Explainable Artificial Intelligence via Bayesian Teaching

Modern machine learning methods are increasingly powerful and opaque. This opaqueness is a concern across a variety of domains in which algorithms are making important decisions that should be scrutable. The explainabilty of machine learning systems is therefore of increasing interest. We propose an explanation-byexamples approach that builds on our recent research in Bayesian teaching in which we aim to select a small subset of the data that would lead the learner to similar conclusions as the entire dataset. We discuss this approach, explicating several key advantages. First, the ability to cover any model with a probabilistic interpretation including supervised, unsupervised, and reinforcement learning (including deep learning). Second, we discuss the empirical foundations of this approach in the cognitive science of learning from other agents. Third, we outline challenges to full realization of the promise of this approach. We conclude by discussing implications for machine learning and applications to real-world problems.