TALK TITLE: The Fragile Side of AI: Understanding Hidden Pitfalls, Systemic Failures, and Their Implications
Abstract
Artificial Intelligence is rapidly reshaping critical domains such as healthcare, autonomous transportation, law, and public safety. Yet alongside remarkable progress, AI systems increasingly reveal diverse failures with significant technical, economic, and societal repercussions. This talk presents a comprehensive taxonomy of AI errors, encompassing algorithmic and perception failures, such as hallucinations where language models confidently generate false or misleading information and deep-rooted biases in data and algorithms that reinforce existing inequalities, particularly in areas like facial recognition and automated credit assessments. A central focus is on adversarial examples, small, often imperceptible input perturbations that can dramatically alter a model’s output. These subtle manipulations expose structural vulnerabilities, as seen when minor changes to a stop sign’s texture cause an autonomous system to misinterpret it as a speed limit sign, posing clear safety risks. Beyond technical flaws, we delve into project-level and organizational shortcomings, including poorly defined objectives, low-quality or biased datasets, lack of rigorous validation under real-world conditions, and deployments rushed by commercial or competitive pressures. Drawing on diverse case studies and recent empirical research, this talk proposes an integrated framework to analyze these failures by their technical or social nature, predictability, mitigability, and the degree of human involvement. It also highlights why adversarial vulnerabilities uniquely underscore the absence of true semantic comprehension in current AI architectures. We conclude by outlining practical strategies and future research directions to enhance AI’s robustness, explainability, and ethical alignment from the earliest stages of design and deployment.
BIOGRAPHY
Dr. Gloria Bueno holds a BSc in Physics (1993) and a PhD in Computer Vision (1998), awarded by Coventry University (UK), where she specialized in advanced methodologies for image understanding. Currently, she is a Full Professor at the School of Industrial Engineering of the University of Castilla-La Mancha (UCLM), where she has been based since 2003. At UCLM, she leads the VISILAB research group, a team dedicated to exploring cutting-edge problems in Computer Vision and Artificial Intelligence. Her academic trajectory includes a Marie Curie postdoctoral fellowship at Université Louis Pasteur (Strasbourg, France), and significant research stays at the Institut Physique Biologique (France) and Centro de Estudios e Investigaciones Técnicas de Gipuzkoa (Spain), where she applied AI-driven approaches to both biomedical diagnostics and industrial inspection systems. Throughout her career, she has served as Principal Investigator on 24 publicly funded research projects, five of them at the European level. Prof. Bueno actively collaborates with globally renowned institutions, including CERN, CEA, CNRS, CSIC, and INRIA, fostering international partnerships that have fueled joint publications and shared technological developments. Her expertise has enabled technology transfer to companies such as SIEMSA, INDRA, and Leica Biosystems. In 2009, she was a visiting professor at Carnegie Mellon University (USA), engaging with leading experts in robotics and computational imaging. She is member of IEEE, SEIB and SPIE, a promoter of the ESDIP scientific society, and a founding partner of the UBOTICA Ltd. Her work stands
at the intersection of fundamental research and real-world impact, bridging the gap between academic innovation and industrial application.