We are happy to continue our series on SUMMETIX being used in scientific work. SUMMETIX is deeply committed to supporting cutting-edge AI and LLM research, and we are also actively contributing to the research community (big news ahead!).
This week, as part of the 28th European Conference on Artificial Intelligence (ECAI), our long-time collaboration partners from the Dialogue Systems Group at Ulm University, together with Tensor AI Solutions GmbH and the University of Bamberg, are presenting a contribution to the 3rd International Workshop on Argumentation for eXplainable AI (ArgXAI).
The paper is named “Towards a Deeper Understanding: Effects of Domain Knowledge Integration for Conversational XAI” and explores how integrating domain knowledge (DK) into explainable AI (XAI) can improve human understanding and model transparency in AI systems. To that end, the authors integrate structured domain knowledge — partially with the help of the SUMMETIX Argument Search API — into interactive dialogue systems.
The findings show that DK can increase user engagement (longer dialogues) and, in some domain-specific conditions, improve perceived plausibility and agreement with the AI’s decisions — but this effect depends highly on the domain and the underlying knowledge. The authors conclude that while integrating DK is promising for conversational XAI, designers must ensure explanation faithfulness, handle scenario dependency and user familiarity, and safeguard against undue persuasion.
