From machine learning predictions of new materials to machine learned interatomic potentials, Artificial Intelligence is transforming the way we do materials science and rapidly accelerating progress. However, AI is not magic it is only a statistical model with limits, sensitivities, and biases. This course equips students with an informed skeptic’s toolkit to evaluate when and where to trust AI predictions and how to probe the assumptions behind them. Starting with glass box interpretable models, students will learn to use feature importance, sensitivity studies, and model ensembles to interrogate models. We will then transition to grey box models and the use of partial dependence plots, loss landscapes, and model weight sensitivity to evaluate robustness and trustworthiness. Finally, we will consider fully black box models for which the students have access to only outcomes and will investigate the use of the previous tools as well as genetic algorithms and other tools to identify decision boundaries and the model input-prediction latent space. By the end of the course, students will be able to confidently question the performance claims of AI tools, evaluate their robustness for materials applications, and recognize both their power and their limitations.