Machine learning and prediction without understanding

Big data is the buzzword du jour, permeating from machine learning to hadoop powered distributed computing, from giant scientific projects to individual social science studies, and from careful statistics to the witchcraft of web-analytics. As we are overcome by petabytes of data and as more of it becomes public, it is tempting for a would-be theorist to simply run machine learning and big-data algorithms on these data sets and take the computer’s conclusions as understanding. I think this has the danger of overshadowing more traditional approaches to theory and the feedback between theory and experiment.
Read more of this post