Objective-Driven AI: towards AI systems that can learn, remember, reason, and plan

Keynote AI Technology

08/02/2024 | 10h50 - 11h15 | Tech & Strategy Stage


  • How could machines learn as efficiently 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 Auto-Regressive Large Language Models fall short.  I will propose a modular cognitive architecture that may constitute a path towards answering these questions.  
  • The centerpiece of the architecture is a predictive world model that allows the system to predict the consequences of its actions and to plan a sequence of actions that optimize a set of objectives.  
  • The objectives include guardrails that guarantee the system's controllability and safety.  The world model employs a Hierarchical Joint Embedding Predictive Architecture (H-JEPA) trained with self-supervised learning.  
  • The JEPA learns abstract representations of the percepts that are simultaneously maximally informative and maximally predictable.  
  • The corresponding working paper is available here: https://openreview.net/forum?id=BZ5a1r-kVsf