Several research questions are triggered by this questioning: 1. How to design more transparent and more explainable Agents, LLMs and Machine Learning models that maintain high performance? 2. How to allow humans to understand and trust AI-based systems, methods and digital angents and assistants? 3. How to evaluate the overall transparency, explainability, ethics and social impact of AI systems and models? Current XAI evaluation practices are fragmented, often limited to technical proxies such as fidelity and sparsity, and rarely incorporate human‑centered measures like comprehension, trust calibration, contestability, or long‑term societal effects. The workshop will promote discussions on multi‑dimensional evaluation frameworks, benchmarks and protocols that combine quantitative and qualitative methods, link explanation quality to downstream decision quality, and integrate ethical and legal perspectives, including fairness, accountability, privacy, safety and environmental impact. Overall, the workshop is motivated by the need to move from ad‑hoc explanation techniques toward principled, human‑centered and value‑aware design and assessment of explainable AI systems. By focusing on (1) designing transparent and performant models and agents, (2) enabling human understanding and appropriate trust, and (3) rigorously evaluating transparency, explainability, ethics and social impact, the event aims to chart concrete research directions and foster cross‑disciplinary collaborations that can make next‑generation AI not only powerful, but also understandable, contestable and socially responsible.
Explainable AI (XAI) has become a central requirement for deploying modern AI systems, especially agents, LLMs and high‑capacity machine learning models, in safety‑critical, socially sensitive and decision‑support scenarios. Despite their impressive performance, these models are often opaque, making it difficult for users, stakeholders and regulators to understand how outputs are produced, to contest decisions, or to assess potential harms and biases. This opacity undermines trust, slows down real‑world adoption, and creates friction with emerging AI governance frameworks that explicitly call for transparency, accountability, and human oversight. The proposed Workshop on Explainable AI aims to bring together researchers and practitioners working at the intersection of AI modeling, human–computer interaction, cognitive science, ethics and policy to rethink how intelligent systems are designed, evaluated and governed. A first core motivation is to explore methods for designing more transparent and more explainable agents, LLMs and machine learning models without sacrificing performance. This includes approaches such as intrinsically interpretable architectures, post‑hoc explanation methods, modular and neuro‑symbolic systems, and interaction patterns that expose model reasoning in a way that remains faithful and robust while meeting real‑world scalability and accuracy requirements. A second key motivation is to understand how to enable humans, end users, domain experts, operators, and affected communities, to actually understand and trust AI‑based systems, methods, digital agents and assistants. Trust here is not blind acceptance but calibrated reliance grounded in meaningful explanations, uncertainty displays, user‑centered interaction design, and alignment with users’ mental models, values and goals. The workshop will therefore solicit contributions on human‑facing explanation interfaces, conversational and interactive explanations, user studies and cognitive aspects of explanation, as well as domain‑specific practices in areas such as healthcare, law, public administration, education and recommendation.