Supervisors info:
Stathis Psillos, Professor, National and Kapodistrian University of Athens
Aristotle Tympas, Professor, National and Kapodistrian University of Athens
Manolis Simos, Ph.D., National and Kapodistrian University of Athens
Summary:
This thesis undertakes a comprehensive exploration of Explainable Artificial Intelligence (XAI) by synergizing perspectives from both technical AI research and Science, Technology, and Society - Science and Technology Studies (STS). Anchored in a thorough literature review, we dissect multifaceted narratives emerging from STS literature on XAI, emphasizing the societal, ethical, and philosophical dimensions of explainability. We then navigate technical avenues, diving into the methods and critiques emanating from the AI community itself. At the core of our discourse is the understanding that AI systems are not merely technical entities but intrinsically woven into the fabric of societal structures and politics. We address the pressing need to perceive AI in terms not only of algorithmic transparency but also of its alignment with human cognition, societal norms, and power dynamics. Through this lens, we unravel the challenges of unclear terminology, the absence of universal objectives, and the intricate interplay between transparency, trust, and interpretability in XAI. Real-world case studies, spanning from music recommendation systems to AI in oncology, offer tangible illustrations, bridging theoretical insights with practical scenarios. These narratives act as crucibles to test and validate our interdisciplinary approach, emphasizing the significance of user-centric designs and the politics embedded within AI systems. By synthesizing these analyses, we illuminate a path towards a more integrated, holistic, and informed approach to XAI—one that champions both technical rigor and societal resonance. As we find ourselves in an era where AI continues to reshape our world, this work sets the stage for more responsible, nuanced, and inclusive advancements in the realm of explainability.
Keywords:
Artificial Intelligence, STS, Algorithmic Transparency, Explainability, Interpretability
