We're acting like throwing more layers on a neural network is going to magically solve complex problems, when really we're just sweeping the lack of underlying understanding under the rug.