[im3w1l]

There was a popular conception (like taught in textbooks even) that one hidden layer was enough since a neural network with one hidden layer could approximate any function. Deep learning became a thing when it was found to work great in practice.