Gary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
Lex Fridman Podcast
A challange to Unsupervised Deep Learning's quality
Deep learning is loved for its ability to accomplish tasks automatically without human intervention. However, it is difficult to program a machine to understand concepts like causality or the workings of a container. Understanding containers alone requires significant effort and logical analysis. While it would be easier to feed a system with videos and have it infer knowledge about containers, building a robust system that can accurately predict outcomes is challenging. Some argue that the focus should be on defining an unsupervised learning algorithm that can learn from videos, but others believe that the hard work lies in devising a design specification and engineering knowledge into the system. Current deep learning systems lack complex knowledge and struggle to naturally connect different concepts.
