From 11-13 March 2026, Doctoral Candidates Eva Paraschou and Katerina Drakos had the pleasure to attend the Winter School on Ethical, Legal and Societal (ELS) aspects of Artificial Intelligence (AI) and Autonomous Systems (AS) at Umeå University in Sweden. The winter school provided a deep dive into the field of Responsible AI, with an interesting range of theoretical lectures and hands-on workshops. Both Eva and Katerina benefited from the multidisciplinary approach to the topic and specific examples related to the health and mental health domain.

Of particular relevance were the following lectures which also led to a number of engaging discussions after the winter school:

• Lecture 2: “Alignment with What Values?”, Professor of Philosophy Kalle Grill led a discussion that was highly relevant to the DCs’ work. As they reach the stage in their PhDs where they must begin operationalising values, they are reflecting deeply on key ethical considerations, such as the distinctions between values and preferences, long-term versus short-term impacts and the concept of collective value.

   

• Lecture 1: “AI: Responsibility in a Changing World”, Virginia Dignum emphasises that AI ethics is not a simple checklist or fixed solution, but an ongoing process of navigating complex moral, legal and social tradeoffs. She argues that developers must explicitly define how they prioritise and characterise specific values, like fairness, to bridge the gap between their design intentions and users’ interpretations.

   

• Lecture 7: “Explainable Artificial Intelligence”,  Leila Methnani presented a spectrum of model interpretability methods, outlining global, local and cohort-level explanations (such as LIME and counterfactuals). She also highlighted the critical human elements of XAI, stressing the need to design domain-specific concepts with stakeholders, understand user mental models and guard against automation bias.

In addition to both DCs’ active participation in the lectures and workshops, Eva presented her poster “The Bias Spillover Effect in LLMs” in the poster session. This poster was a result of an ongoing research project in which Eva investigates the unintended consequences of unilateral fairness alignment in large language models (LLMs). Specifically, her research explores how mitigating bias for a single sensitive attribute, such as gender, can inadvertently exacerbate disparities across other untargeted attributes like physical appearance, sexual orientation and disability status. By evaluating three LLMs under ambiguous and disambiguous contexts, her work highlights the critical need for multi-attribute, context-aware fairness frameworks rather than narrow, single-attribute alignment. A pre-print of this work-in-progress has been uploaded to arXiv. During the poster session, both DCs also took the opportunity to network with fellow PhD students and discuss their poster presentations.

Overall, it was a very beneficial experience for the alignAI consortium and both DCs returned to Copenhagen with renewed motivation and a wide network of researchers working on similar projects.