Interactional Fairness in LLM Multi-Agent Systems: An Evaluation Framework
Adapts measures of interpersonal respect and quality of explanations from organizational psychology to evaluate conversations between agents.
AI bias & safety
Senior Researcher and Deputy Lead of DFKI’s Efficient and Explainable AI (E&E) group
I work on bias and safety in AI. I’m especially interested in how bias and unsafe behaviour emerge when AI agents interact with one another or with people.

I study how communication style shapes agents’ decisions (AIES 2025, oral presentation) and how causal reasoning can help reconcile fairness, privacy and robustness (ICML 2026, position paper). I combine behavioural experiments with interpretability methods to study these questions. I also study how to attribute contributions in human–AI and AI–AI teams. Our AI as a Research Object (AI-RO) paper shows how recording prompts, configurations and outputs makes AI’s role in scientific writing inspectable (MetaCRiSP at IEEE S&P 2026).
I studied sociocultural history and anthropology at Vilnius University, then earned an MSc with Distinction in Data Science, Technology and Innovation at the University of Edinburgh. I completed my PhD in computer science at École Polytechnique with Catuscia Palamidessi at Inria Saclay, and did my postdoctoral research with Mario Fritz at CISPA. I’m a programme chair for the 2027 European Conference on Algorithmic Fairness in Vilnius and served as diversity and inclusion chair at ECML PKDD 2024.
A Coefficient Giving grant supports my work on AI Deception Dynamics. A DFG International Collaboration grant supports my work on contribution attribution with CSIRO’s Data61 and Macquarie University in Australia. I received a Google Women Techmakers Scholarship (2020–21) and was named one of the 100 Brilliant Women in AI Ethics in 2025. The Artificial Intelligence Association of Lithuania nominated my work on interactional fairness for AI Publication of the Year 2025.
From 7 to 19 December 2026, I’ll visit CSIRO in Australia. We’ll study how to identify what people and AI systems each contribute when they create something together.
About the DFG collaborationI’m now Deputy Lead of DFKI’s Efficient and Explainable AI (E&E) group. Simon Ostermann leads the group within the Multilinguality and Language Technology department.
Meet the E&E groupI’ll serve as a programme chair for the Sixth European Conference on Algorithmic Fairness (ECAF 2027) in Vilnius, 25–27 August 2027.
Meet the organising committeeWhat does alignment training reward when oversight is imperfect? Our new paper examines conditional compliance.
Read the preprintMy lecture on bias and safety in AI is now available from the Probabilistic AI School.
Watch the recordingOur ICML position paper uses causality to explain why trustworthy AI objectives can conflict.
Read the paperOur study of hidden bias in LLM-assisted peer reviews was accepted to ACL Findings.
Read the paperCoefficient Giving is supporting our project on deception in multi-agent systems. I’m leading the project as principal investigator.
Explore the projectI presented my framework for evaluating interactional fairness in multi-agent systems at AIES.
Read the paperAdapts measures of interpersonal respect and quality of explanations from organizational psychology to evaluate conversations between agents.
Uses causal models to explain when fairness, privacy, and robustness place conflicting demands on a prediction system.
Introduces AI as a Research Object (AI-RO) and demonstrates how prompts, model configurations and outputs can document AI’s contribution to scientific writing.
Changes author metadata while holding the manuscript fixed to test what influences an AI-assisted review.