FACt Talks

Session chair: Mathias Verbeke

Jerry Spanakis

Title: QUO VADIS (BN)AI(C)? The paper, the product and the law

Abstract: "As AI systems have quickly moved from scientific prototypes to real-world infrastructure, the relationship between research outputs, deployable products and the evolving regulatory landscape becomes increasingly intertwined. In this talk, I will argue that the future of AI research requires intentional alignment across these three dimensions: the paper, the product, and the law. Drawing from recent projects in legal NLP, consumer protection and online enforcement I will reflect on the ultimate test for AI, namely public welfare and the need to move from benchmarks to real-world evaluation."

Gerasimos (Jerry) Spanakis is an assistant professor at the Department of Advanced Computing Sciences (DACS) and at Maastricht Law+Tech Lab (Faculty of Law), at Maastricht University (UM) in the Netherlands. His current work/research lies at the intersection of Social Computing, Natural Language Processing (NLP) and Law, combining fundamental research on Machine Learning, Large Language Models (LLMs), robustness and bias with applied work on socially relevant domains such as legal AI systems, social media transparency and language-centric technologies. His work advances both the scientific understanding of AI systems and their responsible deployment in social, legal and regulatory contexts, building bridges between technical innovation and societal impact.

Antske Fokkens

Title: Well-behaved language models?

Abstract: "This talk will provide a picture of what would be needed to get desirable behavior from GenAI. I will discuss both how AI research can contribute to creating well-behaved systems as well as where other expertise or input from society is needed."

Antske Fokkens' main research interest lies in methodological aspects of research in Computational Linguistics. She is driven by the question of how computational models of language work: what patterns and systems are found in natural language? How can they be modeled computationally? Which computational methods are suitable for modeling or analyzing which phenomena?

Tias Guns

Title: Conversational Combinatorial Optimisation

Abstract: "Combinatorial optimisation is widely used to solve scheduling, sequencing, rostering, routing and other assignment problems. The traditional paradigm is that of model+solve, where an expert user expresses their problem in constraints and a highly efficient solver searches for the optimal solution. Can we make this less static, and more centered around the decision maker? Interactive, explainable, conversational? We'll highlight recent work around LLM-based modeling and explainable constraint solving, developed as part of Prof. Guns' ERC grant CHAT-Opt."

Tias Guns' research focusses on the integration of machine learning and constrained optimisation. The aim is to make constraint solving more human-aware by learning from the daily operational environment and its users.