Keynote

Most complex problems of social relevance—such as climate change mitigation, taxation policy design, and traffic management—are inherently multi-agent and multi-objective, involving diverse stakeholders with frequently conflicting goals. Building human-aligned agents in these domains requires a framework capable of navigating the inevitable friction between disparate objectives. While Reinforcement Learning has become a pivotal tool for designing solutions in these critical areas, traditional RL often falls short by collapsing complex trade-offs into a single scalar reward. In this talk, I discuss how Multi-Objective Reinforcement Learning (MORL) offers a more robust and adaptable alternative by explicitly modeling the multi-dimensional nature of feedback signals. I will present MORL as a foundational framework for conflict-aware systems, showcasing how it can foster key principles like explainability, transparency, and trust. Finally, I will explore how this multi-objective approach provides a flexible mechanism for humans and agents to co-evolve solutions, turning systemic conflict into collaborative synergy.

Dr. Roxana Rădulescu is an Assistant Professor in AI and Data Science at the Department of Information and Computing Sciences at Utrecht University. Her research focuses on reinforcement learning and multi-agent systems, with particular emphasis on multi-objective decision-making and multi-objective multi-agent reinforcement learning (MOMARL), where autonomous agents must balance multiple, often conflicting objectives. She received her PhD in Computer Science (Artificial Intelligence) from Vrije Universiteit Brussel, where her work developed a utility-based perspective on decision-making in multi-objective multi-agent systems. Prior to joining Utrecht University, she was a postdoctoral researcher at the Artificial Intelligence Lab at Vrije Universiteit Brussel supported by an FWO fellowship. Her work spans reinforcement learning, game theory, and multi-objective optimisation, and has been published in leading venues such as JAAMAS and AAMAS. She is also actively involved in the AI research community through tutorials at major conferences and service roles including organizing committees for AAMAS, IJCAI, and ECAI.