IRA FLATOW: This is Science Friday. I’m Ira Flatow. You know, it’s hard to give a simple description of conspiracy theories like Pizzagate and QAnon. The baseless claims involve political figures, child sex trafficking, and secret codes. And despite the fact that they’ve been disproven, these theories are still popular enough to drive people to take violent actions, like bringing a rifle to a pizza parlor or storming the capital of the US. But at the end of the day, a conspiracy theory is a story.

And my next guest is part of a team of researchers that has wondered, can machine learning tell us something about these stories– how they’re structured, how they break from reality, all in the aim of disrupting conspiracy theories in the future? Dr. Tim Tangherlini is a professor of folklore at the University of California, Berkeley. Welcome, Tim.

TIM TANGHERLINI: Thank you for having me on, Ira. And I would be remiss if I also didn’t mention my colleague, Vwani Roychowdhury, who I did all of this work at the Narrative Modeling Group.

IRA FLATOW: Credit where credit is due. You know, I couldn’t help but notice your academic affiliation as I introduce you. You’re a folklorist. How are conspiracy theories folklore?

TIM TANGHERLINI: That’s a great question, Ira, and remember that folklore is cultural expressive form circulating informally on and across social networks. And one of the things that conspiracy theory does is it links together lots of different stories that we’re familiar with as a way of explaining what’s going on in the world. So it’s a totalizing approach to explain all of the things, often threats or worries, that we have in the world.

IRA FLATOW: You know, in the past, I know my image of a conspiracy theorist was the lone guy in the garage with a bulletin board and all those newspaper clippings. But that’s very much changed, hasn’t it?

TIM TANGHERLINI: Right. So the original image that we have is the person in their garage or basement with the wall of crazy, taking little red thread and connecting all the parts of this narrative world that they’re pulling together, all of the beliefs and narratives, which are more or less accepted on faith and kind of are linking them together. What the social media platforms have allowed is more people to be part of the process.

So then it becomes kind of a crowdsourcing of these otherwise disparate parts, these little stories or story parts that are circulating, and people pull them together. They tried some out, and they discard others. So all of the norms, all of the beliefs, all of the values that we have are negotiated often through the process of storytelling and negotiation of the parts of those stories.

IRA FLATOW: When you say “negotiating,” you mean everybody’s bringing a little bit of what they want this story to be like, and then you negotiate what the final story is?

TIM TANGHERLINI: Yeah, it’s a little bit like the noisy bar problem. So I walk into the bar, and people are sitting around, and they’re talking, and I can hear little bits and pieces of those conversations. And what people are doing in conversation is they’re often telling parts of stories or complete stories, and people are interrupting and saying, well, that’s not how it went, or Amy didn’t say that, or Ira didn’t go to the lake like that.

Or somebody comes to the conversation– you’re not going to believe what I just heard about our good friend, Bob. And then that becomes part of the conversation, and that process is, like I say, negotiated. Some people say, no, that’s not what happened. Other people try to emphasize one part of the story or another. And in this context, most importantly, we’re talking about narratives that often are structured as threat narratives or disruption narratives.

IRA FLATOW: Can folklore theory tell us why people latch on to conspiracy theories in the first place?

TIM TANGHERLINI: Well, conspiracy theories are attractive because they help explain things in these low information environments, when we’ve either got poor access to information or low trust in the information that we have access to or a combination of the both. Instead of turning to information sources that might be coming from or perceived as coming from outside our community, we turn to our community members. They’re the ones who have raised us. They’re the ones who have protected us. And they’re the ones who can give us information about what’s going on.

And conspiracy theory is– we want to call it attractive, because it helps explain all parts of the world.

IRA FLATOW: Interesting. OK, but your work isn’t just analyzing stories, but doing it with help from computers, from machine learning. How do you apply computation to something as subjective as human storytelling?

TIM TANGHERLINI: That’s a great question and one that we’ve struggled with. I think it wasn’t until I met Vwani Roychowdhury that we were able to come up with a strategy for working computationally with all of the conversations that were taking place on social media. Social media, in some ways, became the world’s largest self-archiving folklore collection. And so we had all of this data, and the challenge was, how could we find, in these conversations, the underlying generative narrative framework that was allowing people to contribute to this group storytelling of how the world is really put together?

IRA FLATOW: I mentioned Pizzagate earlier, and in the research we’re talking about today, I understand that you applied this methodology to two different stories– the Pizzagate conspiracy theory and Bridgegate, which was an actual conspiracy. Can you refresh our memories about these, please?

TIM TANGHERLINI: Bridgegate was a political payback operation launched by Chris Christie’s advisors and people in the Port Authority to shut down several lanes of the George Washington Bridge as a way to create traffic chaos in Fort Lee, New Jersey, because the mayor of Fort Lee, Mark Sokolich, had refused to endorse Chris Christie’s bid for re-election as governor of New Jersey.

Like all conspiracies, actual conspiracies, these are factual events comprised of malign actors who work covertly, often in an extralegal manner, to effect some sort of outcome beneficial to those actors. So it’s a very small group, and they don’t want their story to come out. They’re deliberately keeping that hidden. And so that story came out through the work of investigative journalists.

On the other hand, we have Pizzagate, which actually has, in subsequent years, fed into the much larger, much more totalizing QAnon conspiracy theory that was centered on a pizza parlor in northwest Washington, DC. It also involved the Podestas and Hillary Clinton and allegations of them all being in cahoots to run a pedophilic satanic child trafficking ring in tunnels underground in northwest Washington.

IRA FLATOW: So when you run your artificial intelligence about the difference between the conspiracy theory and the real conspiracy, what does the artificial intelligence say?

TIM TANGHERLINI: So it’s a little bit more complicated, like so many things with artificial intelligence and machine learning are. So we run a pipeline to try and extract the main actants– that is to say the people, the characters, the places, the things, and the relationships between those that are embedded in the conversations themselves, to figure out the underlying, what we call narrative framework, the connections and the relationships between all of these different actors as a graph.

The conspiracy theory graph– this is the Pizzagate graph– came together very quickly, seemed to have a large number of characters and places, and was connected in such a way that when we deleted the links that were coming from Wikileaks, each of those communities that we had discovered in the graph, Democratic politics, the Podestas, satanism, and casual dining fell neatly apart so that they were no longer connected.

In Bridgegate, the graph had certain features that would not allow us to do that. We could delete, in fact, all of the actors coming from Bridgegate and their relationships, and the graph would stay as one single what’s called a giant connected component. And so we could even delete Bridgegate from the entire graph, and New Jersey politics would continue, for better or worse.

IRA FLATOW: Does this seem to hold true for other conspiracy theories– QAnon, for example, or those surrounding the pandemic this year?

TIM TANGHERLINI: Yeah, so that’s a great question, and it’s one that Vwani Roychowdhury brought to our group. And as you recall, as the virus and the pandemic took hold in March, we had less information than probably all of us wanted, and there had been several years of erosion of trust in the information sources that we had. So this was a perfect opportunity for people to collectively start to come up with explanations for what was going on.

So we did a lot of work in this narrative modeling group, and what we found was that there were multiple belief narratives emerging, and a lot of those were sort of linking up to form larger cycles of belief narratives, which is the precursor to a fully formed conspiracy theory. A lot of these stories are what we would call threat narratives. There’s some sort of threat, and then people in the storytelling figure out strategies to deal with that threat. This is the classic Ghostbusters question. When ghosts appear in the neighborhood, who are you going to call? And the answer to that question is always ideological.

So here, we have a threat, the pandemic, and we have different ways of dealing with that narratively. Some people say, well, it’s a hoax, so that’s a way of saying it’s not a threat at all. Other people see it as a threat, and then what are the strategies for dealing with that threat?

IRA FLATOW: OK, so what do we do now that we know this? I’m asking, how do we apply, knowing the structure for dismantling or disrupting conspiracy theories, before people take actions, and then endanger others?

TIM TANGHERLINI: I think there are two ways that we can work. Now that we have a tool that allows us to, from any large set of conversations on the internet, through this process of we find the actants and all of their mentions and all of the contexts in which they’re mentioned, and we aggregate them. And then we find all the relationships between these actors, and we aggregate them so that it’s not a hairball. We can now find this underlying narrative framework graph, so that would allow me to make a selection of actors and their relationships and generate a story so that I could be part of the conversation.

So that’s a powerful tool. It’s a generative model of storytelling, and it rests on this underlying idea that within any conversation, there’s a limit on who you can mention and the relationships between them.

IRA FLATOW: Well, just can you give me an example of how can you disrupt a forum, an internet forum by using the techniques you’re talking about?

TIM TANGHERLINI: So in some ways, what we did was reverse engineered what was going on already. We wanted to figure out how these conversations are generated. Once you have that, then you can generate stories to introduce into that conversation, if that’s what you want to do. I think, ethically, one has to start to be very cautious. As a researcher, that’s not my position. But we do recognize that it can be used to, if not disrupt, at least contribute to a conversation, perhaps steer it. We have an understanding of the underlying narrative framework, and then we also have an understanding of the strategies that people are proposing to deal with the threat.

Someone could introduce strategies that are, perhaps, not disruptive or not violent or not leading to chaos, but rather, strategies that are much more conciliatory. And perhaps those would get some uptake in various communities.

IRA FLATOW: I’m Ira Flatow. This is Science Friday from WNYC Studios. So in case you’re just joining us, we’re talking with Dr. Tim Tangherlini, professor of folklore at the University of California in Berkeley. So are you saying that if I wanted to be a counterintelligence officer, for example, or I wanted to disrupt a potentially dangerous forum, I could introduce a strategy that is a little bit less dangerous to other people?

TIM TANGHERLINI: So the standard belief narrative structure is I set up a story– the who, the what, the where, the when– and that basically creates a sense of community between me and you. We’re basically telling stories about us or people like us, and then there’s some sort of disruption or threat. And then we come up with a strategy. What are you going to do about it?

In the case of most belief narrative, it’s retrospective. They did this, and this is what happened. And so that’s how we come up with a sense of norms and values and beliefs. If you do this, this is what’s going to happen. But in rumor, in these periods of intense lack of access to or lack of trust in information sources, the narrative stops. It says, here’s the threat. Here’s the disruption. And it says, what are you going to do about it? And that’s what pushes people into taking real world action.

And so if we can find those moments where a threat is being either overcharged and diminish that sense of threat, or if other community members can suggest strategies that are less disruptive, less violent, less harmful, then that would be where you would feel that you had done something quite successful.

On the other hand, of course, this could be used in a very different– it’s a very different and negative ends, too.

IRA FLATOW: What do you mean by that?

TIM TANGHERLINI: Well, I mean what we’re talking about here is what somebody quite jokingly referred to as weaponized folklore. And I think storytelling has always had the chance to be weaponized. We see this in a lot of paroxysms of violence through history– genocide, actual witch hunts– where people are killed, and entire communities are destroyed on the basis of storytelling. They pose a threat. What are we going to do about it? Here’s our choice.

IRA FLATOW: Isn’t that what we just saw in Washington, the attack on the Capitol?

TIM TANGHERLINI: Indeed, it is. We saw that there was a narrative that the election had been stolen and that the actual elected officials of our government were the threat to democracy. And so what are we going to do about it? And the group decided that, with certain encouragement, that they were going to march on the Capitol.

IRA FLATOW: In your AI research, would there be or could there be any way to combat that?

TIM TANGHERLINI: We would certainly be able to find that these were strategies or certain things were being represented as threats, and then it would be up to public safety officials to work with that information that we could provide them that this is where the story is going or these are the story parts that we see, and then other people would make a decision based on that.

IRA FLATOW: It sounds like, because we have so many folkloric type of conspiracy theories now, that possibly the FBI or other investigative bodies might want to employ a folklorist as part of their strategy.

TIM TANGHERLINI: Well, I’m always in favor of people employing computational folklorists. Folklore really is one of the first big data fields in the humanities, where we’re dealing with hundreds of thousands of versions of stories told by thousands of people. And so to really get a sense of what is going on in those stories and how they’re changing and how they influence the norms, beliefs, values, everything that we live by, everything that we’ve grown up with, I think it’s important that we look at the informal cultural processes. And we can do it now, because a lot of these are circulating on and across these social media networks.

IRA FLATOW: Thank you, Dr. Tangherlini. Dr. Tim Tangherlini, professor of folklore at the University of California at Berkeley.

TIM TANGHERLINI: Thank you, Ira Thanks for having me on this morning.

Copyright © 2021 Science Friday Initiative. All rights reserved. Science Friday transcripts are produced on a tight deadline by 3Play Media. Fidelity to the original aired/published audio or video file might vary, and text might be updated or amended in the future. For the authoritative record of Science Friday’s programming, please visit the original aired/published recording. For terms of use and more information, visit our policies pages at http://www.sciencefriday.com/about/policies/