I thin it's more that we're operating in a position of extreme simplification, and don't understand most biological details well enough to apply them in our own

I thin it's more that we're operating in a position of extreme simplification, and don't understand most biological details well enough to apply them in our own systems. But biological motivation is quite common in neural networks. Name a popular technique (backprop, dropout, etc.) and I can probably find you a source arguing it's biological relationship. It's just at an engineering-style level of approximation that scientists may abhor. The ReLU was originally introduced by computational neuroscientists as a more biologically accurate approximation compared to the other activation functions. E.g. https://www.nature.com/articles/35016072 https://www.nature.com/articles/35016072 I would expect similar history behind convolutional neural networks, given how the early stages of the vision system have been understood to operate.