Is it a hoax, a conspiracy theory, a viral meme? Can you tell real information from fake? Meet P. Takis Metaxas. Takis is a professor of computer science at Wellesley College whose research helps users answer these questions to avoid the dangers of online fraud and deception in what many have begun to call the “post-truth era.”

For over a decade, Takis has examined the power of crowdsourcing and social networks—especially Twitter—to influence and, in some cases, shape the outcomes of political events. He is a co-inventor and proponent of TwitterTrails, a tool he helped develop with National Science Foundation (NSF) funding for measuring the propagation of rumors on Twitter.

In the wake of Donald Trump accusing news organizations like CNN of reporting fake news, we interviewed Takis. As his research continues to take on new significance, we asked Takis about the rising tide of fake news, the impact of social media on disinformation for political gain, and ways of navigating the ever-changing information landscape of the digital age. (Interview posted: February 21, 2017)

PIL: How do you define fake news, and where did that phrase come from? Why does it suddenly seem as if fake news is everywhere as we find ourselves living in a “post-truth era”? What power does fake news wield in our society today, at a time when four in 10 Americans often get their news online, especially from social media channels?

Takis: The term “fake news” refers to lies presented as news—that is, falsehoods online formatted and circulated in such a way that a reader might mistake them for legitimate news articles. However, there is some confusion around the term as it has been used more loosely to mean different things to different people. Some use it to denote opinions with which they disagree, or incorrect predictions, though these are not technically fake news.

“Fake news” has been around since ancient times, but technology has made it possible to produce and consume it on a massive scale. Such articles appear on a variety of little known websites, then turn a profit by competing for clicks as advertisements on social media sites. In order to be successful in attracting user attention, they present a fake story of political nature, religious nature, or anything with an emotional appeal. Typically, fake news stories, which may or may not have some remote connection to reality, are planted on social media sites using provocative titles and images. “Clickbait” attracts the attention of unsuspecting social media users who click on links to these stories thinking they are visiting a legitimate news site. These engaged users are drawn in by the emotional appeal, while fake news providers get a share of advertising money from each click.

These made-up stories are not really news; they are a form of propaganda, aiming to trick readers into behaving in ways beneficial to the fake news provider, who is essentially a propagandist. The benefit may be political (e.g., persuading readers to vote as the propagandist wants), financial (e.g., persuading readers to click on advertisements and bring money to the propagandist), religious (e.g., persuading readers that a particular religion is good or bad), entertainment (e.g., persuading readers to spread a joke and show how gullible people are), etc. While thousands of these stories exist, the vast majority do not succeed in getting widespread attention. Those that are successful, however, spread for a variety of reasons. One of the main reasons is that most readers are applying a historically effective model of trusting news, which is typically edited and printed by some authoritative source to the Web. And these days, search engines and social media show results where anyone can be an author, which can make anyone’s opinion look equally authoritative.

While we can often identify and avoid being tricked by fake news, the unfortunate fact is that any of us could fall for one of these lies, especially if it is presented in a way that matches our biases and prior beliefs. In order to recognize fake news, diversity is key. It can be easier to recognize fake news if a group of diverse people, with a broad variety of individual biases (and therefore tendencies to believe or be skeptical about different stories), engages with the information together. In contrast, members of a homogeneous group (an “echo chamber”) are easily fooled when presented with lies that conform to their common biases.

Unfortunately, people tend to form echo chambers in social media and in life. We find comfort and safety with others who are similar to ourselves. And when we are presented with evidence that our beliefs are incorrect, we try hard to avoid challenging our belief system. This is when we are most susceptible to lies; when they are presented in a way that confirms our prior beliefs.

PIL: What is TwitterTrails and how does it work? What questions can TwitterTrails answer for its users, and who is mostly likely to use the open access tool? What, for example, did you and your colleagues learn when you used TwitterTrails to track the infamous #PizzaGate story?

Takis: TwitterTrails is a system we developed with my colleagues Eni Mustafaraj and Sam Finn, and several undergraduate students at Wellesley College. It was designed to help us measure the spread and skepticism of particular claims or memes on Twitter. It starts by collecting all relevant tweets by using a keyword search on Twitter (e.g., “#PizzaGate”). The tweets are then analyzed to answer questions, such as who first tweeted about the topic, who first “broke” the story (made it widely known), how widely it spread, which accounts were important to spreading it, and more. It can answer other questions, too, such as whether there is skepticism around the topic, polarization in the accounts tweeting about it, political bias surrounding the issue, or an echo chamber effect. Throughout the process we also collect relevant data, such as the images shared with these tweets and the URLs of any websites mentioned.

In its present form, TwitterTrails is a tool useful mainly to researchers and journalists, but we include a summary of each story we collect. We also have a blog we use to comment on stories that may be of more general interest.

We investigated the infamous #PizzaGate story and we found several points of note (beware that the data set is large and it may take a couple minutes to load on your computer). The most relevant finding is that it spread in a very dense echo chamber. In Twitter discussions about the topic, hundreds of conspiracy theories were offered as “further evidence” and shared millions of times among a few thousand accounts. As we heard the #PizzaGate story reported by real news outlets, one such user believed this falsehood so intensely that he took it upon himself to attempt to act violently as a result of this story. As he found out, lies can cause real hurt. And despite the widespread coverage debunking #PizzaGate, there are still accounts spreading similar conspiracy theories on Twitter today.

PIL: What do the two metrics of TwitterTrails – spread and skepticism – mean? Why is this a valuable pair of measures for deciphering the veracity of posts on social networks? What have you learned about the spread and skepticism of Twitter vs. Facebook posts?

Takis: We have observed that, overall, claims on Twitter have different spread and skepticism scores depending on their validity. Claims that receive higher skepticism and lower spreading scores are more likely to be false. On the other hand, claims that receive lower skepticism and higher spreading scores are more likely to be true.

Why do these patterns hold? We have found that when people are doubtful of the truthfulness of a tweet they either do not retweet it (resulting in a smaller spreading score), or they may tweet questioning its claim (resulting in a greater skepticism score). I should also point out that these findings are not valid for claims that spread within an echo chamber. The common biases shared among the participants in an echo chamber makes it easier for people to be fooled and promote a lie, as it happened with the infamous #PizzaGate story.

These findings are specific to Twitter, however, and do not hold for every social network. For instance, it seems that differences in interfaces between Facebook and Twitter may cause users of the two social networks to behave differently. On Twitter, a tweet containing fake news and a rebuttal tweet refuting the story get the same “real estate” on a user’s screen: 140 characters (plus, optionally, a thumbnail image). This is not the case on Facebook, where an initial post gets much more exposure in terms of space on the page than any comment refuting it. Furthermore, followup comments on a post are quickly hidden by further comments, all of which makes it harder for readers to see skepticism or disproofs of a claim. These interface differences may help explain why Facebook researchers found that false claims on Facebook live for a long time, even when the truth is easily accessible, for instance, by searching on Snopes. In contrast, we find that far fewer Snopes-disproved claims live long on Twitter.

Unfortunately, while we would love to do so, we cannot include Facebook data in our own research because the company does not allow researchers outside the company to access its data, in contrast with Twitter’s more open policy. Facebook’s widely-publicized central role in spreading fake news in the last year may damage their credibility unless they act decisively to stop this phenomenon.

PIL: What is a Twitter-bomb, a term your colleagues coined? What did your research find about the 2010 Massachusetts election for U.S. Senate and the predictive power of social media? Could your research have applied to last November’s presidential election? If so, how?

Takis: In late 2009 we started collecting Twitter data to investigate whether it could be used to predict the results of the 2010 Massachusetts special election for the late Senator Ted Kennedy’s seat. While we did not find evidence that Twitter could be used to predict the election, we did discover from our analysis that propagandists were employing several different strategies to influence those users who were following the election on Twitter.

Image originally appeared in “Limits of Electoral Predictions using Twitter” by D. Gayo-Avello, P.T. Metaxas, and E. Mustafaraj http://cs.wellesley.edu/~pmetaxas/ICWSM11-limits_predict_elections.pdf

In one such election-influencing strategy, we found nine Twitter accounts created in a 13-minute interval two days before the election behaving as “bots” (powered by code to post tweets automatically). These nine accounts sent 929 tweets in the course of 138 minutes (before being shut down by Twitter), addressed to 573 unique users who were involved in online discussions about the election. These tweets were lies about the Democratic candidate, Martha Coakley. We estimated that the audience of these messages, which were amplified by retweets, reached 61,732 Twitter users. Upon further investigation, we discovered that the bots behind this attack were organized by a propagandistic website belonging to a Republican group from Iowa. We termed this strategy a Twitter-bomb. Since then, we have recorded more Twitter-bombs. If you have ever noticed a tweet in reply to something you posted coming from an account that does not follow you, you may have been targeted by a Twitter-bomb.

We and other researchers have tried to automatically detect them before the recent elections, but they have become increasingly sophisticated. Without access to the whole Twitter database, which we do not have, it is not easy to discover Twitter-bombs because they do not use hashtags or keywords to make it easy to search for them. They just target unsuspecting people who have posted about the elections and send them propaganda as a reply tweet.

Most researchers agree that in the last election, the main issue seemed to be one of fake news spreading widely on Facebook, but it is not easy to determine how effective this strategy was in influencing voters.

PIL:Twitter recently banned several high-profile users in response to online abuse from those who many consider to be trollsFacebook has historically relied on users to report abusive or misleading posts, but has begun considering using third-party organizations to assist in addressing these issues. You have studied crowd behavior and crowdsourcing as a means of control and influence, but is this a sufficient means of policing social media communities? What skills should users of social networks, such as Twitter or Facebook, apply to evaluate content in their feeds?

Takis: It is tempting to look at a social problem that has gained greater attention due to technology, such as fake news or trolling, and think the solution will be exclusively technological. Artificial intelligence and machine learning, as well as crowdsourcing, are strategies that will certainly help. But I believe there are three components that will be critical in addressing these issues.

First, we should draw upon professionals with expertise in recognizing quality in written publications, such as librarians. Librarians are a critical group because they have formal education in doing comprehensive research in a variety of domains, and can detect misinformation much more accurately than the average internet user. It is my hope and vision that groups of academic librarians will provide their expertise to create a Wikipedia-like database that will identify and document websites promoting lies, fake news, and propaganda. Like Wikipedia, this database would have transparent deliberations discussing the addition of any new site. This approach is powerful because it will scale faster than the production of fake news—it is far easier to evaluate a website than it is to create and promote a fake news story. The findings of this database can enable browser plug-ins, much like one created recently by the Washington Post, to warn readers finding fake news in their social media streams. This work can also allow advertising companies to cut off ad revenue by refusing to work with providers of fake news.

But at the end of the day, we will not solve this problem by clicking buttons to receive answers. Much of the responsibility resides with us, the users of the internet. To survive and thrive in the 21st century it is not enough to know the basics of reading, writing, and math. Today, educational competencies include the need to understand two more things: critical thinking and ourselves.

Critical thinking is more than just logic or common sense. It means using the scientific method in our thought process, from generating a hypothesis to evidence collection to drawing a conclusion. First, to generate a hypothesis we must be able to articulate a clear, unambiguous statement central to the issue we are trying to evaluate. Next, we need to search for evidence both for and against our hypothesis—it is easy to bias one’s thinking by forgetting to address the hypothesis with adequate skepticism, no matter how true we believe it to be. We must also evaluate each piece of evidence for validity, only holding on to evidence we know to be trustworthy and unbiased. At the end of this process we must draw a conclusion about whether or not our hypothesis is supported, regardless of our original opinion.

In order to use the scientific method effectively, it is necessary to be aware of our own perspective and point of view. Understanding ourselves deeply means examining our stereotypes, biases, preconceived notions, hopes, and fears—even those which we hold unconsciously—because otherwise they may mislead us. After all, we have all believed a falsehood at some point or another. All these factors have been influenced by our culture, friends, education, experiences, and other aspects of our environment. Biases and stereotypes, especially, are used by all people as a shorthand to make quick judgements, but when they are misapplied the results can be disastrous for ourselves and our peers. The more aware we are of our own weaknesses and shortcomings, the less likely we are to fall prey to them. ΓΝΩΘΙ ΣΑΥΤΟΝ (knowledge of self) was, according to the ancient Greeks, an essential part of education; this is still true today.

Applying self-aware critical thinking is a difficult process! It can be tiring and onerous, and require much more mental effort than making assumptions or using biased shortcuts. But, as social media continues to show us, the stakes are high. With practice, this process can begin to feel intuitive and natural. And in our online world, there will be plenty of opportunities for practice.


P. Takis Metaxas is a professor of computer science and founder of the Media Arts and Sciences program at Wellesley College. He is also an affiliate at Harvard’s Center for Research on Computation and Society (CRCS) and the Faculty Director of the Albright Institute for Global Affairs.

In addition to teaching, Takis studies how the web is changing the way people think, decide, and act as individuals and members of social communities. Much of Takis’ research is in web science, an emerging interdisciplinary field that connects computer science to the social sciences (sociology, political science, psychology, economics) and natural sciences (biology).

Takis’ research publications on web spam and propaganda have received Best Paper Awards from the International Journal on Advances in Security (IARIA), Lecture Notes BIP, and the Web Science 2010 Conference Proceedings.

Media coverage about Takis’ work has appeared in The Washington PostScience (podcast), The AtlanticThe Huffington Post, and MIT Technology Review, and many other mass media outlets.

As an undergraduate, Takis studied mathematics at the University of Athens (Greece) and computer science at Brown University. He received his masters and a Ph.D. in computer science from Dartmouth College and has been a visiting scientist at MIT, Sydney University (Australia), and Harvard University.

Smart Talks are informal conversations with leading thinkers about the challenges of higher education, teaching, and lifelong learning in the digital age. The interviews are an occasional series produced by Project Information Literacy (PIL).

PIL is an ongoing and national research study about how students find and use information for courses and for use in their everyday lives and as lifelong learners. Smart Talk interviews are open access and licensed by Creative Commons.

Picture

“Takis Metaxas: Separating Truth from Lies” (email interview), by Alison Head and Kirsten Hostetler, Project Information Literacy, Smart Talk Interview, no. 27 (21 February 2017), is licensed under a Creative Commons Attribution-Non-commercial 3.0 Unported License.