Berkeley Talks: Computational folklorist on how storytelling becomes belief

(Music: “Silver Lanyard” by Blue Dot Sessions)

Anne Brice (intro): This is Berkeley Talks, a UC Berkeley News podcast from the Office of Communications and Public Affairs that features lectures and conversations at UC Berkeley. You can follow Berkeley Talks wherever you listen to your podcasts. New episodes come out every other Friday.

Also, we have another show, Berkeley Voices. This season on the podcast, we’re exploring the theme of transformation. In eight episodes, we’re looking at how transformation — of ideas, of research, of perspective — shows up in the work that happens every day at Berkeley. You can find all of our podcast episodes on UC Berkeley News at news.berkeley.edu/podcasts.

(Music fades out)

Timothy Tangherlini: Thank you all. My God, look at the size of this crowd. You have to tell all of my students to turn out like this.

Thank you all for coming. I hope you had a chance to read our land acknowledgement. I think this is very important. And I want to welcome you all back or welcome you here to UC Berkeley. And I’m just delighted that I was invited to be part of this wonderful weekend.

I started off with a title, “Parler Games: Conspiracy Theory, Insurrection and Computational Folkloristics,” I know is a mouthful, but so is my last name. And I had started off really by thinking about this in the context of a storytelling approach. And because I’m a folklorist, I work so much with storytelling and so much with the way people frame belonging and not belonging inside and outside. But then something happened in the news and I realized I might need to change the title of my talk. And poor little one on the right.

And then it even got better because you almost never have a public figure endorsing the main premise of your talk, and yet there it was. If I have to create stories so that the American media actually pays attention to the suffering of the American people, then that’s what I’m going to do. And so it was almost, I had to say you just stole my thunder. And so I was a little bit more angry at this individual for taking the whole premise of today’s talk, which is that stories play an enormous role both in the way we frame our understanding of the world around us, but also can inform our decision making. And this is not a new phenomenon; this is something that’s been going on for quite a while.

I want to start with a story. Remember, I’m a folklorist, so I’m going to bring us back to 19th century rural Denmark. And I want to tell you a story actually that was collected from a farmer living out in Jutland, which if you recall third grade Danish geography, is the peninsula of Denmark. And there’s a group of hidden folk out there down by the river who menace the local population by stealing their pets and eating them. A guy told me the following: His father worked in Taastrup. And at that time, the forest went all the way down to the town of Taastrup. Now there’s practically a mile up to it. As you know, in the forest, there were elves who had hollows in their backs. They often came to him when he was a shepherd and they tried to steal his bread from him. That’s really what they wanted. OK, just a side note. Anybody been to Denmark? Yeah, smørrebrød, the open-face sandwich. That’s really what he’s saying, “They’re trying to steal my open-face sandwiches.” When he says, “They’re trying to steal my bread,” that’s what he means. Just a footnote there. Apologize for that. They also went into town and were bad about stealing there and taking the townspeople’s cats. It’s said that they ate them. People had to watch out that their cats weren’t alone in a room because then they would get snatched.

And you think, OK, well that’s an interesting story from the 19th century, but how does that translate up into the present day? Well, fast-forward to 1994 in Denmark in the phase of what certain largely populist political parties were saying was an overwhelming influx of refugees and immigrants. And this is a story that was collected by a colleague of mine. I heard the story about the dog that was stolen and eaten by immigrants about four years ago. It was in a circle of reasonable people, but the teller knew the people it happened to. These people took a walk with their dachshund, I think to now it would be like your French bulldog, when all of a sudden a couple of immigrants snatched the leash out of their hands and disappeared with the dog. When the police found the culprits, dinner was just finished. The entire family was sitting around the table and only the well-gnawed bones were left.

What do we get from this? Tradition and belief provide resonance across decades if not centuries. Like the hidden folk, immigrants lead a life that’s hidden from the, quote, “native population.” They have a separate culture and language. They work the least popular jobs, and there’s minimal chance for assimilation into Danish culture. Often then, physical characteristics set them apart from ethnic Danes. And keep that in mind; the idea of insider and outsider and the dehumanization of outside groups is something that is constructed through storytelling. And so that’s a little prelude into our talk today where we’re going to be looking at storytelling, folklore, and belief, but we’re going to be looking at it at a very large scale, pretty much internet scale.

Just to remind you, because I know you all went to Cal or have some association with Cal, so you probably already know this, but recall, if you haven’t already remembered, that folkloristics is a study of vernacular, informal culture and its circulation on and across social networks. You can think of your favorite social networks, something like the Instagram; I call it the Insta Snapface. TikTok. In contemporary culture, this is going to have a lot in common with social media. The informal dimensions of everyday life are created, circulated, and negotiated. This is something that’s very important where there’s always a give and take. When I’m telling you a story, you might interrupt me, say, “That’s not the way it goes.” We’re constantly negotiating what is good, what is bad, what we recognize as being to our group.

I’m reminded of the squirrel dance and no sausage day. Do many of you celebrate no sausage day? No, of course you don’t. It’s a stupid thing. Nobody would do that. It’s not a thing that we celebrate. It creates no meaning, nor do we do the squirrel dance. And so what we’re doing with storytelling is we’re often trying to align things that create meaning for us with the other people in our group. We’re interested in these dynamic processes of variation and stability. How and why do ideas persist? And how do these ideas support and create beliefs?

We’re also interested in the relationship between cultural expressive forms, groups and the dynamics of group membership. Who gets to be inside? Who is outside? And ultimately then, it means that folklore is remarkably efficient because it’s circulating all the time, and we’re constantly involved in these negotiations. It’s incredibly efficient at encapsulating and communicating cultural ideology. Things that you might think are innocuous or don’t have that much weight, actually, because of their commonality are incredibly laden with beliefs, norms, and values where norms are how you expect people to behave, beliefs are pretty much what you think the outcomes are going to be of those behaviors, and values is simply a rank list over the outcomes of that system. That’s what stories do, that’s what a lot of our informal culture does. And it’s small packets because we’ve worked through it so much.

Now, don’t confuse belief with truth, whether it’s justified or unjustified. Belief is a social construct, and it influences, I think, in profound ways how we interpret facts, our understanding of how the world works. Now, we don’t know the mechanisms yet. This is something that we’re working on. It’s even hard to measure belief, but it’s pretty clear that storytelling plays an important role in this, partly because belief is often related to causal claims. Another downstream effect of that, because belief is related to causal claims, it can influence the decisions that we make. I’m going to do this because I believe that.

You’re thinking what the heck does this have to do with folklore? But I hope that I’ve brought you along, albeit quickly, to recognize that one of the things that folklore is so very good at doing is creating a group that recognizes in this communicative setting that we’re part of a group, and that we’re now going to engage in some sort of conversational interaction that will help us align our little models of the world that we’re all walking around with with the other person’s little model of the world that they’re walking around with, and then that might influence the decisions that we make in the real world.

You can think of the used car example. You’re going to buy a used car; you’ve done all the research. You’ve decided that the best car for you is, we’re in Berkeley, so it’s a Subaru Outback. You’re absolutely sure that this is the best car for you, but you go out for drinks with a couple of friends of yours and one of them says, “Oh yeah, no, we just bought a Jeep Grand Cherokee, and it’s just a fabulous car.” And that weirdly blasts away all of your research and you go out and buy a Jeep Grand Cherokee, which, if my brother’s experiences can give caution, then don’t buy the cheap Grand Cherokee. Really, you should stick with the Subaru Outback.

How does this go to legend and rumor? Well, legends, this is something that I work on quite a bit, is very closely related to rumors. There’s sources of information that thrive in low trust or low information environments, or a combination of both. If you don’t trust information sources, you’re going to turn to your friends, family, your community to get that information. Or if you don’t have access to information, you’re going to ask the person next to you, “Hey, what the hell is going on?” And I know many of you are thinking that you should turn to the person next to you in this lecture and go, “Huh? What the heck is going on?”

Legend in this context should be seen as short, believable. One thing happens, monoepisodic narratives they’re told as true; very informal settings, often conversationally. And rumor is simply a legend that has, to use the vernacular, gone viral. Now, rumor also has a feature that it pushes decision-making into the real world. A legend is often told retrospectively, like the stories that I told you at the very beginning of this lecture about the elves stealing people’s cats and eating them or the immigrants stealing people’s dogs and eating them. We know what happened. Where rumor says, “Here’s a threat, here’s a problem. What are you going to do about it? Or What should we do about it?”

And so I want to give you a little model of legend structure. And recognize that in rumor, this part here, result, is often missing. Basically has six parts, but only three are really essential. A lot of times you say, “Hey, did you hear what happened to my friend Bob?” No, no. That’s actually a real question. No, no, of course you didn’t, no. And I can tell you, “Oh yeah, no, you know Bob. This was last week down in El Cerrito at the Hotsie Totsie, the pub.” That kind of gives you a who, what, where, when, and we have that setting, and we already have established there an inside group. We all know Bob. We like Bob. He’s a little bit weird, but we like Bob. Wondering what he’s doing at the Hotsie Totsie, but that’s OK.

And then something happens. When something happens, you have to make a decision about what you’re going to do next. And a lot of times there’ll be an evaluation, say, “Oh my God, that happened to Bob?” And then there’s a result of what finally happened.

One of the things I’ve always been annoyed by is the complicating action is not terribly theorized, and so we came up with a model that breaks it into two distinct parts. Basically there’s a threat or disruption. We start a story, we’ve created a community, there’s going to be some sort of threat or disruption, and then we’re going to come up with a strategy for dealing with that threat or disruption. That’s where it becomes ideological. It’s basically asking, “Here’s what’s happened. What should we do about it?” And here’s a story about what people decided to do about it. And if the strategy succeeds, then that gets endorsed by the community. And if it fails, it gets rejected.

I don’t know if anybody’s seen the film, The Exorcist. I use this a lot as an example. My wife got upset when I showed it to my children when they were very young. I was like, “No, no, it’s research.” But recall The Exorcist, which is a phrase I’ve always wanted to say in public. The story starts with a broken family and a movie star who moves to Washington, DC to film a movie. And one morning while she’s having her breakfast, her daughter spider walks down the stairs, spewing blood and speaking in a very menacing voice. And as any parent, she thinks, oh, it must be a stomach flu. And so she gives her child a little bit of Tylenol and some Pepto-Bismol to … That doesn’t work. I don’t know if you’ve ever looked at the labels on Tylenol and Pepto-Bismol; these are not for satanic possession.

She does what any parent does and she brings her child to the doctor, the physician, the pediatrician. The pediatrician looks at Reagan and checks her vitals and looks in her throat and says, “Oh, I’m going to prescribe some antibiotics.” And again, really, the doctor should have been prescribing anti-satanics, not antibiotics, because again, those fail. The next thing we see is Reagan’s being examined by a whole room of experts. And the experts come up with all sorts of theories, and they’re all wrong.

Now, fortunately for this mom, she’s right near Georgetown University, which has an excellent department of exorcism down on the basement floor of one of the buildings there. And she goes, and she finally finds an exorcist. And the exorcist is weirdly successful, although he dies in the process of exorcising Satan from little Reagan. But what’s interesting then is that she goes back to Hollywood, I believe they get married again, and Reagan probably lives happily ever after. This is a strongly ideological film. If you look through all of the strategies, they try all of the strategies of modern science, and it’s not until they come to the Catholic strategy … Not really Catholic, but the strategy of using a Catholic priest for exorcism that Reagan is cured. There is this ideological component to strategies and application.

Just very quickly, we have this model that we have to use. We have to have a formalization of this model so that we can apply computational methods to it. And looking at my time, I think I’m just going to show you the slides but not tell you what’s on them. I’m a professor; we do this all the time. Student’s like, “Wait, how is this on the exam?” It’s like, “Oh, no, it’s in the slides.”

This brings us to the idea of narratives, of threat, and the role of fear. A complicating action is often going to propose a threat or disruption that is linked to the shared fears or concerns of a group. Now, remember, this is dynamic, so those are negotiated over time. What are we afraid of as a group? And a strategy to deal with that threat or disruption is culturally acknowledged as potentially efficacious. The strategy that you come up with, that mom comes up with for Reagan, yeah, seems that could work, but it doesn’t. This is the Ghostbusters problem. Does anybody recall this classic film? It’s a classic of modern cinema. It’s called the Ghostbusters. I don’t know what they say in French, I’m sure it’s Les Ghostbusters. Anything that’s important, we say in French in the humanities. What does the Ghostbusters ask? It’s one of the most profound questions of the 20th century.

Audience 1: Who you gonna call?

Timothy Tangherlini: Who you gonna call? Yeah, but it’s before that, when ghosts appear in the neighborhood, when there’s a threat or disruption, who are you going to call? And the answer to that question is always going to be ideological. You can either take it on yourself, you can call an expert, you can band together and do something. But that is really a fundamental aspect a lot of these stories.

If we’re looking at stories that are told as true that are circulating widely in a community, many of them are about some sort of threat or disruption, and then the storytelling starts to explore strategies for dealing with that threat or disruption. If we can catalog these narratives, we can get an idea of the various sources of threat or disruption, that’ll tell us a lot about the group, the various strategies proposed for dealing with those threats or disruptions, also incredibly good information, and then the outcome of applying those strategies to those threats or disruptions. And by doing that, we’re going to learn a great deal about the group in which those stories circulate.

One of the questions I’ve had is actually why so much fear? And this is some research that we’re doing along with some neuroscientists. It might well be that there are circuits in the brain that are highly attuned to trying not to get eaten. And at the first perception of threat, a circuit turns on. And as we try to get more and more information, that channel might get modulated. That’s a slower channel. And so there might be some neurophysiological relationship to belief and threat narrative. That’s research that we’ve just gotten funding for, and greatly interested to see if we come up with anything.

That brings us to social media and storytelling. Now, you probably recall from philosophy classes, George Boole, the famous philosopher and also logician, he says, “In every discourse, whether of the mind conversing with its own thoughts or if the individual’s intercourse with others, there is an assumed or express limit within which the subjects of its operation are confined.” This, in some ways, gives us an idea of domain so that when we start talking about Jeep Grand Cherokees and Subaru Outbacks, were probably unlikely to be also talking about spaceships and martians. There’s some limits to what can be admissible into a conversation. One of our goals then is to, in any social media conversation, try to estimate what those limits are of the conversation and also try to understand the relationships between the entities in that conversation.

Now, if you spend any time on social media, you’ve discovered that posts are notoriously fragmentary. People are referring to conversations that started long before, earlier threads, other forms, other platforms. It really is like parachuting into a conversation and you have no idea what’s going on. We’re going to try and use computational methods to find out what’s going on without having to read all of these posts, because there are tens of millions if not hundreds of millions of posts; you can’t actually do it.

But in the aggregate for any sort of domain, these posts and discussions should constitute a collaborative storytelling process, resting on some underlying narrative framework. That’s our assumption. And this means that we can start to do some things. We can find all of the characters and places in the stories that are being referred to on social media, and we can maybe estimate whether those are insiders or outsiders. We can maybe figure out the negotiations over what happened, the complicating action. Did it constitute some sort of threat or disruption? The origin and status of potential threats and perhaps consideration of potential strategies. And if there is an end result that the social media people can comment on, then maybe we’ll also be able to get this retrospective evaluation of end results and how that then fits into this narrative world.

Now, there’s some things that you have to be careful about when you work with social media and virtual communities online. There’s an illusion of what I would call close homogeneous communities. Close homogeneous communities, those are the communities that you’re familiar with from everyday life. These are people that you share values with. You may not agree with them. Thanksgiving is always a good opportunity to explore disagreements in close homogeneous groups. But you’re brought together by shared beliefs, values, norms, which I consider ideology. And they’re marked by degrees of what we call homophily. That’s simply birds of the feather flock together. You’re more likely to gravitate towards people who you share something in common with.

Real life communities may, however, require a little bit more give and take. And they may have what I call social breaks associated with them. Things travel a little bit more slowly. And if I start spouting off about the Illuminati and space lasers and things like that, people will say, “Tim, that’s not a real thing,” and they’ll all move away from me in the bar. And so as social beings, we like to be with people who share our values, but in real life social settings, there’s a degree of slowness which is necessary.

On virtual communities, it’s at the speed of light or very close to it. And the community that you are interacting with may not actually be what you think it is. In fact, in real life, and we can think about Bob going to the Hotsie Totsie, you avoid sitting down to pizza and beer with malevolent robots. On social media, you actually do not know if you have not just sat down to the equivalent of pizza and beer with a bunch of malevolent robots. So what?

Stories and action; I think this is really important. Stories give a basis and a justification for people to take real life action. They can be retrospective justification, but they can also be motivating justification. Here’s what’s happened. What should we do about it? A famous French theorist proposes that stories represent repertoires of schemas of action. You could have collective exploration of those schemas. How should we act given a particular situation? And Anne Swidler mentions the stories present a toolkit of habits, skills, and styles. That allows a group to figure out how we as a group should react to what we might see as a disruption or as a threat.

And so one of the reasons why I think that is, so what is that, for example, people believed many of the COVID conspiracy theories because of the storytelling environments in which they were embedded. It’s much more the storytelling than anything else. And that’s how Bill Gates actually, and you’ll see in just a moment, becomes effectively Satan.

And so this brings us to the realm of experimental computational folkloristics, three words that nobody ever thought would be uttered together in public. Like so many things, Berkeley is at the forefront of this field as well. And this is where we start to unravel narrative at internet scale.

One of the things that’s kind of interesting if we start to think about conspiracy theories is you’ve all heard little bits of these in different places. But what a conspiracy theory is able to do is to take simple threat narratives and link them together to form complex representations of threatening groups and their interconnections. Has anybody ever seen those shows that have either the detective or the lunatic in their basement with what we call a wall of crazy? Yeah? All those pictures of people and red thread going between all of the different pictures and then these great big questions and usually a slogan over the top of the wall of crazy like “The truth is out there.”

We want to estimate those walls of crazy for any social media group. And as we started doing this work, we discovered that these walls of crazy, these what we call narrative frameworks have a huge amount of stability across time once they’re established. It didn’t surprise me when I heard someone using in a political speech talking about, quote, “The threat of immigrants,” to say that they’re eating the dogs and cats, they’re eating their pets. Because as I showed you at the very beginning of this talk, those stories have been in circulation for a very, very long time, and in very different places as well.

One of our questions was how are these networks held together? Are there certain topological features of the narrative framework that tell us something about narrative belief in community? Are there certain features of narrative framework topologies that are more important in solidifying ideology? We had to come up with a model; and I’ve already given you the basis of it.

And so what we did was we said, “Well, all of the actants,” these are the people, the institutions, the places that interact in these stories, “can be the nodes and the network, and the pairwise or multi-way relationships can be the edges.” We’re going to create a giant network representation of these domains. These relationships are going to be context dependent, dependent of whether it’s health, politics, power, corruption, religion, et cetera. And then we come up with this very simple little pipeline of interlocking computational methods. You can immediately see what we were doing to extract these social media network narratives. And then we run some algorithms to try and find the groupings, this very large network space, and those are the individual narrative communities that have been linked together.

And so we did a series of experiments on this. We looked at anti-vaccination movements, we looked at the rise of QAnon, we looked at conspiracy in the time of COVID-19. For the anti-vaccination narratives where we were really focused on what are called mommy blogs, we discovered that, interestingly, vaccines were not the threat. If vaccine preventable diseases are the threat, the obvious strategy is vaccinate. If vaccine preventable diseases aren’t the threat, if instead vaccines are the threat, then the obvious strategy is don’t vaccinate. Then a lot of the discussions were, “How can we not vaccinate?” And that’s how you start to get these exemption narratives taking over. Incredibly stable once these beliefs were found, once they started to emerge, stable over nine years, even though all of the membership of these mommy blogs had long cycled off. They got picked up by new members to the blogs afterwards.

We looked at Pizzagate and the rise of QAnon. And one of the things that we found was that domains that don’t have any real connection were being connected by people interpreting, in this case, a source of information, the WikiLeaks dump of emails. Casual dining, democratic politics, the Podesta brothers, and Satanism, which really don’t have that much to do with each other … Casual dining does not immediately make you think of Satanism. Or if it does, come see me afterwards; I want to talk to you, include you in one of our studies. But through these narrative linkages, they all of a sudden became, and you can see in the projection there, one big narrative network. If we removed all of the edges coming from WikiLeaks, the narrative network would fall apart. During COVID-19, we saw something else happening. A bunch of different stories all of a sudden started to get linked together via things like Bill Gates, cells, cell phones, little semantic weirdness going on there, and 5G and feeding into the QAnon narrative.

That was an interesting situation, to discover that we can find these narrative networks, we can understand what is seen as a threat, we can see who’s inside, who’s outside, and we can start to see ideas of strategies for dealing with those threats. We can also see that topologically, these networks are incredibly fragile. They’re not robust. You can delete parts of them and they just fall apart. Very different from an actual conspiracy like Bridgegate.

In Bridgegate, you could have deleted all of the actors. You could have taken out all of the people who were involved in Bridgegate and New Jersey politics would’ve continued without missing a beat for better or worse. I don’t know how you feel about New Jersey politics. But with Pizzagate, if we deleted the edges coming from WikiLeaks, the entire narrative framework would just fall apart. The problem like that is, in another film … Has anybody seen Terminator 2? Yeah, fabulous movie. What happens to the terminator? If you shoot him a lot, falls down, but is very easily able to reconstitute itself. We say that these conspiracy theories, they’re not robust, but they’re incredibly resilient. Even if you attack them at the level of these wall of crazy edges, it’s very easy to put those edges back in.

And so that brings us to a conspiracy based on a conspiracy theory, Jan. 6 and a Parler. That’s just a quick timeline going in backwards order as to what happened. And this postage stamp is not actually a postage stamp. Yeah.

Audience 2: Can I just ask, what do you mean by delete a …

Timothy Tangherlini: Oh, yeah, that’s a good question. What do I mean by delete an edge? We’ve estimated this narrative framework, which is we’re showing as a graph, nodes and edges. Some of those edges are only coming into the graph because people are saying, from the WikiLeaks dump, “I found this connection.” There are low probability edges that have been added in from the discussions themselves. If we can find those edges … In this case, in Pizzagate, it was really easy. These were edges coming from the interpretation of WikiLeaks. We could delete those edges experimentally, just experimentally. What happens to the network if we delete those edges? It falls apart into very distinct conversations: Casual dining, democratic politics, and Satanism. Yeah, great question.

The postage stamp is not a postage stamp; that’s an AI generation of a postage stamp with a prompt, “Make a postage stamp commemorating Jan. 6,” and that’s what it came up with. A little bit terrifying.

We looked at Parler. We took the entire dump of Parler. And if you don’t remember Parler, it was self-described as a haven for patriots. Not much used until Nov. 10 when Stop the Steal was banned from most big social media platforms and everybody from those groups flocked to Parler. It called itself a harbor for free speech, which is an interesting interpretation of free speech. You could post or parley up to 1,000 characters. It went abruptly offline on January 10th, partly because it was being used as one of the main planning platforms for the events of Jan. 6. Just before it went offline, somebody had the idea to check how secure the platform was, and the answer was not secure at all, so they dumped all of the data from Parler and made that available to researchers.

We didn’t use any of the user information in our work, we were mostly just interested in what were the conversations going on there? This is just some idea of the volume. The volume was absolutely massive, so we had to sample over it. And then what did we do? We downloaded the Parler set, we discovered the actants and their relationships. You can’t really see, it’s a cool picture. I want to make it as a poster. Is there a way to turn off the lights? Anyway, not that important. That’s a cool picture.

Audience 3: [Inaudible]

Timothy Tangherlini: Yeah. Yeah, playing with them randomly is cool. It gives a little light show. You guys get it. It’s a little bit better. We consolidate some of these, we do some magic with computers, and then we generate a series of narrative framework graphs. We’re trying to understand what the story was and how it evolved over the time from Nov. 10 to Jan. 10. And so we wanted to take this graph apart, see what was at the basis of it, and then also try and understand what people were talking about. We used some AI stuff here called Vert Topic, and we generated labels on the various communities.

We found some main actants, not surprisingly, Biden, Trump, then some slightly more surprising people, Sidney Powell, Scott Fitzgerald, Lynn Wood. Hunter Biden was already there. And if we go over here, we see China, Republicans, God, President Trump, and Obama. You can see already from this estimate what people are talking about on Parler.

When we go down to what is generating the edges and we label the groupings of sentences related to these edges, we see a huge amount discussion of Antifa and Black Lives Matters, treason, traitor. You can see all of these groupings here; those are based on some similarity in what we call a word embedding space. The assignation of Democrats to treason and to outsiders is unequivocal. And then there’s another whole group of discussions linked to things like Islam and Muslims, abortion, pedophile, and weirdly also … Well, not weirdly, actually, the Catholic Church here. You can see these groupings of the different narratives that are characterizing this Parler domain.

The topics over time march up and down intact with political events. And if we look at the very beginning of this, we can already see that the discursive space has put Democrats as satanic and as leftists and as socialists. You’ve already constructed a threat narrative from the very beginning. Everything that happens afterward is based on the Trump group and, quote, “patriots” as the insiders. And what are we going to do to beat back the threat that’s constructed narratively of these other groups?

And when Flynn gets this pardon, this is when there’s an enormous amount of interest in then taking it to the streets and really going to war literally with the democratic threat. It’s not subtle. There’s nothing subtle in any of these graphs. What’s interesting is that in these millions and millions of posts, we’re able to find these different signals and we’re already able to see early on that there is a convergence on a strategy of forceful and violent response to what has largely been then framed as a theft of the election.

And so here’s probably when Clark writes his letter to Georgia, you really see this very clear condensation of that narrative of the Democrats stealing an election from Trump, and that it really is an opportunity for patriots to come to the savior salvation of the American people. It’s not terribly nuanced, but a lot of these stories aren’t nuanced. You don’t necessarily need nuance to get people to take real world action.

We go through this and ultimately find that on Jan. 6, there is an insurrection we’ve already seen, starting already on Nov. 14, organizing about something to happen right around the beginning of January narratively. What’s interesting is in the aftermath of Jan. 6, how quickly the narrative flips so that it’s not something that we’re doing as a strategy to right a wrong, but rather that Jan. 6 was largely a false flag operation with Antifa in cahoots with the Capitol Police. It just pops out of these narratives graphs. There’s no secondary analysis going on here, these are just the nodes and the relationships. And we see that in the aftermath, there’s then a very strong conversation narrative that tries to correct what actually happens. It’s a revisionist history.

Conversations on Parler were ideologically consistent throughout. There’s very little room for dissent. It was basically an echo platform. New events that came in were fit into the existing belief framework, and the existing belief framework drove real world actions. There was a rapid escalation to organizing for violent intervention early on, and Democrats were seen as an existential threat from the start. They brought in fears of technology, communism, socialism, and race. And in all of these graphs we see that Trump and the patriots are aligned with God. The actions were seen as patriotic, supported by Trump, God, and violence was justified and necessary, and a very strong, what’s called Christian nationalist movement slant to a lot of these conversations.

What else are we doing? We’re trying to understand these narrative groupings at this very large scale. You have to understand that these things can flip flop very easily. Semantically, it’s very easy to change the story. I think that tech will kill me. Tech, we’re trying to figure out what is an insider, what is an outsider? And so we’ve done some work trying to figure out parsing at the level of sentences what the person, the message that is being sent. And when we do this over the COVID-19 data set, we find an insider group here where faith, family, and American is super insider, and billionaires, the rushed vaccine and politics are the real outsiders. Weirdly, Mussolini is oddly close to the middle. We don’t know what’s going on with that. We have no idea what he’s doing there. We actually don’t know what he’s doing there, but Mussolini was very prominent in the COVID-19 discussions.

And then one of the questions is, while a friend of a friend is a friend and an enemy of an enemy is a friend, what is a friend of an enemy? This incomplete information and act of alignment in narrative networks is what we’re working on as well right now. In the process of storytelling, we try to categorize groups or groups of people. And so part of that negotiation of the storytelling is trying to figure out where they fit in the scale of insider to outsider. And can we do this with just free-form text? And so we were playing around with different discussions, in this case about gun control, and trying to understand the differences between groups not so much based on overt statements, but on implicit statements in this work.

And our final work is actually trying to work on graph summary. Because you saw some of these graphs were so big that it’s hard to really get a sense of what’s going on in those graphs, so we want to take apart the graphs and understand the semantics, not so much just of the topology of the graph, but the semantics of the stories being passed. We’ve done this both for Parler and have some great visualizations, but the math is perhaps beyond what we want to go into right now. It’s on a Friday afternoon, after all. But I’ll give you the online quiz next week and you’ll actually have to solve the equations. I apologize.

But we also did it for French Twitter. And what was interesting with French Twitter is we found we could find these same types of insider, outsider, but the labels on what was considered insider, outsider, what the main threats were were significantly different from those in the America. That’s confirmatory that our method is working, but it’s also not terribly surprising.

Positive directions. Are there anything good that can come of this? And that’s something I think is very important as a humanities scholar and just as a human being. I already know that we can use it for bad. It’s very easy. We can take our models, give the model to ChatGPT after sampling over the graph and create valid posts, attach those to some robots, and we might actually be able to hasten the end of the world. Certainly use it for bad.

A great deal of attention in the study of misinformation and disinformation has been trying to attack the misinformation itself. And the problem is that you get minor effects, but they don’t persist.

My curiosity is can we use the structure of the storytelling that I’ve just belabored now for almost an hour to change the conversation? Can we question exclusionary ideas about who belongs and turn them into more inclusive ideas in the storytelling itself? Move towards telling stories that are inclusive as opposed to exclusive. Can we question ideas of what is threatening? Is it the vaccine preventable diseases or is it the vaccines? I think the vaccine preventable diseases are probably more threatening. Is there a way that I can change the discussion of threat and disruption? Can I change the strategies? Can I develop ways to steer conversations to more inclusive and less destructive strategies? If people are eating the dogs and cats, they must be hungry, so what can I do to alleviate hunger in those communities? There’s different ways of approaching this. It doesn’t have to be just, oh, that’s disinformation. You can actually, in positive ways, nudge the storytelling in different ways, and then maybe we can change the results so that it isn’t, “That was a false flag operation.” You can reshape the narrative so that one gets to an outcome with a net social good for everyone.

Thank you so much for listening. Yes, I think we have time to entertain some questions. I’m not sure if I’ll be able to answer them, but I’ll certainly try to entertain them. You had your hand up first. Yes.

Audience 4: I understand all about the mythology and storytelling, but not once did you mention lies and how you deal with the concept of lies being the basis for a story?

Timothy Tangherlini: Right. This goes to beliefs, right?

Audience 4: Right.

Timothy Tangherlini: This is the difference between beliefs and truth. And so in a lot of communities we come up with at times completely contradictory sets of beliefs. And from some standpoint, we might look at those as purely lies.

Now, there are certainly people who are injecting lies into conversations for malign purposes. Didn’t have time to go into that whole branch of the discussion there. One of the questions is to what extent am I … What are my goals of telling those lies? Is it something that I’m doing deliberately? And I think lies suggest deliberate intent, whereas belief is something I may have a belief and it’s a strongly held belief that for somebody else is like, that’s just … I can’t use the word, but BS. Yeah. There’s a tension there, and it is one that might go to the intent of the teller. Yeah. Good point. You had your hand up, then you, and then you.

Audience 5: When you’re assessing this data, are you doing it by date so you can track … you can see a change when it’s like this date is when it looks like things are going to happen on Jan. 6?

Timothy Tangherlini: Yeah, we have some change point detection methods where we can see where either something that there’s a change in the narrative structure that front runs an event that we discover later on happen, so it could be predictive, and then we can also find change points that are subsequent to events. That’s an important consideration. Yeah, we gather as much time-varying data as we possibly can, so everything’s actually down to the minute, or actually probably a second.

Audience 5: And what was the time when you thought things were going to happen?

Timothy Tangherlini: Some of the things front ran events. So Jan. 6 was wildly anticipated. We already see that something was going to happen there. There were other things that were completely reactive. The million MAGA match then led to a spike in conversations about, “We should do that again.” I think were you next? And then the question in the back. Yeah.

Audience 6: Are you able to find evidence of a conspiracy behind all the conspiracy theories?

Timothy Tangherlini: Yeah, that’s a great question, and one that we’re really interested in. We do know that there was a conspiracy, and it was in plain sight on Parler to react en masse and violently on Jan. 6. And that’s a reaction to a conspiracy theory. We were really interested in is there a topological difference in the storytelling that comes out and drips and drabs when a conspiracy is found? Because recall, a conspiracy, their whole goal is to never be found out.

That’s the point of a conspiracy. You have to wait for the narrative to come out. It takes a very long time for you to actually estimate what the underlying narrative is of something like Bridgegate. It took years, whereas a conspiracy theory seems to jump into a fully formed state very, very quickly. QAnon really get much added to it after about six weeks. There’s manipulation, blah, blah, blah. There’s not really much being added to it. Part of that’s because it’s storytelling. You can create the whole universe in story very quickly. It’s very hard to conspire and not get caught. Yeah, that’s actually something that we’re working on. You had a question about here.

Audience 7: Two parts. You answered part of the first one, the predictive model quality. I’ve seen a lot about those metrics [inaudible]. Are you finding it successful as a predictive model similar to trends on Twitter or something? Is that what you’re seeing when you follow something like this? And then the second question is, for those that use this for various activities like North Korea or a government that wants to do something and create conspiracies, do they use these kind of predict or computational models to analyze what would be the best kind of misinformation to spread based on how ripe it is for acceptance?

Timothy Tangherlini: Yeah. Answer to the second question first, yes. That’s the simple answer. But yeah, no, they’re obviously using these kinds of models. But there’s also a brute force element to it. You pump out a lot of stuff and see what sticks. It’s like the spaghetti cooking model where you keep testing to see if it’s going to stick. But they’re getting better.

And so once you’ve estimated what the domain is for a community, it’s very easy to generate posts that fit the pre-existing parameters of the model and then reinforce those using bots. The thing is reinforcement actually makes things stick. The more you see something, the more likely it’s going to stick. The first one is predictive modeling. Can we predict when something is about to happen? We can predict what a range of potential strategies are that people are talking about. We wouldn’t be able to predict whether or not somebody is going to actually do that. Yeah. Yes.

Audience 8: I’ll let this be the last question. You left us hanging. What happened to Bob at the Hotsie Totsie?

Timothy Tangherlini: We’ll have to go out for a beer and we’ll talk about that. He’s doing OK now, though. He’s fine now. Great. Well, we’ve been cut off by our hosts. I think we all have to leave and …

Audience 8: No, we don’t have to leave. We can stay for a few more minutes.

Timothy Tangherlini: There’s a question here and then there’s a question in the back, and we’ll make that one … As we go all the way to the back, we’ll make that the last one because that’s the back wall and there’s nothing beyond it.

Audience 9: You got into this a little bit with the foreign actors, but you mentioned that people shouldn’t be sitting down to have pizza with a, what is it, malignant bot.

Timothy Tangherlini: A malevolent robot. Yeah.

Audience 9: What percentage is the bot versus real people?

Timothy Tangherlini: Oh. It depends on the platform. Shockingly high. There’s a wonderful tool out of Indiana called Bot or Not. And it gives you a bot profile for Twitter accounts. And there’s certain things that bots used to do that were predictable. Unfortunately, the people who control bots, and you could control bots if you’d like, read the paper. And it’s very clear they read the paper because bots stopped behaving that way. They added more stochasticity. I’m pretty sure that this narrative generative framework, this generative narrative framework, has definitely also made its way back into malevolent hands. Would also like to see if it can have some benevolent or positive effects as well. You got to be careful. You don’t want to get involved in social engineering; gets really weird really quickly. But at least now we know some of the mechanisms, at least narrative mechanisms that are driving some of these conversations. Very last question there in the back.

Audience 10: These narratives are still going on today. Where do we go from here?

Timothy Tangherlini: Well, as far as storytelling, that’s something that makes us human. We’re always telling stories. That might be little stories, that might be seemingly insignificant stories, but that’s the way we create community. That’s the way we gauge how people around us are feeling and thinking.

Now, some stories have been weaponized, but this is nothing new. Stories have been weaponized for a very, very long time. And you can get people to think badly about other groups very quickly through stories. You can also hopefully get people to think good about other groups. The inscription on the Statue of Liberty really is one of these wonderful, inclusive gestures, and that becomes part of a grand narrative that the United States is a welcoming country for immigrants and refugees, is the whole premise of economic and social development. It’s also the premise of the public land-grant universities, of which Berkeley’s is a great example. We can tell stories that lift people up. A lot of stories are about fear and threat; and that might just be something physiological. But can we nudge those stories in ways where the fear and threat is dissipated not through some strategy of exclusion or violent action, but really through a strategy of inclusion? Great. Thank you so much for listening.

(Applause)

(Music: “Silver Lanyard” by Blue Dot Sessions)

Anne Brice (outro): You’ve been listening to Berkeley Talks, a UC Berkeley News podcast from the Office of Communications and Public Affairs that features lectures and conversations at UC Berkeley. Follow us wherever you listen to your podcasts. You can find all of our podcast episodes, with transcripts and photos, on UC Berkeley News at news.berkeley.edu/podcasts.

(Music fades out)