I totally agree that the ML black boxes need to be made comprehensible by us, and this has to be done fast, before the machines become too big and ubiquitous to be understood. There is recently a paper on Science that talks about this issue of Explainable AI: http://www.sciencemag.org/news/2017/07/how-ai-detectives-are-cracking-open-black-box-deep-learning
I recently asked Kai-Fu Lee, a famous VC in China who worked in Google, Microsoft and Apple, about his view on explainable AI. He said it may be of interest as a research topic but not so much in the industrial/commercial side, for the networks are so complicated with their huge number of parameters, and human is too stupid for understanding them.
Anyway, within this year I hope to build a startup that helps to crack the ML black box.
