Small Data for Big Tasks: A Paradigm Shift Towards AGI
Abstract
In this talk, I will begin by offering a brief overview of the evolution of artificial intelligence, highlighting its various peaks and valleys, and the current state of the field. I will argue for a shift in our focus from "Big Data for Small Tasks" to "Small Data for Big Tasks" in order to advance toward our ultimate goal of achieving human-level general intelligence. Following this, I will delve into our academic quest for a unifying theory of artificial intelligence, emphasizing the importance of incorporating social and physical commonsense into our research. Additionally, I will introduce a framework for analyzing the challenges of achieving Artificial General Intelligence (AGI). This framework includes descriptive criteria of AGI and introduces two key components: a Comprehensive Cognitive Architecture and a Dynamic Environment with Physical and Social Interactions (DEPSI). Finally, I will outline the critical future research directions in the field that could pave the way for the next generation of AI.
Bio
Song-Chun Zhu
Director, Beijing Institute for General Artificial Intelligence (BIGAI)
Chair Professor, Peking University & Tsinghua University
Song-Chun Zhu, Ph.D. (Harvard '96), is a well-known scholar in the fields of artificial intelligence and computer vision. He holds dual chair professorships at Tsinghua and Peking Universities in China, where he also serves as Dean of the Institute for Artificial Intelligence (IAI) and the School of Intelligence Science and Technology (SIST) at Peking University. Founder of the Beijing Institute for General Artificial Intelligence (BIGAI), Zhu was previously a professor at UCLA, where he led the Center for Vision, Cognition, Learning, and Autonomy (VCLA Lab). His scholarly impact includes 350+ peer-reviewed papers and multiple seminal awards, such as the Helmholtz and Marr Prizes. An IEEE Fellow, Zhu has co-chaired CVPR conferences in 2012 and 2019.