I have seen some effort to create reasoning with deep learning. Wonder if it's any good. AFAIK the idea is to bet what is causing changes modify it weight and s

I have seen some effort to create reasoning with deep learning. Wonder if it's any good. AFAIK the idea is to bet what is causing changes modify it weight and see if you get better results. This way network learns it's dependencies explicitly and can kinda "reason". Anyone in the field could confirm?