Trustworthy AI explanations unveiled

Prof. Barbara Hammer from Bielefeld University will explore a central question in explainable artificial intelligence (XAI): When can we trust an explanation? Her lecture takes place on 11 December 2025, 10:15–11:45 AM, in JvF25/3-303 – Conference Room (Lamarr/RC Trust Dortmund) as part of the AI Colloquium.

Explainable AI aims to extend or replace black-box machine learning systems with components that humans can understand. Examples include prototype-based approaches, counterfactuals, and feature importance measures such as SHAP. Yet despite the rapid growth of XAI methods, it remains unclear what information these explanations provide, why different techniques produce diverging results, and to what extent humans can rely on their output.

Hammer will illustrate how XAI can be applied in critical infrastructure and highlight current challenges, including uniqueness and plausibility. She will present recent work addressing these issues by examining feature-based explanations and their game-theoretic foundations. Interactions between features can significantly influence model interpretation, especially in large language models and multimodal foundation models. New approaches allow these interactions to be quantified, opening promising opportunities for trustworthy machine learning.

Prof. Hammer leads the Machine Learning Group at the CITEC Research Institute. Her research spans trustworthy AI, lifelong learning, and hybrid symbolic–sub-symbolic systems. She has held international visiting positions and is active in major scientific committees and initiatives, including ERC WaterFutures and DFG Constructing Explainability. She is a member of Academia Europaea and serves on the Scientific Directorate of Schloss Dagstuhl.

AI Colloquium
The AI Colloquium is a series of lectures dedicated to cutting-edge research in the field of machine learning and artificial intelligence, coorganized by the Lamarr Institute for Machine Learning and Artificial Intelligence (Lamarr Institute), the Research Center Trustworthy Data Science and Security (RC Trust), and the Center for Data Science & Simulation at TU Dortmund University (DoDas).