How AI Helps Us Tell Conspiracy Theory Fact from Fiction

January 25, 2021 - 8 minutes read

Conspiracy theories are everywhere these days, and social media is perpetuating how fast people learn about them. What once were doubtful ideas with little evidence supporting them, conspiracy theories have moved into the realm of causing harm and danger to whoever or whatever is thought to be the cause of the theory. Forums that draw a fuzzy line between freedom of speech, freedom from moderation, and the ability to post something without fact-checking are at the center of the spread of these theories.

People can join social media networks like Facebook and Reddit to further entrench themselves into their beliefs and worldviews, allowing them to develop their theories, plan to take action, and “investigate” the ideas in the theories. Using machine learning applications to graph out and connect facts and ideas within theories, it’s possible to distinguish between a real conspiracy and a conspiracy theory. With these tools, we can catch false narratives earlier, before they spread out to everyone who’s likely to take part in sharing them.

Tracing Conspiracy Theories

A research group at the University of California Berkeley, located outside of San Francisco, has developed an automated way to deduce whether an idea on social media is a conspiracy theory or not. The researchers have seen success when applying their algorithm against the COVID-19 pandemic, anti-vaccination movement, and the Pizzagate conspiracy theory. Next, they’re going to use the algorithm to study QAnon theories.

The researchers defined real conspiracies as those that are purposefully hidden from the problem and involve real people and real actions used for a negative motive. On the other hand, conspiracy theories are those that are constructed through collaboration, develop in the open, and have little to no factual evidence supporting them. Often, conspiracy theories are intentionally complex and play upon fear. They usually try to explain everything, not just one thing.

Conspiracy theorizing is no longer the concept of a lone person hanging photos on their walls and piecing together a grand idea by connecting everything together. It now involves an online gathering of people who share ideas and riff off of each other to create a collective story. The good news is that because conspiracy theorists have moved online, they’re easier to study and their stories are easier to trace from origin to rumors to the final narrative.

For the UC Berkeley researchers, Pizzagate was a perfect subject of study. It grew in narrative in October 2016 during the weeks leading up to the presidential election. It took a month to finalize the narrative which involved characters from a variety of domains that were otherwise unrelated and previously unlinked. This included the lives of the Podesta brothers, Democratic politics, pedophilic sex trafficking, and casual family dining. The connective thread for these seemingly unrelated topics was the leaked emails from WikiLeaks of the Democratic National Committee in late October 2016.

Defining the Narrative

The research group from Berkeley used advanced artificial intelligence (AI) development to create a set of machine learning tools that connect places, people, things, and relationships outlined in each conspiracy theory. Machine learning algorithms can process a large amount of data in one niche to analyze and categorize the information into more organized buckets. Using these algorithms, the researchers analyzed almost 18,000 posts from Reddit and 4chan during the timeframe of April 2016 to February 2018.

The model views each post as a part of an unknown story and tries to use subsequent posts as pieces of information to uncover the true narrative. It determines what the major elements of the theory are: places, people, and things. Then, it works to connect all of the elements. To test if the algorithm created an accurate graph of relationships, the researchers compared the output to illustrations from The New York Times. The result was a close alignment with the publication’s illustrations, and the algorithm had even gone above and beyond to produce finer details about the places, people, things, and their relationships to each other.

Comparing Relationships

To test if the algorithm could deduce if a conspiracy was real or a theory, the researchers had the algorithm analyze Bridgegate, which was a political operation launched against the Democratic mayor of Fort Lee, New Jersey by Republican Governor Chris Christie’s administration. The narrative graph for Bridgegate took from 2013 to 2020 to uncover while the Pizzagate narrative took a month to develop and stabilize. This was the first feature that distinguishes conspiracy theories from actual conspiracies: the time it takes to establish and stabilize the narrative.

The second feature that the researchers found that points to a conspiracy theory is the removal of elements from the narrative. The Bridgegate conspiracy remained as a single, connected network despite removing key relationships and people from the scandal. In contrast, the Pizzagate theory was fractured immediately upon removing key figures from the graph. The Pizzagate fragmentation more closely mirrored the reality of the theory: none of these disparate figures and objects were actually related other than through a false narrative.

artificial intelligence app development

Ethical Considerations

The researchers are aware of the ethical issues and challenges that their work raises. For example, their methods could be reverse-engineered to generate more robust conspiracy theories that look a lot more legitimate in their graphs. They could also be used to supplement current conspiracy theories with more realistic elements and relationships. But the researchers say that they’ve found more and more instances of weaponized storytelling without people knowing about their methods.

Theories vs Facts

Conspiracy theories have the potential to unfurl the fabric of our societies and politics, and it’s imperative that we develop tools to identify them. Using these tools, we need to alert authorities and researchers to growing movements of new theories, allowing officials to mitigate the narrative and debunk falsehoods.

Have you seen any mention of conspiracy theories on social media? Which theory did you see information about and which platform was hosting the posts? As always, let us know your thoughts in the comments below!

Tags: , , , , , , , , , , , , , ,