Neural Networks: Finance Prediction

With machine learning on the rise it seems there are two camps in finance. The first camp thinks machine learning will fix all of our problems and automate our jobs away and the second camp thinks machine learning is too far off for finance to worry about. I think the reality lies somewhere in the middle. Yes, machine learning can fix a variety of finance problems or do them better than current methods, BUT it's not going to fix everything and surpass all current methods. We're at a point where people are experimenting with machine learning in finance and we need to figure out where it can help our industry.

Neural networks are just one type/family of machine learning models. I see a lot of people stating it can only be used for image recognition as they tout how great and wonderful random forest and gradient boosting is for finance. Don't get me wrong there has been success in the area of CART models however finance needs to think outside of the box. I have no evidence of if or how neural networks will add value to finance but here is my prediction. Neural networks are great at image recognition which we all know by now. Building financial models such as a probability of default (PD) model has many inputs/variables. (Yes, I know neural networks can't be used in pricing due to transparency but they could be used in servicing.) All of these inputs can be thought of as a pixel in a larger picture. The picture itself will make no sense to a human but there should be unique images for similar candidates. In a PD model we are predicting whether someone will default or not. A neural network should be able to recognize the inputs and classify the picture as either high or low risk of default. With this image recognition approach the model would be able to predict a variety of different levels of risk as well.

My latest YouTube video covers this hypothesis a bit more in detail with visuals. Do you think this is a reasonable way to use neural networks in finance?