Fake News Is Not the Only Problem

Go to the profile of Gilad Lotan
Gilad LotanBlockedUnblockFollowFollowing
Chief Data Scientist @betaworks, co-founder @scalemodel | NYC, Data, Visualization, Social Network Analysis geek, Adjunct Professor @NYU, @globalvoices
11 hrs ago

Fake News Is Not the Only Problem

Bias, propaganda, and deliberately misleading information are much more prevalent and do more damage.

Points: Buckle up for a longer than usual Points post, friends (and don’t worry, there are lots of illustrations). Gilad Lotan, Chief Data Scientist at betaworks, chaperones us through some trends in how conversation and attention are being shaped online and argues they present a larger problem than “fake news.” How can the ideal of an informed public be defended? — Ed.

There have been so many conversations on the impact of fake news on the recent US elections. An already polarized public is pushed further apart by stories that affirm beliefs or attack the other side. Yes. Fake news is a serious problem that should be addressed. But by focusing solely on that issue, we are missing the larger, more harmful phenomenon of misleading, biased propaganda.

It’s not only fringe publications. Think for a moment about the recent “Hamilton”-Pence showdown. What actually happened there? How disrespectful was the cast towards Mike Pence? Was he truly being “Booed Like Crazy” as the Huffington Post suggests? The short video embedded in that piece makes it seem like it. But this video on ABC suggests otherwise. “There were some cheers and some boos,” says Pence himself.

In an era of post-truth politics, driven by the 24-hour news cycle, diminishing trust in institutions, rich visual media, and the ubiquity and velocity of social networked spaces, how do we identify information that is tinted — information that is incomplete, that may help affirm our existing beliefs or support someone’s agenda, or that may be manipulative — effectively driving a form of propaganda?

Biased information — misleading in nature, typically used to promote or publicize a particular political cause or point of view — is a much more prevalent problem than fake news. It’s a problem that doesn’t exist only within Facebook but across social networks and other information-rich services (Google, YouTube, etc.).

I worry that focusing on fake news will not help us strengthen trust in institutions and create a more informed public.

The Curious Case of Hillary’s Health

Remember that whole media cycle around Hillary Clinton’s health? No? When a number of conspiracy theories, powered by some real events and amplified by mainstream media, culminated in the New York Times sending a notification that Clinton fainted? You know, that week when hundreds of articles were published on a topic many people promptly moved on from and forgot?


The case around Clinton’s health is an interesting one to track. While there are articles going back to 2007 mentioning a coughing fit, this recent cycle began as this YouTube video (posted on August 4) and started to make its way through 4chan, Reddit, and the social web; it now has 5.5M+ views and 16k comments.

The video is craftily edited, pausing and piecing together troubling visual imagery of Clinton coughing and making strange faces.


As the video picked up steam and started to gain visibility, it generated millions of views, thousands of subscriptions, and tens of thousands of shares, all within its first week, according to YouTube’s statistics.


On August 22, Clinton went on Jimmy Kimmel Live, dispelling the rumors but giving the topic more attention.


The next spike in attention happened on August 29, when Donald Trump seized on the topic, challenging Clinton to release her medical records. Look at the list of publishers who covered this:


As more and more mainstream media outlets wrote about Clinton’s coughing, the heightened level of attention and responses meant that terms such as #HillarysHealth and #HackingHillary start to trend on Twitter and Facebook. Here’s a screenshot from September 6 showing #HackingHillary as one of the trending topics:


But this was not only happening on social networks. When I search Google for “Hillary health” right now, this is the kind of content that appears on top:

Same goes for YouTube:


Everything came to a peak when Clinton actually fainted (or had a health “episode,” as Fox News put it) at a September 11 memorial ceremony.


One of the challenges around this media episode is how facts and fiction intermingle, how visual storytelling plays into our emotions, picking disturbing images, stitching together snippets of video, creating a compelling narrative — all amplified by a few real coughing fits, as well as a real case of pneumonia.

Fact checking doesn’t help, especially when the choir is being preached to. What’s more, with the support of mainstream media, countless articles were being published about Clinton’s health — drawing more and more attention to the issue. And when she actually became sick, conspiracy theorists and journalists alike coalesced around the topic, publishing continuous coverage and sending out dire notifications.

“Hillary’s Health” Viral Video: data visualization highlighting the first group of users who shared this video showing Clinton’s “Bizarre Behavior” (now has over 5.5M views).

Polarized Networked Spaces

As an Israeli, the topics of political polarization, filter bubbles, and information warfare are things I’ve been obsessively studying for many years. Israeli society has been subject to these phenomena through a number of wars and military operations.

With increased political polarization, amplified by homophily — our preference to connect to people like us — and algorithmic recommender systems, we’re effectively constructing our own realities.

Two years ago I wrote about how social networks helped Israelis and Palestinians build a form of personalized propaganda during the last Israel-Gaza war. The shape of conversations and responses to events typically looked something like the graph below, where one frame of the story tends to stay on only one side of the graph, while a completely different take spreads on the other.

Typical Polarized Social Networked Space for Information Spreading about the Israeli-Palestinian Conflict.

In the cases that I was investigating, neither side of the graph’s frame was false per se. Rather, each carefully crafted story on either side omitted important detail and context. When this happens constantly, on a daily basis, it systematically and deeply affects people’s perception of what is real.

A more recent example from the Middle East is that of Ahmed Manasra, a 13-year old Palestinian-Israeli boy who stabbed a 13-year old Israeli Jew in Jerusalem last Fall. A video [warning: graphic content] that was posted to a public Facebook page shows Mansara wounded, bleeding, and being cursed at by an Israeli. It was viewed over 2.5M times with the following caption:

Israeli Zionists curse a dying Palestinian child as Israeli Police watch…. His name was Ahmad Manasra and his last moments were documented in this video.

But neither the caption nor the video itself presents the full context. Just before Manasra was shot, he stabbed a few passersby, as well as a 13-year old Israeli Jew. Later, he was taken to a hospital.

The visualization below shows how two very different versions of this story spread within disparate echo chambers, before any mainstream media (Reuters in this case) picked it up.

This dynamic unfolds continuously, especially around political events, helping us construct our own realities and reinforce our existing beliefs. And the pace at which content cascades through the network is staggering. Even with tools that give journalists the ability to “find content that’s about to trend,” in many cases it is already too late.

Noah Feldman writes about the completely different information realities that the Israelis and Palestinians have constructed:

Israelis believe theirs is a democratic society in which the police enforce the law rather than break it; Palestinians think Israeli security services shoot first and ask questions later…
Facts are real, and can be true or false. But how we determine those facts is highly inflected by our circumstances — which can lead to interpretations that seem crazy to the other side.

Media Hacking, Algorithmic Gaming

In a co-authored essay, John Borthwick and I define Media Hacking as the usage and manipulation of social media and associated algorithms to define a narrative or political frame. In that essay, we showed a few examples of ways in which individuals, states, and non-state actors are increasingly using Media Hacking techniques to advance political agendas.

More recently I’ve written about another form of media hacking — where Trump supporters successfully gamed Twitter’s trending topics algorithm to make the #TrumpWon hashtag trend worldwide after the first US presidential debate. As I was analyzing this data, it was striking for me just how organized this group of supporters seemed to be. They seemed to have been coordinating somewhere, all publishing to Twitter with the same unique keyword at the same time (a known tactic to get something to trend).

In the following weeks, this happened over and over again. For example, after the Podesta email leaks, Trump supporters online were synchronizing usage of the same unique hashtags that changed on a daily basis (#PodestaEmails28, #PodestaEmails29, #PodestaEmails36, etc.). A new, unique keyword that attains high velocity of shares is more likely to trend.

(In his excellent “Media in the Age of Algorithms,” Tim O’Reilly notes that “Google has long demonstrated that you can help guide people to better results without preventing anyone’s free speech.” Tim suggests that Google’s Panda algorithm update, which rewarded higher quality sites, solved a similar problem for Google as Twitter’s gameability. Based on many queries that I’ve run (see the “Hillary Health” screenshots, above), I’m not convinced.)

Here’s a Media Hacking example from this week. There’s a detailed conspiracy theory known as “pizzagate” (source: snopes). Leading up to the elections, a pizzeria’s owner noticed heightened attention on social media — a spike in Instagram followers, along with numerous menacing messages. The harassment became so serious that the owner contacted the FBI and local police.

If we look at the underlying communities driving the #Pizzagate hashtag, we can quickly identify a number of conservative clusters of users, a group of #BernieOrBust affiliated users, and a group of users who associate themselves with #Gamergate.

This kind of information analysis is a first step in the identification of agenda setting through algorithmic gaming.

Highly Organized Community Activation

Other than clear coordination among networked users around what to post and when, there are a handful of other ways in which people seem to become activated.

The more I dive into the data around the content seemingly organized groups of Trump supporters are sharing, the more I’ve been able to identify a few recurring links, such as the US Freedom Army website. From a message to their members:

Tweeting is our main recruiting tool. Tweeting instructions are in the mid-month report but if anyone needs them now let us know.
When you join your voice to a large group you have power and all the people who are not being heard can speak loudly as one. Critical to this effort is our use of Twitter. If you can help us on Twitter your efforts are very much appreciated. If you need Twitter instructions please advise us.

They clearly see Twitter as a critical recruiting tool and ask for help propagating information on the platform. There’s more specific information on this LinkedIn post:

If any of you know of anyone on Twitter with more than 20,000 followers who would like to work with us on recruiting or loan us his account please advise us ASAP. We have three enlistees whose Twitter accounts (@oceanbcake, @AbninfVet & @snikpmis) met the conditions in the above paragraph and are helping us directly.

In a recent response to Twitter’s purge of Republican accounts, Andrew Anglin at the Daily Stormer asks users to create bogus “black accounts” and start flooding the social network with content “in a manner which is indistinguishable from normal black tweeters.” He claims to already have thousands of these accounts. These types of “bot” accounts are some of the most difficult to identify because they are powered by real humans and may be “activated” in a coordinated manner to send out very specific messaging at certain points in time. (For more, see the many resources available from Phil Howard and Sam Woolley.)

Distrust in the Establishment

A consistent theme I’m seeing not only in the US, but around the world is a decrease in trust in media institutions.

It is especially striking when coming from the president-elect himself.

The loss of trust in institutions, especially mainstream media, is worrying (read this Mathew Ingram piece) because it means there’s no consensus on who is telling the truth, what is based on facts, and what is missing important context.

Back to the Old World

These trends and techniques are already making a very real impact across continental Europe. Make Europe Great Again (#MEGA) is a slogan that’s starting to take shape online.

If we look at the groups of users organizing around the #Marine2017 hashtag, we can identify the following cohorts:

Source: Scale Model.

A number of French groups supporting Marine Le Pen, as well as groups specifically opposing the EU (“Noeuro,” in dark blue), appear in the visualization. But more interestingly, we see two additional user segments which are not yet connected to the rest of the network but are organizing around the same hashtag: one is an #altright-affiliated group of users (in red) with accounts such as @dualkoondog and @thefinn12345; and the other is Russian (in light blue) with accounts such as @rykov and @JewRussophile.

Over time, these networks will evolve and likely grow more and more intertwined.

What Is Real

No. Clinton did not fund ISIS.

No. She does not have Parkinson’s or some bizarre neurological disorder.

But the web that we’ve built — the social web, search engines, and spaces governed by algorithmic systems attuned to social signals (clicks, shares, likes, comments) — makes it increasingly difficult to locate a definitive response to fabrications like Clinton funding ISIS.

There’s a broad range of not-fake-but-not-completely-true information. Leaving out information makes for a much more cohesive story but also may nudge a reader in a desired direction.

And in a world where stories form so rapidly and organically, who gets to decide what is real?

We may be able to learn a lot from spam detection mechanisms by coordinating across entities and limiting access to the criteria for what exactly is considered spam versus ham. But then, to be effective, we would not be being transparent about our actions.

There are other models of automated filtering and downgrading for limiting the spread of misleading information (the Facebook News Feed already does plenty of filtering and nudging). But again, who decides what’s in or out, who governs? And who gets to test the potential bias of such an algorithmic system?

If our collective goals include increasing trust in institutions and supporting an informed public, there is a lot of work to be done, and not only by Facebook. A number of actors, including publishers, social networks, content distributors, and forums, are all important in this space. By pointing our fingers at Facebook and looking at the extremes of fake news, I fear we’re missing out on an opportunity to actually make a difference.

Questions, thoughts? Feel free to respond below or find me on Twitter: @gilgul.

  • BlockedUnblockFollowFollowing
    Go to the profile of Gilad Lotan

    Gilad Lotan

    Chief Data Scientist @betaworks, co-founder @scalemodel | NYC, Data, Visualization, Social Network Analysis geek, Adjunct Professor @NYU, @globalvoices

  • Follow
    Data & Society: Points

    Data & Society: Points

    Points…“is categorized, measured, calculated, and represented.”

    Previous Post
    Next Post