Powered by Indico
v3.2
Abstract: Artificial Intelligence is often described as a revolutionary approach to the understanding and modeling of complex systems. Across disciplines, AI-based methods—particularly Machine Learning and Neural Networks—are increasingly adopted as general-purpose tools for prediction, decision-making, and scientific discovery. This seminar aims to provide a critical and interdisciplinary reflection on such claims by situating modern AI within the broader landscape of computational models and the scientific method. We will introduce the basic principles of Machine Learning, emphasizing its continuity with well-established computational techniques rather than its portrayal as an entirely new paradigm. Fundamental notions from computability theory will be recalled to clarify the intrinsic limits of computation and learning, highlighting that not every well-defined problem can be effectively computed or generalized from data. Neural networks will be discussed both as computational models and as learning algorithms, tracing a conceptual path from early formal neuron models to contemporary large-scale systems such as Large Language Models. Through selected case studies—including the rise and fall of IBM Watson’s AI medical system—the seminar will distinguish between myth and reality, and reflect on the role of Artificial Intelligence as a tool within, rather than a replacement for, the traditional scientific method.
Bio: Gaetano Salina is a senior researcher at the Istituto Nazionale di Fisica Nucleare (INFN), Sezione di Roma Tor Vergata. He earned his Laurea in Physics from Sapienza University of Rome in 1985, with a thesis on the APE project—an early dedicated supercomputer for lattice gauge theories—supervised by Prof. Giorgio Parisi. Across his career at INFN, Salina’s work has blended theoretical modeling with advanced computing, electronics, and data/knowledge-base approaches, spanning topics from lattice gauge theory and parallel computing to complex systems and data-intensive applications.