[unimpossible]

Setting aside whether Deep Learning can do everything, would you want to do everything with deep learning? We don't use Quantum Physics for mechanical engineering, we use Newtonian Physics because it is at the right abstraction level for these purposes. Similarly, there are other abstractions possibly to talk about causal reasoning or consciousness (he did say everything!)

Also, he mischaracterizes Symbol Systems. A symbolic system has two important properties - Composability (you can combine parts to make new things just like words are combined to make sentences) and Distal Access (a part can stand for something else. For e.g., a word like Justice stands for a complex concept. Very similar to naming something). Nearly 40 years ago, Newell also talked about the various levels of an AI system - Implementation Level, Algorithmic Level, and the functional level. Neural Networks etc are at the implementation level, Deep Learning lies at the Algorithmic level, and symbolic systems are at the functional level (i.e., they describe function not how that function is implemented). A symbol system is not at odds with a neural network or deep learning although there are those who say that neural networks can never implement a symbol system.