In the near future, the transformative potential of machine learning could revolutionise areas such as medicine. This opportunity comes, however, with its own unique challenges, chief among which is the inherent difficulty of taking the workings of complex “black box” machine learning models and making them readily interpretable to a multitude of users. The question developers of machine learning models must grapple with is how they can ensure that their intended users—ranging from clinicians to medical researchers to patients—can trust and understand the recommendations made by their models.
In this Turing Lecture (delivered March 11, 2020), Mihaela van der Schaar introduces a number of cutting edge approaches her research team have developed to turn machine learning’s opaque black boxes into transparent and understandable white boxes.