> Is a neural network really different from a very large cascade of nonlinear filter elements? It's different in theory, but not by much practically. Reverse e

> Is a neural network really different from a very large cascade of nonlinear filter elements? It's different in theory, but not by much practically. Reverse engineering the former is a fool's errand, but so would be a sufficiently complex version of the latter. In terms of Kahneman's "Thinking Fast and Slow," ML, at least in its current stage, is like the fast thinking system. It's essential but is exponentially more valuable when combined with the slow system, which is still elusive in AI practice.