How the machine ‘thinks’: Understanding opacity in machine learning algorithms