The tales people hear in their community guide their economic behaviour. – Interview with dr. Tim Tangherlini

Dr Tim Tangherlini is a professor of folklore in the Department of Scandinavian and the School of Information at the University of California, Berkeley, where he also serves as Associate Director of the Berkeley Institute for Data Science. Trained as a folklorist and ethnographer, his research focuses on how stories circulate within and across social networks, and how individuals use them in negotiating ideologies within their social groups.

He visited Ljubljana as part of the workshop AI Methods for Research of Folkloristic Narratives. In this short interview, we talked about Danish folklore, witches and milk, strategies for tackling online misinformation, and the use of computational methods in studying K-pop dance.

What is the main topic or question you want to answer with your research?

My research is essentially concerned with cultural analytics, which is just a data-driven way of looking at cultural phenomena. As I am a folklorist, I am particularly interested in vernacular culture, from storytelling in 19th century Denmark to rumours and conspiracy theories in today’s social media.

In what ways does artificial intelligence come into play in your research?

I use large language models to try to estimate latent semantic aspects in the corpus of stories or posts on social media. We try to understand patterns in the often very noisy data and we use LLMs to uncover these patterns.

One of the patterns in cultural data you were able to recognise with AI has something to do with witches and milk. How did that come up in your research?

We used the standard statistical methods to measure the correlation, but we also found the story. In the data corpus of Danish folktales we worked with, many stories were not classified as stories about milk and witches, even though there are many stories about witches stealing milk or threatening the milk supply. We used a large language model to group together clusters of stories on this theme, and then we linked these stories to the people who told them.

We have discovered that these stories are linked to the emergence of dairy farming cooperatives in Denmark in the 1880s. Once cooperatives made milk production more stable and shortages disappeared, the stories about witches stealing milk also became less common in Danish folklore. What is really interesting is how people’s economic decisions were influenced by the stories that circulated in their communities. To the extent that these decisions reduced the worries reflected in the stories, the stories themselves lost their purpose and gradually disappeared. The tales people hear in their community guide their economic behaviour.

Could you elaborate on your current research on K-pop and dance in social media?

Another way I use AI in my work is in collaboration with a colleague with whom we are developing a search engine for dance-based K-pop videos on TikTok. Basically, we’re using AI models for pose detection to see where the dance style and certain movements come from, and that helps us understand the dance better.

Have you discovered anything interesting so far?

We are now able to identify where dancers repeat movements in a dance, and we can begin to develop a dance vocabulary for an individual dancer or a group. Therefore, we can characterise different K-pop groups based on their dance movements.

The other thing we’ve been looking at is dance group challenges on TikTok. What are the differences between very creative dancers who dance to snippets of a BTS song and those who are not so creative? Some dances are good, some are bad. What are these differences? What is creative and what is bad in the context of TikTok dance competitions?

You also study conspiracy theories through the lens of folklore and storytelling. 

Conspiracy theories are stories presented as truth. They form part of our collective storytelling, and on social media, their spread can be seen as a kind of social natural experiment. Before the rise of social media, these theories were mostly confined to everyday conversations and difficult to track. Today, however, social media makes it possible to collect and study them systematically.

For example, Pizzagate did a very good job of archiving the conversations, they did the data collection for us and we got a very good ready-made corpora. Our research question was: Can we determine the boundaries of a conspiracy theory based on posts on social media? Who are the main actors, what are the relationships between them, and who is part of the group or not? Can we recognise structural or topological features? To investigate this, we created a network of actors and their interactions and used it to construct a narrative network for the conspiracy theories.

We have found a difference between a conspiracy theory and a conspiracy that is part of the real world. With a conspiracy theory, the aim is to spread the story as widely as possible, you want others to know about it. In a real conspiracy, on the other hand, the people involved try to keep the story secret; they don’t want it to be uncovered.

Can you point to any particular patterns or cases you have observed?

We have dealt with two events in the USA that took place at almost the same time. The first was the Pizzagate conspiracy theory that Democrats were using underground tunnels to transport children so they could cannibalise them as part of a satanic cult.

Then there was an actual conspiracy to close a bridge between New York and New Jersey, which caused a five-day traffic jam. That was an actual conspiracy, and it took five years for that story to actually come out. Pizzagate, on the other hand, came to light almost immediately. We looked at the narrative structures and realised that there are indeed differences between a real conspiracy and a conspiracy theory.

You are also proposing some measures to stop conspiracy theories?

Well, we can recognise them. People often want to attack misinformation at the level of threat or disruption. People see vaccines as a threat, and that makes them reluctant to get vaccinated, and you try to attack them by saying vaccines are not a threat. That works in some cases, but you have to come up with strategies that resonate with the people who believe these stories. But you also have to shift the conversation to change the idea of inside and outside. We can use the structure of storytelling to create inclusive narratives instead of threat and exclusivity narratives.

Why do you think social media provides such fertile ground for the spread of conspiracy theories?

This is related to what I call social brakes. Just 20 years ago, most stories were told face to face. If you thought the story I was telling you was somehow strange, I could see in your face. There was a social brake in storytelling. There was also a time lag, I could reach as many as five, maybe ten people. With social media, these social brakes have gone away, instantaneous communication has been strengthened.

On social media, the presence of bots makes the situation even more problematic. I like to illustrate this with an example of going out for pizza with friends. In the past, I would go to a bar, find my friends, and enjoy pizza and beer while interacting with people I knew and had built rapport with over time. On social media, however, it’s more like walking into a bar and joining a table of malicious robots without realising they’re not human. This can be dangerous, because these bots can influence behaviour without your knowledge.

What are the current trends in your field of research?

Multimodal models. When we do research on social media, we only look at texts. But now we try to include all other media such as memes, text, sound and video snippets. And then we are trying to take advantage of the power of LLMs to estimate aspects of the narrative that in the past we forced into classification models that were not always reliable.

Do you have any book recommendations for our readers?

There is a very interesting book I have just read, The Report from Iron Mountain by Leonard C. Lewin about a conspiracy theory in the USA. I can also recommend Narrative Economics: How Stories Go Viral and Drive Major Economic Events by Robert J. Schiller and Six Degrees: The Science of a Connected Age by Duncan J. Watts.