Deep learning is definitely the way of the future but it still has some hurdles to overcome. Compared to more traditional methods like Random Forests or Linear Models, there are a LOT of knobs that you need to turn. How many neurons do you need? How many layers should there be? These are just a few of the questions that you need to answer before you can even start building the model. While there are rules of thumb you can use, there are an infinite number of architectures you can create which only amplifies the difficulty. Once someone overcomes these hurdles and creates an easy to use framework, I bet neural networks will be expanded to even more applications that what they are being used for today.