The black-box metaphor that is used in data science refers to the fact that in machine learning one can only see the input and output of an algorithm without knowing how it accomplishes it. A black-box data scientist is responsible for just watching the input and output of an algorithm without knowing it's internal workings.