I2SC Lecture Series

On Gender Gaps in Natural Language Processing 

In the field of natural language processing (NLP), gender gaps are thought of and treated similarly to the gender wage gap in economics. I will present previous work on the Gender Gap Tracker at Simon Fraser University and more recent work done at Saarland University on pronoun use with large language models, to illustrate how gender gaps are technically operationalized in NLP. These examples will serve as motivation to discuss bigger-picture epistemological gaps in how “gender” is treated in NLP - typically as binary, immutable, and directly inferable from characteristics such as a person’s name, pronouns, and so on. Drawing on methodologies, frameworks and viewpoints from other disciplines, including work I was involved in on the use of intersectionality in AI fairness, I will end with my thoughts and hopes for closing the epistemological gaps, with the goal of addressing the technical gaps more effectively and justly. 

Legal Aspects of Artificial Intelligence: Balancing Privacy and Utility

The regulation of artificial intelligence poses new challenges for the law. For example, personal data is subject to data protection law even though the impact of the data processing might appear insignificant in many cases. In particular, this may apply to machine learning applications, which often use personal data of millions of different people.

The European Union is currently changing the legal framework for data processing. How will this framework affect the development of AI systems, the gathering of personal data for training purposes, and the application of trained models? Is the privacy of individuals still guaranteed - and if so, is data protection hindering technical progress?  The lecture will provide an overview of current developments and analyse them from an interdisciplinary perspective.

The Rise of Generative AI: Exploring Trust, Believability, and Social Media Dynamics 

We have entered a new era of artificial intelligence (AI), characterized by the rapid proliferation of Generative AI (GenAI) and its numerous applications. As humans, our interactions with “content” in various forms, including text, images, and videos, have undergone a profound transformation due to the integration of AI technologies. It has become increasingly challenging to discern whether we are communicating with a human or an AI model. Concurrently, the content generated using Large Language Models (LLMs) and AI-synthesized images and videos using Diffusion models have significantly altered our perspectives in the real world and on social media. 

In certain instances, we encounter AI-generated fake profiles during job screening or marketing. In other cases, we are confronted with AI-generated fake videos, such as deepfakes, on social media, which purport to depict actions that never actually occurred. Notably, in an era of clickbait-incentivized social media platforms, deepfakes and other forms of appealing AI-generated content significantly increase human engagement, leading to various potential outcomes, both positive and negative. 

It remains uncertain how we can establish trust in AI or how society will be impacted on a large scale by AI-generated content. Several questions arise: How much will share, incorporate, and interact with content created by AI? Will people “trust” the outcomes generated by AI? Will people believe what the majority share on social media? 

In this presentation, I intend to engage in critical discussions on the topic of trust in AI, present empirical research findings on the believability and shareability of synthetic content, and explore the broader societal implications of GenAI applications and interactions. 

Unraveling Romantic Communication. Correspondence networks of Early Romanticism

The Early Romanticism in Jena and Berlin is considered the outstanding intellectual revolution of young German authors and scholars at the turn of the epoch around 1800. The group operated publicly and dispersively, yet network-forming; they reflected and practised "Geselligkeit," for example, through the communication form of letters. For the analysis of these epistolary communication processes among these authors, scholars, intelligentsia, and the people connected to them, the project aims to build up a database that consists of the letters exchanged among the key protagonists of Early Romanticism (such as, among others, Friedrich, Dorothea, August Wilhelm, and Caroline Schlegel, Novalis) and their correspondents between 1790 and 1802, covering various 'prehistories' up to the dissolution of the Jena Circle.

Using methods from knowledge management and network research, the project draws on additional data sources, e.g. the Integrated Authority File (GND) of the German National Library, to work towards a holistic picture of the relationships between those actors, the flow of knowledge and influences, informal and formal collaboration processes between them. By analysing communication networks, we aim to highlight important actors, identify groups and their formation processes, search for signs of loss or (re-)establishment of contact as well as set all these factors in a temporal and comparative context. Ultimately, through these results we intend to get a better understanding of knowledge transfer and production in the Romantic circle.

Theory vs. Computation? On building many Bridges over the Gap

In those areas of Digital Humanities where computational methods are integral to analysis, it becomes essential to critically examine the interplay between humanities theories on the phenomena of interest and their computational analysis. While concepts such as operationalization and modeling have been under discussion for some time, there's still a notable absence of a comprehensive perspective on the entire research process which also includes humanities’ theories. I will suggest to view computational research processes in Digital Humanities as input-process-output workflows in which theoretical concepts serve as integral components at various stages. Drawing on computational literary studies as a case study, my presentation aims to elucidate the connection between theory and computation by proposing a methodology for enabling theory-driven validity.