Goal-oriented AI : towards machines capable of learning, reasoning and planning

Abstract

How can machines learn as effectively as humans and animals ? How could machines learn how the world works and acquire common sense ? How could machines learn to reason and plan ?

Current AI architectures, such as large-scale auto-regressive language models, are insufficient. I will propose a modular cognitive architecture that could be a path towards answering these questions. The centerpiece of the architecture is a predictive model of the world that enables the system to predict the consequences of its actions and plan a sequence of actions that optimize a set of goals. The objectives include safeguards that guarantee the system's controllability and safety. The world model uses a Hierarchical Joint Embedding Predictive Architecture (H-JEPA) trained by self-supervised learning. The JEPA architecture learns abstract representations of perceptions that are simultaneously maximal in terms of information and predictability.

Yann LeCun

Yann LeCun is VP & Chief AI Scientist at Meta and Silver Professor at NYU affiliated with the Courant Institute of Mathematical Sciences & the Center for Data Science. He was the founding Director of FAIR and of the NYU Center for Data Science. He received an Engineering Diploma from ESIEE (Paris) and a PhD from Sorbonne Université. After a postdoc in Toronto he joined AT&T Bell Labs in 1988, and AT&T Labs in 1996 as Head of Image Processing Research. He joined NYU as a professor in 2003 and Meta/Facebook in 2013. His interests include AI, machine learning, computer perception, robotics, and computational neuroscience. He is the recipient of the 2018 ACM Turing Award (with Geoffrey Hinton and Yoshua Bengio) for "conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing", a member of the National Academy of Sciences, the National Academy of Engineering, the French Académie des Sciences.