One side of the situation is "predictive analysis", which ignores all warts in the data in fealty to R2. The notion espoused by the predictivists is, all that matters is correlation. Such fealty to old data predicting future data was what allowed the Great Recession to happen.
Now comes a new post on the book's site, and a quote I have to get behind:
A specialist in high content screening might naturally take the ratio of these two features of cells because it makes good scientific sense (I am not that person). In the context of the problem, their intuition should drive the feature engineering process.
In terms of the Great Recession, one might have said
A banker in housing markets might naturally take the ratio of house price to income because it makes good micro-economic sense (danger to the firm) as well as macro-economic sense (danger to the whole economy), and conclude that corruption was afoot.
In the case of cell science, there's God's Laws to obey. In the case of the Great Recession, they're fungible Man's Laws. In the former case, getting wrong will be punished, in the end. In the latter case, not so much.