[HUMAN-CENTRIC AI (CHEEM) PROGRAME FRIDAY LUNCH TALKS SERIES] THE VIEW FROM THE 1980S : AN AI ODYSSEY
6 Nov 2020
Recently some AI gurus have begun to call for researchers to not only look at the past 2-5 years of research ideas in AI when conducting research, as is often reflected in the kind of references cited at the end of most papers in top AI journals and
conferences. To build what is possible for tomorrow, we should draw from AI research from 20, 30, or even 40 years ago, that is full of groundbreaking and valuable ideas that can power the next major push toward a promising AI future.
In fact, the root of deep learning is also from 40 years ago, in the 1980s. But there were many more groundbreaking ideas from the 1980s, other than deep learning. This talk will review some of these ideas, including deep learning, particularly centering on the year 1980, and the 1980’s, which could be characterized as the Golden Age of AI. As Song-Chun Zhu, prominent AI researcher of UCLA and now Peking University has articulated, the current singular emphasis on object-centered issues, such as “classification” in the field of computer vision, only touches on the 5% “visible matter” of AI-relevant issues. There is still the 95% “dark matter and dark energy” aspects of AI and computer vision, which concerns task-centered issues that have not been adequately addressed – e.g., given an object, what is it for, how to use it, how to refine it and design new and better objects from it, etc. This culminated in the recent UCLA group’s position paper calling for a “new paradigm shift for AI”.
In this talk, Dr Ho Seng Beng, Deputy Director, Social & Cognitive Computing Department at IHPC will review how the odyssey of AI ideas from the past, particularly all the way from the 1980’s, is relevant to this important new endeavor, which could help propel us to the AI future of 2030’s, 2040’s, and beyond.