Reality checker: Artificial intelligence as a tool for unravelling the threads of a conspiracy theory


Researchers use the infamous “Pizzagate” story and “Bridgegate” controversy to investigate a made-up conspiracy compared to a real one.


LOS ANGELES — From elaborate scientific hoaxes to clandestine government plots, conspiracy theories abound in this age of social media and internet forums. The speed at which these false narratives proliferate, and the profound effects they can have on real-world decision-making, have led professors at UCLA to further explore what differentiates a conspiracy theory from an actual conspiracy. By combining sophisticated artificial intelligence with insights about the way stories are structured, the study demonstrates how fragile the narrative framework of a conspiracy theory really is. Moreover, the research shows how easily the theory unravels when relatively few of its elements are removed.

The work is based on two clear target narratives. The first, known as “Pizzagate,” is a debunked conspiracy theory that went viral in 2016. As a result, several high-ranking Democratic officials, including Hillary Clinton, were accused of running a human trafficking and child sex ring out of a Washington D.C. pizza restaurant. The second narrative, dubbed “Bridgegate,” is an actual conspiracy that took place in 2013. That’s when associates of then-governor of New Jersey, Chris Christie, colluded to create massive traffic jams by closing access lanes to the George Washington Bridge in an attempt to punish one of his political opponents.

For the study, published in PLOS One, researchers turned to a form of artificial intelligence, called machine learning, in which computer algorithms improve automatically through experience. Using this tool, the group created an automated graphical model that analyzed three main structural components of the narratives: the actants (people, places, things), the relationships between actants, and the sequence of events involving the actants.

Data for the research comes from two online repositories at the UCLA library. For Bridgegate, the researchers drew from an archive of nearly 400 news reports. Locating solid source data for Pizzagate proved to be more challenging, however, due to the fragmentary nature in which conspiracy theories are often discussed. Overall, the team analyzed approximately 17,500 posts related to Pizzagate that had been archived from the discussion boards of social news aggregate websites, Reddit and Voat.

“Finding narratives hidden in social media forums is like solving a huge jigsaw puzzle, with the added complication of noise, where many of the pieces are just irrelevant,” says Vwani Roychowdhury, one of the paper’s lead authors, in a media release.

Conspiracy theory graph from UCLA study
Researchers produced a graphic representation of the narratives they analyzed, with layers for major subplots of each story, and lines connecting the key people, places and institutions within and among those layers. Courtesy: UCLA (Click to enlarge)

Unlocking the elements of theory vs. reality

The AI allowed the researchers to sift through the massive amount of data generated by these posts in order to fit the pieces of the puzzle together and make a meaningful narrative. The resulting visual representation (right) illustrates clearly how each element of the narrative is connected. Nodes are created for each actant, lines depict relationships between actants, and layers represent each major subplot of the story.

Based on this graphical presentation, researchers were able to tease out differences between the conspiracy theory and the true conspiracy. The authors suggest that an actual conspiracy consists of a large number of densely interconnected actants that are confined to a single domain of human interaction, such as New Jersey politics in the case of Bridgegate. Even if an element of the story is removed from the framework, the narrative holds up because of their multiple points of connection.

By contrast, conspiracy theories have a more tenuous structure. They consist of a smaller number of essentially constant actants that have fewer connections across multiple domains. As a result, the deletion of any component of the framework leads to total collapse of the narrative.

“One of the characteristics of a conspiracy theory narrative framework is that it is easily ‘disconnected.’ If you take out one of the characters or story elements of a conspiracy theory, the connections between the other elements of the story fall apart,” explains Timothy Tangherlini, a lead author.

Another difference the researchers uncovered between conspiracies and conspiracy theories was how quickly the narrative emerged and became stable. In the case of Pizzagate, the complete network of actants and relationships emerged within the first month and remained largely the same during the entire three-year data collection period. On the other hand, the entire cast of actants associated with Bridgegate took nearly six years to be fully described.

Conspiracy theories debunked more quickly?

An important takeaway from this study is that its methods do not require comprehensive collections of pre-existing data. In fact, any number of internet resources can be mined for relevant information to a specific story. That’s because people interested in a particular topic tend to seek out similar forums and form tight-knit communities. This means that the approach could be used to identify conspiracy theories in near real-time, a potentially useful application for monitoring breaking news stories that might be particularly vulnerable to reinterpretation by conspiracy theorists.

One current news story the group has examined is COVID-19. In another recently published paper, they applied their methods to track the evolution of several coronavirus-related conspiracy theories. They demonstrate how the narrative frameworks of four main conspiracy theories have begun connecting to each other. Moreover, the researchers suggest the possibility that these theories could eventually merge to form a single conspiracy theory. They were also able to identify hints of brand-new conspiracy theories as they were emerging.

The authors highlight the importance of their research in the realm of public security. Conspiracy theories touch on the deep-seated insecurities and fears of many people, and in some cases, they have even incited violence. In the case of Pizzagate, an armed man was arrested after firing an assault rifle inside the pizza restaurant, acting on the mistaken belief that he was saving trapped children.

“Given the challenges that conspiracy theories present to democracy and a free and open society, we believe that the ability to automatically discover the underlying narrative frameworks for these accounts is of paramount importance,” the authors write in the study.

Like studies? Follow us on Facebook!

Follow on Google News

About the Author

Judy Minkoff, PhD

Judy Minkoff holds her doctorate in immunology and molecular pathogenesis from Emory University. She has over a decade of experience in preclinical laboratory settings working on viruses and vaccine development. She was a medical writer for two-and-a-half years and has been a freelance science writer and editor since 2016.

The contents of this website do not constitute advice and are provided for informational purposes only. See our full disclaimer

Leave a Reply

Your email address will not be published. Required fields are marked *