[samus]

Backprop is just applied calculus. People simply didn't think about using it for neuronal networks yet.

[varjag]

It was thought of as early as in 1960s by Rosenblatt but he did not come up with a practical implementation at the time. Lotsa things look obvious in hindsight.