• Wheeler, B., Searching for Productive Causes in Big Data: The Information-Transmission Account. (2015)

The extreme case of model interpretability is when we are trying to establish a mechanistic model, that is, a model that actually captures the phenomena behind the data. Good examples include trying to guess whether two molecules (e.g. drugs, proteins, nucleic acids, etc.) interact in a particular cellular environment or hypothesizing how a particular marketing strategy is having an actual effect on sales. Nothing really beats old-style Bayesian methods informed by expert opinion in this realm; they are our best (if imperfect) way we have to represent and infer causality. Vicarious has some nice recent work illustrating why this more principled approach generalizes better than deep learning in videogame tasks.
— http://hyperparameter.space/blog/when-not-to-use-deep-learning/

Deep Convolutional Neural Networks for Pairwise Causality
What's the relation between hierarchical models, neural networks, graphical models, bayesian networks?