I2SC Lecture Series

Pitfalls, Challenges, and Benefits of Interdisciplinary Research: Reflections from a Psychologist Collaborating with Computer Scientists for Over a Decade

The Interdisciplinary Institute for Societal Computing at Saarland University is dedicated to fostering research at the intersection of society and technology, which is why “interdisciplinary” is central to its name. In this talk, I would like to share my personal experiences as a psychologist who has worked collaboratively with computer scientists for over 13 years. I began this journey somewhat naively and have since undergone a long learning process. This experience has convinced me that anyone interested in interdisciplinary work should periodically reflect on the pitfalls, challenges, and benefits of such collaboration. I hope my lecture provides one of those opportunities for reflection. 

Language Understanding as a Constraint on Consensus Size in LLM Societies

The applications of Large Language Models (LLMs) are going towards collaborative tasks where several agents interact with each other like in an LLM society. In such a setting, large groups of LLMs could reach consensus about arbitrary norms for which there is no information supporting one option over another, regulating their own behavior in a self-organized way. In human societies, the ability to reach consensus without institutions has a limit in the cognitive capacities of humans. To understand if a similar phenomenon characterizes also LLMs, we apply methods from complexity science and principles from behavioral sciences in a new approach of AI anthropology. We find that LLMs are able to reach consensus in groups and that the opinion dynamics of LLMs can be understood with a function parametrized by a majority force coefficient that determines whether consensus is possible. This majority force is stronger for models with higher language understanding capabilities and decreases for larger groups, leading to a critical group size beyond which, for a given LLM, consensus is unfeasible. This critical group size grows exponentially with the language understanding capabilities of models and for the most advanced models, it can reach an order of magnitude beyond the typical size of informal human groups. 

From Stories to Sonnets: Data-Centered NLP for Creative Works

In this talk, I'll share two recent studies that use natural language processing (NLP) techniques to model creative works like stories and poetry. In the first part of the talk, I'll discuss NLP approaches for story detection and analysis, focusing on how NLP methods can help us study storytelling at large scales and across diverse contexts. In the second part, I'll discuss the poetic capabilities of large language models (LLMs), focusing on audits of the vast pretraining datasets used to build these models. Both studies will highlight the challenges in creating open evaluation datasets for creative works and the importance of interdisciplinary collaboration between NLP and the humanities.

Misinformation and Hate speech on WhatsApp - Insights from a large-scale data donation study

In this talk, I will present a study looking at the prevalence of misinformation, political propaganda and hate speech on WhatsApp. Using a large WhatsApp data donation program in India and Brazil covering thousands of users, we systematically analyze private group messages to understand content prevalence, virality, and user profiles in problematic content spread. Our analysis from data in India shows a high prevalence of political content, with significant misinformation and hate speech against Muslims. This misinformation was notably prevalent in caste-based groups and had been previously debunked by fact-checkers, indicating a failure of fact-checks to reach these groups. This research is the first quantitative analysis of everyday WhatsApp use and highlights challenges with end-to-end encrypted platforms. It provides a baseline for developing moderation policies to combat misinformation and promote responsible use of encrypted communication channels. The study also develops novel data donation methods and tools to collect representative samples from hard to study platforms. These approaches can be scaled to other platforms to enable data collection in the post API age.