Interdisciplinary Perspectives on Explanations in AI

This course introduces doctoral researchers to the theoretical foundations, methodological approaches, and practical applications of Explainable Artificial Intelligence (XAI) from an interdisciplinary perspective. It examines how explanations are generated, communicated, interpreted, and evaluated across different domains, combining insights from computer science, human sciences, and health sciences

The course explores explainability techniques for AI systems, human-centered explanation design, cognitive and behavioral aspects of explanation, trust and transparency in AI, and the evaluation of explainable systems. Particular attention is given to the relationship between explanation quality, user understanding, decision-making, fairness, accountability, and societal impact.
Applications may include recommender systems, digital health, decision-support systems, education, human-computer interaction, scientific discovery, public services, and other domains in which AI systems support or influence human decisions.

Upon completion of the course, the student will be able to:
• explain the fundamental concepts of interpretability, explainability, and transparency in AI systems;
• distinguish between inherently interpretable models and post-hoc explanation for black-box AI models;
• XAI in Knowledge-Based Systems
• explain interdisciplinary perspectives on explanations, including contributions from cognitive science, philosophy, social sciences, and behavioral research;
• assess the role of explanations in supporting decision-making, trust calibration, behavioral change, and human-AI collaboration;
• critically analyse challenges related to fairness, bias, privacy, transparency, accountability, and the ethical deployment of explainable AI systems;
• design and conduct a small-scale research project involving the development, evaluation, or application of explainable AI methods in a selected domain;
• communicate and present research findings on explainability and human-centered AI to both technical and non-technical audiences.

Participants are expected to have a basic understanding of data science, and AI. Familiarity with Python programming and common machine learning frameworks (e.g., PyTorch, TensorFlow) will be helpful for project implementation.

Prof. Alexander Felfernig, Mehrdad Rostami

The learning material consists of a combination of research articles, book chapters, and online resources covering next-generation search, large language models, conversational AI, and query prediction. Selected readings will be provided through Moodle prior to each lecture.

The course will be completed through a combination of lectures, discussions, exercises, and project. Students are expected to:
• actively participate in lectures to understand theoretical foundations and practical applications,
• complete exercises on designing and evaluating a conversational search prototype,
• implement a small project and document the process,
• prepare a short written report summarizing the methodology, results, and reflections,
• present their project outcomes in class and provide peer feedback.

The course will be delivered contact lectures, and guided group work. The learning environment (e.g., Moodle) will be used for sharing materials, submitting assignments, and facilitating discussion.
• Guided teaching: ~15 hours of lectures.
• Independent/self-study: ~20–30 hours, including readings, project implementation, report writing, and preparation for presentations.

Teaching methods and learning activities include:
• Lectures introducing theoretical foundations and practical applications,
• Small group work for designing and evaluating prototypes,
• Individual project implementation and report writing,
• Online learning materials and resources provided via Moodle,
• In-class presentations and peer feedback sessions.

The 3 ECTS course consists of lectures, project work with report and presentation, all of which are mandatory.