at how many engagements it got on Facebook. One of the things that started to really become clear really early on in the data is that at the early stage of something, that’s when there’s the most interest, and also if something turns out to be false, it typically got way more attention than any attempt to actually go out and correct it, which was a really big concern as we started to identify that.

So where were you publishing this? Where was this?

The emerging research project was part of a fellowship at the Tow Center for Digital Journalism at Columbia University.

And was there any awareness at the time by Facebook, for instance, about there being an issue with the virality of lies or misinformation?

I don’t think that Facebook was really concerned or talking about or thinking about the virality of lies and misinformation at that point in time. I don’t think it was something that they were being pushed on or confronted with to a huge extent. Clearly it wasn’t something that was part of their product discussions at that time either in terms of what they were actually rolling out.

So what was it like when you actually published? Was there a response from Facebook?

No, Facebook never responded to the paper itself. But late in 2014, I did have a phone call with someone who worked at Facebook, someone who I knew outside of Facebook as well, and they were just kind of interested, and they said, “What is this project you’re doing?” The thing that really stands out to me about it is that this person, they were really interested in finding out how much of what we were doing was automated. The truth is that we had some automation. We could automatically determine how many shares something was getting, and we could track that over time automatically, but we had to go out and identify these rumors and these claims ourselves, and we had to go out and find the articles ourselves. And when I described that manual processes to this person who worked at Facebook, it was like through the phone I could just sort of sense their interest completely dissipating, because when you are as big as Facebook is, even at that time – they didn’t have more than 2 billion users then, but they were still really big – if you can’t automate it, I think from their point of view, if we can’t get engineers to do this, we’re not interested in doing it. And that kind of bias toward the automation, toward artificial intelligence over human decision making and human labor I think is at the core of Facebook’s origin story as an engineers’ company.

And why is that problematic?

When you have a bias toward anything, on one level it’s problematic. Specifically to Facebook, what becomes problematic is when you are so big and you’re relying on automated systems to help manage that, to help manage what content should be promoted, what content should be taken down, when you hand that over to automated processes, you really allow a lot of things to happen without any kind of oversight. And sure, they run tests. And they kind of check the code, and they see what’s going on. But to actually have a handle on what’s happening on your platform I think becomes really difficult, so you have people who just, you know, if I don’t see something break in terms of code, if I don’t see a product go down, I think it’s running fine. If you’re actually looking at stuff with human eyes, if you’re kind of evaluating, “Well, what’s really performing well on our platform?,” and digging in on that level, then you start to get a different picture, a qualitative picture that’s a little bit different than the quantitative thing.

Virality Of Misinformation

What sorts of examples came up in that initial research of things that you were watching go viral, things that were spreading? What examples do you remember from that?

One example that we saw of a completely false story that went huge and was really concerning at the time was a story – in the fall of 2014, one of the big concerns in the United States was Ebola, a concern that it would take hold and start to spread in the United States. What we saw were some articles that were created that were completely false claiming that Ebola had come to the United States to a great degree. There was one story from a website called National Report that claimed an entire town in Texas had been quarantined, and you could imagine what the reaction is from people reading that. What we saw was that the original story, 100 percent false, got roughly about 350,000 shares, likes, comments on Facebook. We saw about five attempts from news organizations, often local ones in Texas, to try and debunk it, and they got I think it was roughly about a fifth or a sixth of that engagement. That example really played out how what we were seeing of the false claim got way more attention, got so much traction on Facebook, and then the attempts to debunk it, even when there were multiple attempts by somewhat large news organizations, they just couldn’t keep up. They couldn’t overtake it

Was there a mechanism at that point in time to be able to flag content that was either suspicious or untrue or anything like that? Facebook had a basic flagging ability at that time, something as spam or what have you, but I don’t believe there was a specific category for something that was false. I don’t think it was something that they were particularly worried about on their platform. It wasn’t something that they were really building for in terms of dealing with. And we saw it again and again, where you would have 100 percent false stories, so not like somebody got something a little bit wrong or there was a misleading headline – 100 percent false stories created to deceive, engineered for Facebook, continually taking off. And some of them were really kind of serious ones, like the Ebola one. You’d have other ones with strange, completely false biblical prophecies. One of the big hits from 2014 that I remember was a claim that there would be a day where there would just be complete darkness throughout the earth, and that got just hundreds and hundreds of thousands of engagements on Facebook.

I started to see at this point in 2014 that there were people who were basically building businesses off of this. They ran multiple sites that were all false, that all presented themselves like real news websites, and they just made everything up because it just seemed a lot easier I think for them to make money that way. They put their stuff out on Facebook, and it just caught fire.

Making Money Off Misinformation

Who were the people at that point in time? We obviously know about the Russians later, but who were the people at that point in time who were pointing out fake content and for what reason?

The people that I encountered initially who were doing what I sort of considered to be the pure fake news were typically Americans or some Canadians, and they basically saw themselves as entrepreneurs in a lot of ways.

They were people who realized that if you put ads on these completely fake websites and you got a lot of traffic from Facebook, that was a good way to make money. So they were publishers; they were entrepreneurs. In some cases, they were just very straightforward saying, “I know it’s not true, but it’s satire,” and in other cases they tried to make the argument to me that, “Well, yes, I have this website, and it looks like a conservative web site, but it’s all fake.” And their argument was that they were exposing the supposed gullibility of conservatives. So sometimes they contorted themselves to give these examples that “Yes, I’m making fake content, and I’m spreading hoaxes, and I’m making money from it, but I really have a good goal here.”

And they were making money from it by ad sales on their websites?

Yeah. So you would come to one of these fake articles, and throughout the page surrounding the story itself, you would see ads. The more people that came to that page, the more ads that were loaded, and if someone clicked on an ad, they would also make additional money on top of that.

What were you thinking at the time about what you were discovering? Because this was the front edge of obviously something that’s become major. Also just tell me if it’s untrue, but you’re credited with coming up with this concept of fake news. Was that what you were calling it at the time? Tell me what kind of – bring me into your head.

My reaction was, I was actually really pissed off, to be honest. I was seeing people who were just spreading misinformation for money and were really unrepentant about it, and it was performing extremely well. I was looking at the efforts to debunk this stuff and seeing that it was being completely ineffective in a lot of cases, so it was really frustrating to me. It was really concerning to me. So I started really talking about this a lot on Twitter and talking about it a lot in the media, saying look, we have this problem here where the completely fake stuff is outperforming the real stuff on Facebook.

That was when I really saw the power of what Facebook had in this area and that the incentives seemed to be giving the people doing the fake stuff the advantage. So in the fall of 2014, I just started using the term “fake news” to describe this stuff, and it just seemed like a logical term to me. It’s a site that looks like a real news website, it’s an article that’s written in the journalistic style, but it’s all fake, so I started calling it fake news. When I ended up writing my research paper, I had a section about fake news specifically.

It’s not like I sat down and said, “All right, I need the perfect term to describe this.” It’s just sort of what came naturally.

Coining The Term “Fake News”

… So in terms of it coming naturally, basically you have a chapter that was entitled – you just started talking about fake news.

“Fake news” just seemed like the right term to use.

I didn’t think about its place in history or anything like that. It was just sort of like, “This seems like a good way to help people understand what I was seeing.” I was trying to get people to pay attention. I was trying to get journalists to pay attention. I was trying to also get Facebook and other companies like Twitter to pay attention to this as well.

And what sort of reception did you get from the companies like Facebook, for instance?

On a positive level, someone from Facebook reached out and wanted to talk about the project I was doing. The negative thing was that when they saw that it wasn’t really something that could be automated, something that could be integrated into their systems that was kind of the last I heard from them. But the one thing that I did do for them was I had at that point started building a list of these completely fake websites, and I sent that along to Facebook, and I said:  “You should look at these; you should be aware of them. You should think about how you’re going to handle them. I think at that time I really suggested that I don’t really understand why these sites should be treated the same as real news sites on your platform. There should be some kind of way of distinguishing them. And at a certain point, Facebook did actually try something. They rolled out a satire label on content from places like The Onion and some of these other sites so that when you shared a link from them, it would show up in people’s News Feeds with the word “satire” with some brackets around it.

They tested that for a while, and then they got rid of it. Really that was the last I saw of any kind of overt effort in this area.

… It’s all well and good to flag satire, but you’re finding a lot more than just satire, right?

Yeah. Satire is one of these things where different people have different definitions of it, and that’s fair. But when someone is just completely making up hoax stories about a quarantine in Texas for Ebola, there’s no satire there. There’s a public panic that takes place, so I think one of the things that we’ve seen over time is that people take that “satire” label; they wrap themselves in it to kind of avoid accountability. Labeling something as satire that’s actually meant to deceive people is probably not a good solution.

Legitimizing Misinformation

And you mentioned this idea that there is this almost level playing field between the legitimate news organizations and the news content on Facebook and fake content. Go into a little bit of depth about that in terms of that in your process of discovery, was that something you were becoming increasingly aware [of] and concerned by?

One of the profound changes that we’ve seen with the rise of a platform like Twitter or Facebook is that it creates this kind of flatness, this level playing field. So The New York Times can have a story that gets shared on Facebook, and it’s going to show up; there’s going to be a headline and a photo. Well, if it’s a teenager in Macedonia who’s created a story or a fake news publisher who’s created a story, it will also show up with a headline and a thumbnail. There isn’t a lot that distinguishes these kinds of things when you encounter them on the platforms, and when you create that perceived kind of level playing field, what you’re actually doing is advantaging people who have no reputation, who have no credibility, whose content can look the same as places that are actually putting a lot of effort into what they’re producing. When that happens, the dynamic is that the people who manipulate, the people who actually appeal to emotion, they can have an advantage over folks who are actually trying to give just kind of straightforward information.

What do you mean by emotion and appealing to emotion?

One of the clear ways to succeed with fake news or really hyperpartisan news is to go at people and get an emotional reaction. And it’s not just with news. I mean, any kind of content on these platforms that makes somebody really happy or angry or very sad tends to get them to do something, and that can be sharing; that can be liking; that can be commenting. Over time, I think what a lot of publishers realize, both ethical and unethical ones, is that if we can think about what the reaction is we want from people in this headline, in this thumbnail, in this combination, and the more that we can get an emotional reaction, the better off it’s going to perform.

So this is where things start to get a bit skewed, where the incentives of the platform lead people to pushing to extremes – extreme happiness, extreme sadness. It’s death to be in the middle of just reporting something plainly. You need to get somebody to have a reaction to your headline and to your image, because then they might do something, and if that human does something, if they like it or share it, then the algorithms see that people are engaging with it, and they show it to more people. So it sets off this kind of chain reaction.

… Facebook at the time was saying that it was a neutral platform. Was it actually a neutral platform?

Neutral platform: This is like the mantra of all of these companies. They all really wrap themselves in this idea that, you know, we’re not favoring one thing over another. What people like is what people like. It’s a cop out at the end of the day, and it doesn’t acknowledge the role that they play in one, creating a platform; two, setting the rules on it; and three, creating the systems that enforce those rules. So to imply that because an algorithm is deciding what to show to somebody, or not to imply that that doesn’t have bias, is completely untrue.

This whole idea that they are unbiased platforms, they are platforms for speech. Twitter described itself as the “free speech wing of the free speech party.” That all sounded great when the Arab Spring happened. That all sounded great when people weren’t in a massive way abusing and manipulating these platforms. But what they did was they opened the door for people to say: “OK, it’s a level platform, and I’m just going to appeal to the biases of your algorithms. I’m going to appeal to human bias, and I’m going to have stuff run like wildfire on it as a result.” So no, they were never level playing fields. They always set the rules.

They always built systems that favored some things over others, and as a result of that, when people figured out the rules and figured out what really worked and what got the most distribution, those were the ones that won. It didn’t matter if it was accurate. It didn’t matter if it was real. It didn’t matter if someone had spent a lot of time on something versus no time on something. What mattered was who played the game better on these kind of platforms.

And this was while Facebook was becoming the predominant news distributor globally?

Yeah. In the period between like 2012 and 2014, 2015 Facebook becomes, even more than Google, the dominant traffic driver to news sites, particularly in the United States and a lot of English-speaking countries. Facebook becomes just like the dominant player in news, and what happens is, when it becomes a place where the traffic is coming from, the source just spraying traffic out to news sites, then everybody starts to orient toward Facebook.

We see early on organizations like Upworthy that were really engineered to create content that spread on Facebook. We see BuzzFeed as well, not a news company early on, trying to figure out what people wanted to share. And as Facebook gives out more and more traffic, more and more people start to focus on Facebook, and at a certain point it just becomes an obsession for everyone in the news industry about how they’re performing on Facebook, how much traffic are they getting from Facebook, how do they get more. And that becomes something that happens really on a global level.

Facebook’s Reaction To Warnings

At the time – and correct me if I’m wrong – but at the time, wasn’t Facebook working on an internal study about the “cascading rumor” effect?

Right.

Can you tell me a bit about that?

Do you remember what year that was published?

I think that was 2014.

It was 2014. OK, yeah. One indication that we have that Facebook was aware of how information was spreading was their internal team managed to publish a public paper about “Rumor Cascades,” and it gave us an idea of kind of the flow of information on the platform. They also published a paper that looked at what happens when a discussion gets “Snoped,” is what they call it. People are having a discussion, maybe a link has been shared, and then somebody comes into the discussion and adds as a comment a link to the Snopes website. Regardless of whether that Snopes link was saying that something was true or something was false, what they found was that it actually kind of killed the discussion. When somebody came in and said, “Well, actually this is true,” or, “Actually this is false,” nobody really wanted to talk about it anymore.

I’m not sure if we understand what Snopes is, so are you able to tell me that in a way that’s less specific to that? Yeah. … First of all, did the Facebook study find that there was actually a problem with cascading rumors on their platform?

I think the study certainly quantified that, that this was a really powerful platform for seeing information go tremendously viral. And there was also I think an element there where you could see that whether it was true or whether it was false, it was really spreading regardless of the accuracy of the information. At the core of it, that’s the thing. When algorithms are making a lot of decisions, it’s based on the interactions with it; it’s not based on the quality of the content. Really, at the end of the day, it’s about what is getting humans to react with it, and they saw that stuff, whether true or false, was definitely cascading, causing a lot of conversation, causing a lot of shares.

And was there any indication after they released their own study or after you release your work that they tried to address the problem in any way?

They didn’t really talk about the issue of misinformation on their platform in 2015. And even that “Rumor Cascades” paper, it was released, but there wasn’t a big deal made of it, and there wasn’t follow-up research either. So it came out, and it was nice to see that they had thought about this internally, but there wasn’t a larger product strategy, and executives of Facebook didn’t really want to talk about that. I think overall it’s fair to say they didn’t consider it to be an urgent issue, a big issue. They didn’t consider it to be something that they needed to put resources behind.

Going back to your conversation about automation, right, with the Facebook rep that you were speaking to … first of all, are you able to tell us who that was at Facebook that you were talking to?

No.

OK. … What did it mean when that Facebook representative is telling you, “Well, we’re probably not all that interested if we can’t automate this thing”?

Look, in fairness, at the time, I was interested in taking that research product I was doing and seeing if we could build a sustainable product or a sustainable company or nonprofit around it, so for me a conversation with Facebook at that time was really interesting to get a sense of are we building something that might actually be able to be used by them in some way. The disappointment in that conversation was twofold. It was one, they just don’t seem to be interested in anything that actually takes human effort, and this problem is a complex problem; they wanted a simple solution. And two, of course it told me that well if we were going to make this thing sustainable, we should think about doing it outside of any kind of relationship with Facebook.

And why was it so laborious? Why couldn’t it be automated, for instance?

Even to this day, automatic detection of rumors and automatic detection of true versus false is still really hard. There’s a lot of researchers who have been working on it. Even in 2014, there was a decent body of research that had happened about identifying rumors on Twitter, but nobody had built a system that was really that accurate.

I mean, it’s kind of funny. I think one of the reasons why research into rumors and misinformation wasn’t really widespread in 2013 or 2014 is that a lot of the best researchers and a lot of the best data were inside these companies Twitter and Facebook, so they might be looking at stuff, but we wouldn’t know publicly. A lot of the best data to actually do this research and run these kinds of classifiers and other tools you would be building, I mean, the best opportunity was to do that inside these platforms, and they weren’t really doing it. We had this scenario where a lot of the best talent gets hoovered up by Twitter and Facebook and Google. The people who also have access to the most interesting and best data in these areas work at those companies as well, so when the companies decide it’s not a priority for them, you’ve got some of the best talent and all the best data not being used for these kinds of things.

Were you applying pressure at the time to actually try to move them toward studying this and addressing it?

I was agitating publicly for sure. As I started to encounter stuff during the research process, I was trying to make noise on Twitter. I was trying to get first other people in the media to pay attention to this and go out there and debunk things, but I was also hoping that yeah, this kind of pressure and this awareness of it might make them look and say, “OK, what are things that we can do?” So there was yeah, there was definitely an effort on my part to try and get some attention around it.

Facebook And Ukraine

… Tell me how Ukraine played in at the time in your purview. Where were you when you were learning about it, and what was it that you were curious about in gleaning from [it]?

For me, when we started working on the Verification Handbook, we were dealing with people who in some cases were working for like Amnesty International or other NGOs that were going into war zones or where war crimes or human rights violations have been committed, and we were seeing how they were using these tools and techniques to verify what really happened. Through that process of talking to people at NGOs and these kinds of researchers, I started to become aware that Ukraine was a really big focus for everyone.

The Russian incursion that had happened was really interesting, because you had things like Russian soldiers on the frontlines and the government. The Russian government was saying there are no Russian soldiers in Ukraine, but you had Russian soldiers on the frontlines posting to Instagram and geotargeting their post so you could see exactly where they were in Ukraine, and there were researchers out there who were using this material, this so-called open source material to really prove that this government was lying.

At a certain point we also started to really hear a lot of reports about websites being set up in Ukrainian or in Russian that were spreading completely false stories, and there would be networks of websites, and there would be networks of Twitter accounts. For me, Ukraine was really eye-opening in terms of a state-sponsored propaganda effort that really prioritized information warfare online. I still think to this day Ukraine is a really important moment in the history of online propaganda, because we saw the techniques that Russia was doing then, and they just continue to evolve. But that was a state–on-state attack, and we hadn’t really seen such a clear information warfare effort.

… So what was going on in Ukraine prompting you to further your research and to delve into something more deeply? Tell me how that affected –

I think watching what was happening in Ukraine made me feel like this was a really urgent global issue. It’s really easy to just be focused in English language like I was for my research, but then you would look and see what was going on in Ukraine. It was a literal war zone, so this really was information warfare, and the stakes of that were incredibly high, admittedly much higher than kind of a silly hoax that was spreading widely on Facebook. But you could connect the dots between the two and realize that the techniques could transfer and the approach could transfer and that all of the foundation was there for extremely viral misinformation.

Facebook’s Response To The 2016 Election

… Going back to Facebook’s published this thing about cascading rumors, and there are people like Renee DiResta at the time who were talking about other types of rumors going viral on platforms. But when do you think it actually gets on their radar screen that this is a problem?

I don’t think Facebook looks at it as a problem until a few weeks after the 2016 election. I don’t think they were really focused on it or caring about it to any extent until they started to feel like there was a threat to their company and their business based on the conversation that started happening.

… Was this one of those experiences where you’re just kind of seeing a phenomenon that you know is really important and really global, and no one’s paying attention, and you’re just kind of tearing your hair out as a result of it?

I think there were a lot of frustrating moments where I was looking at stuff, and I just saw these pieces falling into place. I was seeing completely false stuff getting insane engagement, and efforts to debunk it and knock it down were basically ineffective. We saw conflicts around the world where a different type of online propaganda was really being weaponized and used, and it felt like you were kind of shouting into the dark to a certain extent. There were journalists who cared about this; there’s no question about that. But society as a whole wasn’t thinking about it, wasn’t caring about it. It wasn’t a part of the conversation.

And the companies themselves, like Facebook?

You didn’t get any sense that Facebook and Twitter were really looking at this in any meaningful way. For them, I think at the time they saw it as this is a small percent of what’s on our platform; we’re an open place; we care about free speech; we’re not going to take any action in this area.

Part of the thing is that you’re saying, and going back to the equal-playing-field thing, is that this is baked in, right? In some way that – I mean, you’ve used the word “rigged,” right, in some of your writing. Can you explain that and explain it to me in a way where I take it you’re sort of discovering that as you’re going into it, that this is like actually how these platforms work and what actually works on there.

Right. Watching a fake story about a completely false quarantine in Texas really take off and very quickly get a lot of velocity and watching those kinds of sites just pump stuff out, it really became clear that they had kind of cracked the code. What they realized is that our stuff looks like everybody else’s stuff on this platform, looks like a real news story except they have the advantage where they can make anything up. They could take a breaking news event and create the worst-case scenario for it and write a story about that. What I think at the core is that they realized is that you had this collision between human nature and human bias and these machines, these algorithms that have been built to favor whatever the human seemed to engage with the most.

If you triggered the humans and they did stuff, then that triggered the algorithms, and then it was just like this this rock rolling down a mountain picking up steam. I think they just figured out that the system didn’t really care about quality or accuracy. It cared about what was getting people to spend time on it and engage with it. And if you feed people stuff that appeals to their bias, if you feed people stuff that gets them an emotional reaction, then you’ve basically – you trick the system into thinking that this is something I need to show more people.

Facebook And The Distribution Of News

… What do you think of Facebook’s move into news as sort of the seminal thing that brought about this kind of age of misinformation?

Facebook’s big priority on news I think is a huge and major, major step in creating an environment where misinformation can go far more viral than ever before. I think it’s a really important moment, because we never before in the history of humanity have we had a platform with that many people on it in that many countries that was also in many cases available in your pocket and also a place where there was news and your friends and your family and events. It’s just – it’s unheard of. So when you mix that cocktail together, chances are something new is probably going to come as a result.

When legitimate news organizations go on Facebook, like The New York Times or like Frontline, were we legitimizing that as a place for people to get their news?

Yeah, I think the fact that so many big, well-known news brands really pushed into Facebook pretty aggressively legitimized it as a place to get information, and I think that also strangely created the opportunity for people who weren’t legitimate as well, because if the legitimate players are there, and you’re not legitimate but your stuff actually looks kind of similar to theirs, then you’ve just won.

You’ve just gotten a huge advantage, whereas before you would have needed to have newspaper boxes on the street, maybe a satellite dish or these kinds of things, but now all you need to do is set up a website and then share links to it, and your stuff on Facebook is going to look similar enough that you’ve just gotten a huge leg up.

… Do you think that users were able to discern between real content, fake content, ads and news content? Was there any way to distinguish between them?

There’s a weird dynamic that happens where the average person thinks they are really good at spotting fake things, thinks they’re very media-savvy. But when you actually test them – and lots of researchers have done this – most of us perform terribly. We did a poll after the 2016 election where we showed people real fake news headlines that had been spread on Facebook, and we found that in a lot of cases, when people were familiar with them, they were more inclined to believe that they were true.

That’s not completely surprising. There’s often a connection between having seen something and being familiar with something and thinking that it’s real or it’s credible. But there’s also been a lot of research about people being able to tell the difference between real content and sponsored content, and they often can’t. When you have an environment where there’s so many different types of content all intermingling and all in the same basic kind of design bucket, it’s going to create tremendous confusion for sure.

And Facebook knew this?

It’s hard for me to imagine that if Facebook had studied whether people can distinguish between this or that that they wouldn’t have found that there was a lot of confusion there. And it’s not 100 percent unique to Facebook. We’ve always found that sponsored content is hard for people to identify, but Facebook was a new kind of combination of so many different types of things. The idea that somehow humans would innately know the difference on Facebook doesn’t seem to make sense to me.

The Algorithm And The News Feed

Can you also just explain really quickly that it used to be, for instance, that in your News Feed, it was as it came, and then the algorithm introduced something else. Can you just explain that evolution kind of quickly?

One of the big differences and changes for Facebook was when they started to use an algorithm to decide what people should see in their News Feed. Basically what this means is that rather than you seeing the thing that is most recently posted by your friends or the pages you like, Facebook has created a program that weighs certain things and downgrades certain things to decide what they think you most want to see and what they think you most need to see based on your past behavior.

It’s basically a set of instructions that decides the order of what you see in the News Feed, and this is a radical change, because all of a sudden, it doesn’t matter whether somebody just posted something five minutes ago; you may never see that, because the algorithm may decide that that’s not important. What also becomes really big about this is that suddenly there’s kind of a key to the Facebook kingdom, because if you can figure out what the algorithm is more likely to show, then all of a sudden you’re going to have your content seen by more people. So the algorithm becomes an insane obsession for people in media, for people in advertising in every part of the world who’s trying to figure out, how do I get more attention on Facebook?

Misinformation In The 2016 Election

Great. That was really helpful. So getting to the election, bring me into the genesis of how you start looking into what you were eventually publishing in November.

In about the spring of 2016, the editor in chief of BuzzFeed comes to me and says: “All right, it’s time for you to stop running Canada. It’s time for you to go back to your misinformation beat.” And he didn’t say, “We’ve got this crazy election going on; we need you to look at it,” but there was a sense that this was an important issue. Facebook had grown bigger; the election was generating a huge amount of attention. So over the summer of 2016 I start really digging back into this world, and one of the things that I really want to do is I want to do a story and figure out if the kind of Russian activity we’d seen in Ukraine before was happening in the U.S. election.

I had a feeling that there would be some kind of Russian intervention, because it was just too big of an opportunity for them to not want to play in. In that summer, I talked to a researcher named Lawrence Alexander who had done some good work identifying Russian propaganda campaigns, and we started hunting basically. One of the things we looked for was pro-Trump outfits that existed outside of the United States. We were interested, you know, would they create things in other countries around the world either trying to whip up pro-Trump sentiment there or even have that flowed over to the U.S.?

In the midst of that research, we started to see this small cluster of sites in Macedonia. We saw it in Italy; we saw it in a few countries, but the Macedonian one was a little bit puzzling. We saw that it seemed to be people who were really just trying to make money from pro-Trump information. It wasn’t a case where we could call them up and they would say, “Yes, I’m a big Trump supporter.” So we decided to dig in more in these sites in Macedonia, and we ended up finding roughly about 150 pro-Trump websites being run, the vast majority from one town in Macedonia. We also found that a Macedonian publication – and The Guardian had talked about this cluster before, … but what we decided to do was to examine the content that they were publishing. We wanted to understand what were these stories and where did they come from. And very quickly, in looking at the most shared stuff on Facebook, four of the top five were completely false stories, and they were all really either pro-Trump or anti-Hillary Clinton stories.

This is what we saw across all of these Macedonian sites. We saw that they were copying and pasting completely false stories oftentimes from fake news sites in the U.S. They were taking real news stories and oftentimes putting a false headline on top of them, and they were basically almost the most perfect expression of what was performing well on Facebook, because when I talked to them, most of them didn’t really care about who won the election. They weren’t in this for politics; they said they were in it for money, and Trump earned them money. I remember one guy, I think he was 15 or 16 years old, telling me, “Americans want to read about Trump, so I’m writing Trump stuff.”

And what they realized – and this is why I think one of the really important things about the Macedonians is that they give us the perfect expression of the biases of Facebook’s algorithm at the time. What they did, the pro-Trump stuff, because that’s what performed best on Facebook. Not only did they do pro-Trump material, but they also did false pro-Trump material, not because they consciously were sitting around making stuff up, but because they would see that stuff perform well elsewhere on Facebook. They would copy it, and they would do it themselves.

For example.

Some of the stories we saw: Macedonians publishing Hillary Clinton being indicted; the pope endorsing Trump; Hillary Clinton selling weapons to ISIS, just completely over-the-top false claims that then performed extremely well on Facebook.

And they were making money from this.

They were making money from this. One of the things that we saw was that it’s not like there was just suddenly a cluster of pro-Trump sites out of nowhere. It’s not like they were running pro-Trump sites and nothing else. They also had websites about motorcycles. They had websites about cars. Any kind of niche that they could make money in, particularly like health, they were also running those sites.

And as you’re discovering this, what are you thinking?

At first it’s just such a strange and perfect story about the internet; that really, teenagers in Macedonia are making big money from pro-Trump content? For me, on the beat of fake news, to discover this weird cluster there was completely wild. But the thing that I remember talking about with our editor in chief, Ben Smith, was the real economic incentives that were there and that were really powerful. A user in the United States is worth more to a publisher, to an advertiser, than a user in Macedonia. So if you’re in Macedonia, you publish in English; you don’t publish in Macedonian.

The second piece of the economic equation is that the average income for an adult in Macedonia is about $300 U.S. a month. These guys could make that easily in a day with one decently performing story, so for them it was a completely game-changing, life-changing, moneymaking opportunity because the 2016 election was getting so much attention and so much interest, because Facebook was this massive traffic firehose they just exploited that to their advantage.

Can you just set the scene of what this town looks like and feels like a little bit?

Veles is kind of a – I don’t know, a bit of a postindustrial town. There was a big factory there that was the main employer that eventually got shut down because it was basically polluting and toxifying the area. I remember going into Veles with the reporter who actually wrote the stories that eventually got that factory shut down because he was showing how they were making this toxic … and as we were looking down on it, we were standing on a mountain, and what he explained was that the mountain we were standing on was all of that toxic material that they would be dumping from the factory. In Veles there is a fake mountain that you can stand on to get an amazing view of the city there, of the city that fake news really became a massive moneymaker for.

At that point in time, you said you were interested in the Russians and whether they were having a hand in the election or playing any sort of role. This is specific obviously to Macedonian kids making some money, but were you – we’ll get to the publishing of the article, but were you also simultaneously looking into the Russians?

We continued to try and hunt – you know, particularly on Twitter, we were trying to identify Russian troll accounts, but we never really nailed it. We never found accounts that we could definitively link to Russia. And I think one of the reasons for that is the data that was clearly going to link them to Russia was really in the hands of Twitter and really in the hands of Facebook. They knew the IP addresses of who were accessing these accounts. They knew if they were created by people in Russia. They knew if ads were being paid for in rubles to run on Facebook.

So from the outside, even though I strongly suspected that yes, of course there must be some Russian element here, proving it was very difficult, and so we never published anything about Russian activities in the election.

And were you asking Facebook for any of the data?

No, I never asked Facebook for data about Russian accounts because I knew I wouldn’t get. Companies like Twitter and Facebook and to a certain extent Google, they don’t really release much. They don’t like to talk about specific accounts. They don’t like to give you a lot of data, so you really have to go out and find it and gather it on your own. Twitter is much more open with its data than Facebook has been, but you still have to kind of find your own ways around the limitations that are there to get data to try and make these stories.

So when it comes to the Macedonia story, what happens when you go to Facebook for comment about the fact that you found a town of teenagers that are gaming Facebook to spread misinformation?

You know, I actually don’t think we went to Facebook to ask them about that. We published the story because it was about these entrepreneurs in Macedonia, and we didn’t hear anything from Facebook after the Macedonia story was published from what I recall. And yeah, there wasn’t – when the Macedonia story was first published, it got some good reception, and we all thought it was just a wild story about the internet and about the economics of the internet, but it got published and people enjoyed it and sort of went on from there. That story – then all of a sudden we get to the election, and all of a sudden we find out from a New Yorker article that Barack Obama has been obsessively reading it and talking about it. Then after the election the story goes kind of really viral for the first time, because I think all of a sudden people start to think, you know, was this the reason Trump won? It’s kind of a natural human reaction. Something surprising happens, you try to make sense of it.

And this is where I think people started to glom onto this idea of fake news and saying, “Oh, that must be why Trump won, because it’s so unbelievable to me that there has to be some kind of crazy explanation.” Teenagers in Macedonia seems like a pretty good crazy explanation.

Is it a good explanation in your view?

It’s a terrible explanation. The teenagers in Macedonia did not get Trump elected, and fake news in general did not get Trump elected. There were a lot of factors. And yes, there was a huge amount of misinformation that was being spread during the 2016 election. I saw an uptick not only in the volume of it but in the in the velocity and in the engagement it was getting. We would see stories that would just go crazy on Facebook that had little or nothing to them, but as long as they were pro-Trump or anti-Hillary Clinton, it seemed like that’s what was being rewarded.

Facebook And Polarization

I know you did the study afterward about hyperpartisanship, but if you can kind of bring me through the election as if you’re – you know, it’s not ex post facto but in terms of what was happening in terms of hyperpartisanship during the election, what you were recognizing and learning.

One of the big factors that emerged in the election was what started to be called hyperpartisan Facebook pages, so it wasn’t just like your typical sort of conservative organization or liberal organization.

These were Facebook pages that kind of lived and died by really ginning up that partisanship. “We’re right; they’re wrong.” But not even just that. It was also, “They’re terrible people, and we’re the best.” So these websites that were attached to them were running articles that often had misleading headlines, and the Facebook pages were getting tremendous engagement.

We took a group of liberal hyperpartisan Facebook pages and conservative hyperpartisan Facebook pages, and over a period of several days, we fact-checked every single piece of content that they put up on their Facebook pages, and what we found when we compared the level of accuracy for a piece of content to its level of engagement on Facebook is that the more false or the more misleading it was, the more shares it tended to get. Again, we see the incentives on the platform was rewarding the most extreme, the most inaccurate content, and the stuff that was true would get far less shares.

How could that not have had an influence in the election then?

One of the challenges that we have is really quantifying how much a given piece of misinformation changes someone’s mind or not. It’s really hard to know if somebody reads an article, and that’s the defining factor of why they changed their mind. So determining the effect of a piece of media is really difficult. I do think that when you look at the misinformation that was out there for people who are already leaning one way or another or who had already strongly made up their mind, I think it was just kind of more of a push for them. It reinforced them even more, so it created a climate of hyperpartisanship.

The second thing that I think it did was certainly create a lot of confusion, which is often a goal of people who are creating and spreading misinformation, confusion about what did Hillary Clinton really do or not do, confusion about what Donald Trump did or didn’t do. That could potentially factor into people’s decisions, but to actively say that this person consumed three false articles and that made them vote for Trump is almost impossible to determine.

The Danger Of Misinformation

… If you can’t necessarily measure the impact of misinformation, what concerns you about it?

Well, one of the things that is dangerous about misinformation is that it simply takes up space from accurate real information, so if it’s replacing that, then over time, certainly with repetition, that really changes the way people start to see the world, and it can change behaviors. And the key there is, is the ongoing repetition of it. The replacement of real information with false or misleading information, that’s a danger to society.

One of the other pieces that really concerns me is that if we start to lose that shared sense of reality, that shared set of facts, then how do we actually make decisions as a society? How do we actually enable democracy to flourish if we don’t agree on what’s really going on or if we all agree on things that actually turn out to be false? I do think there’s a genuine threat to this, and we have to be aware of how much it starts to pervade our information space and takes up space from real information.

Let me just go back for a second. Were there any sort of seminal pieces or noteworthy pieces of misinformation that you think we should highlight during the 2016 campaign that were, you think, seminal moments of falsehood spreading?

One really notable false story that spread was the idea that the pope had endorsed Donald Trump. I think it’s become kind of a joke or a shorthand for fake news during the election. But for me it’s really interesting, because when we tracked it back, we discovered that the first site to have published that claim was run by a guy who lived in California. His day job was as a pilot, and he had been running a network of completely fake news sites. But rather than doing political news, his sites actually came up with this formula that I called the “local viral hoax.”

They would take a celebrity, and they would take a place name, and then they would put the two together. So it would be like Matt Damon had this to say about Roanoke, Va., or Matt Damon had this to say about Portland, Ore. Literally we identified tens of sites that he had been running, and he simply took the same story, replaced the celebrity’s name, replaced the place name and just churned out versions of it. And it worked because it was a volume play. People in those locations would see it and would share it.

If you only did a couple of cities, you wouldn’t make a lot of money, but if you had tens of websites publishing hundreds of articles, suddenly you start to make a lot of money. So at some point this guy decided to just try this political one. It did OK for him. But the site that it really did well for was one called Ending the Fed, which as far as we can tell is run by somebody not in the United States, was a hyperpartisan site, and they published it. They credited another site that they aggregated it from, and they got a huge amount of engagement for it.

Facebook’s Response To The 2016 Election

… I mean, Mark Zuckerberg – there’s the famous moment at the Techonomy conference, right?

Yeah.

Bring me into what that moment was and what your reaction was.

Leading up to Mark Zuckerberg’s comments, I was regularly in touch with people at Facebook at that point. I had published my study showing that the most viral fake news outperformed the most viral real news about the election, and they were really not happy with that story of mine. They pushed back a lot before we published. I showed them all of my data before we published.

They said that this is still a very small percentage of the kind of content on Facebook, and they strongly pushed back on that. I learned after I published that analysis that Facebook had actually considered putting together a public rebuttal of my story; that senior people, executives and communications people were preparing an official rebuttal to sort of say, “This story does not tell the full picture.”

For whatever reason, they decided not to do that, so at a certain point, I get a heads-up from one of the PR people at Facebook saying: “Hey, I think you want to pay attention to what Mark’s going to say tomorrow at this event. He’s going to address some things.” So we send a reporter to the event, and we wait to see what he’s going to say. I think it’s really important to keep in mind that Facebook I think at the highest levels was starting to get concerned about fake news at this point.

… What he said was not an accident. It was not an off-the-cuff comment. That was a comment that they had I think thought about a lot and decided what message they want to say, and the message was that it was crazy to think that this had had an impact. And it really went over terribly. I think instantly he started to get blowback for that.

What did you think at the time when he says it?

I thought it was an incredibly dismissive comment, and I thought that it showed that their view of it at that point was that this was going to blow over and they just had to sort of hold the line and this would go away.

Facebook In The 2016 Election

… Let’s go back to the genesis of that story then. Bring me into what it was that you were looking to study and when it was –

The election happens, and everyone’s kind of freaking out and wondering what the heck went on. I remember seeing someone on Twitter talking about comparing some really misleading or false stories to other real stories, and I remember thinking, well, that’s kind of an incomplete analysis, but then also thinking, but I have the data to do a good analysis. At that point I just wanted to try and figure out, how can I quantify in a reasonable way what I had been seeing and experiencing as somebody who was totally focused on this?

I had seen an insane amount of misinformation, but the tricky part is that you can’t really compare a fake news website to The New York Times. The New York Times has a recurring audience, a much bigger audience. Fake news lives and dies by virality. They need to have the hits, so you can’t compare overall traffic. What I decided was that I wanted to compare the virality. Of the biggest viral hits of known fake stories that I had catalogued and gathered, what if I compared those to the biggest viral hits from real news organizations about the election?

So I put together a list of 19 of the biggest most dominant news organizations in the United States, and then I went through my ever-growing list of fake news websites and articles that we had debunked during the election. … One of the things that I wanted to do with the analysis was understand at different points during the election what was the difference between viral real news and viral fake news. So I looked at the biggest hits from nine months before the election, from six months before the election and three months before the election, and I honestly expected, of course, the real news hits to be much more viral, because big news outlets are successful on Facebook. And we saw that.

Nine months out, no comparison. Big viral for real news. Six months out, big viral for real news. But suddenly things switch in the critical period of three months before the election. The most viral fake news stories on Facebook got more engagement than the most viral real news stories on Facebook, and that aligned with anecdotally what I had been seeing, and I thought that was the right way to communicate to people that when you think about the viral hits, when you think about the stories that got tremendous exposure on Facebook, you can start to look at this and say there’s no reasonable reason why these completely false stories should be outperforming scoops and big reporting from major news organizations. There’s a fundamental flaw there.

So what did that say, though? Why nine months and six months are the legit stories going viral, and how do you explain that?

I think that we see the real news being more viral nine months and six months out before the election because one, that’s a time where people are certainly paying attention but not obsessively focused with the election. I also think that we saw new players enter the closer we got to the election. We analyzed the new websites being registered, the new domain names being registered for hyperpartisan sites, and we absolutely saw a spike as we got well into 2016. People saw an economic opportunity in fake news and hyperpartisan news.

OK. So three months out – why the shift then?

Three months out we have the urgent, most important part of the election. The intensity is there, which means there’s even more attention focused on it. And by three months out, we see teenagers in Macedonia with tons of sites. We see people in the United States in other countries with fake sites, real sites. So there’s lots of players. There’s people there to make money; there’s people there to win votes; there’s people there to sway hearts and minds.

… In a nutshell, can you just describe the finding of the analysis with the fake news versus the virality study?

I think the most important takeaway of it is, is that the stuff that was completely false was getting far more exposure and attention than the stuff that was completely true. When you think about the best performing content of each, the false stuff won. And for me that’s a really good proxy of understanding that a huge amount of people were exposed to completely false information, and that is not what you want to have happen during a democratic election.

… Bring me – tell me the sort of finding there.

What you see is false stories like the pope endorsing Trump or Hillary Clinton selling weapons to ISIS getting hundreds of thousands, or sometimes close to or above a million shares, likes, comments. That’s an insane amount of engagement. It’s more, for example, than when The New York Times had a scoop about Donald Trump’s tax returns.

… And when you spoke, when you brought your data to Facebook before you published –

Yeah.

 –  what was their critique of what you’d found?

… Facebook is basically arguing that well, you know, The New York Times overall has way much more engagement on Facebook if you total up all of its articles and all their engagement. And of course that’s true. The New York Times has tons of journalists. It has a huge Facebook page. But how is it that a kid in Macedonia can get an article that gets more engagement than a scoop from The New York Times on Facebook? How is it that some random person who makes up a hoax on a website they’ve only created three days ago can score hundreds of thousands of engagements more than a scoop from a major news organization?

It speaks to the weird incentives. It speaks to a real – it speaks to how unleveled this supposedly level playing field was. It was allowing these completely unethical players to compete and beat a lot of the biggest news organizations in the world, and I don’t know how you as a platform are comfortable with that when it comes to completely false content.

And what do you make of the argument that it is actually a small amount of content that’s on the platform, that fake news is only a small amount of content that spreads on Facebook?

I think it’s true that if you look at the overall amount of content on Facebook that completely false information is a small percentage of that. That’s true. The question is, what impact does it have on the platform? That’s why looking at the virality of it, looking at the kind of exposure it gets and how it performs compared to other types of content is important. I mean, obviously there’s 2 billion people on Facebook. You don’t have 2 billion people creating misinformation every single day.

… In the three months leading up to the election, the most viral fake stories overall got far more shares, likes and comments than the most viral real stories from 19 of the biggest news websites in the United States.

Fact Checking On Facebook

… How effective has fact-checking been at Facebook?

The interesting thing about the fact-checking program with Facebook is the thing that’s most important is the part you don’t see. Forget about the labels. What matters most is that when one of these fact checkers labels something as false, Facebook is instantly taking that and factoring it into the decisions in the News Feeds. Instantly that article starts to get a lower amount of distribution in the News Feed. That’s where it’s really important. We have this really interesting dynamic where they’ve got human labor of fact checkers actually being used to tell the machines what to do, and ultimately that’s one of the kinds of cycles, one of the kinds of relationships that is going to help Facebook deal with this at scale, figuring out how to take good-quality human labor and mix that with the machines to do this on a really global level.

And you have faith that that can be an effective system?

I think it can be one piece of effective systems. The problem in the end is that there’s only so many fact checkers in the world from country to country, so you’ll never have enough fact checkers to fact-check every piece of garbage that’s on Facebook. There’s just too much of it. But what’s hopeful about it is that over time, as more and more fact checkers label things as false, then the machines actually start to learn, and maybe we get to a point where the algorithms of Facebook can credibly determine that something is not credible.

We’re not there yet. It may be several years from now, but Facebook is the kind of place that has the resources to do it, now has the data to do it. It’s going to be interesting to see what happens in the coming years. Do they actually really make huge advances in automatic detection of false content? If anyone’s going to do it at this point, it’s actually going to be Facebook.

Now, Mark Zuckerberg has said that he doesn’t really feel comfortable being the “arbiter of truth” and that we should be in some ways careful what we wish for –

Right.

 – in giving this company the power to arbitrate what’s true or what’s not. What do you think of that?

This line about “arbiters of truth” is an appealing line. It’s in one sense because of course you don’t want Facebook becoming the world’s censor, but it kind of ignores the reality that Facebook actually has the potential to be the world’s censor. Facebook has more impact over speech in the world than pretty much anything else right now, so if their position continues to be that we don’t want to have to take responsibility for this, then we’re all kind of in a bad situation, because they do have to take responsibility for it. I don’t want Facebook sitting down and deciding necessarily what’s true and what’s not, but Facebook also can’t just throw up its hands and let its algorithms decide to promote things and not promote other things. So figuring out a way of doing this that’s credible and transparent and that does not result in mass censorship is Facebook’s big challenge.

originally posted on pbs.org