By now, everybody should know that Artificial Intelligence is about to produce a dramatic impact on many sectors of human activity. In the last ten years, thanks to the development of machine learning in "deep networks", we have experienced spectacular breakthroughs in diverse applications such as automatic interpretation of images, speech recognition, Go and chess playing. Most importantly, algorithms are now competing with the best professionals at analyzing skin cancer symptoms or detecting specific anomalies in radiology; and much more is to come. Worrisome perspectives are frequently raised, from massive job destruction to autonomous decisionmaking "warrior" robots. In this talk, we shall open the black box of deep networks and explore how they are programmed to learn from data by themselves. This will allow us to understand their limits, to question whether their achievements have anything to do with "intelligence", and to reflect on the foundations of scientific intelligence.

Marc Mézard is a theoretical physicist. He received a PhD from Ecole Normale Supérieure in Paris, did a postdoc in Rome, and became the head of the statistical physics group in ParisSud University. He has been the director of École normale supérieure since 2012. His main field of research is the statistical physics of disordered systems and its use in various branches of science biology, economics and finance, information theory, computer science, statistics, and signal processing. In recent years his research has focused on information processing in neural networks. He has received the Lars Onsager prize from the American Physical Society, the HumboldtGayLussac prize, the silver medal of CNRS and the Ampere prize of the French Academy of Science. He is a member of the European Academy of Science.

Public Lecture Series sponsored by Friends of KITP
January 30, 2019

__________________________________

Learn more at: https://www.kitp.ucsb.edu/

Follow @KITP_UCSB for updates on Twitter:   / kitp_ucsb  

__________________________________

The position of the KITP is that ownership and copyright of all online material slides, text, audio, video, and podcasts belongs to the author. KITP is providing dissemination for these materials but does not claim ownership. Any person citing these materials for scholarly purposes should provide an appropriate scholarly reference.