Reproducability crisis in machine learning systems that inform immediate or future actions

I have by now accumulated a list of articles on a growing debate over naive scientists, especially social scientists, botching their research by using machine learning techniques they don't understand. They can't 'show their work' or provide enough information for others to reproduce their results. This article by Harrison is a short and succinct summary that includes links to other publications, and you should check out the article by Gibney in Nature Magazine.