This is an interesting read, pretty much in-line with what I had previously speculated about deep learning replacing traditional model-based predictive analytics (e.g., linear regression, SVM). But the author, a director of AI at Tesla, goes much further, and speculates that deep learning will replace most complex logic that we currently write by hand (so-called Software 1.0). Obviously, someone still needs to write the foundations of deep learning with Software 1.0.
The author says:
…we’ll be left with a choice of using a 90% accurate model we understand, or 99% accurate model we don’t.
I am not sure deep learning has achieved the 99% accuracy he is claiming; I certainly have not heard of such high accuracy even in the (now commoditized) image-recognition tasks. Nevertheless, there is probably nothing fundamentally flawed in deep learning that makes it impossible to achieve such high accuracy, given enough data and better network configurations.