AI-based software are of increasing relevance but also add additional challenges for the software testing process, such as the oracle problem of uncertainty in the precise expected outcome of the software or gigantic numbers of potential test scenarios. At the same time, AI techniques, both data-driven machine learning and logic-driven constraint solving, can aid testing techniques to be more cost-effective and focus on finding faults early. In this talk I will discuss a selection of applications at this intersection, involving metamorphic testing, reinforcement learning, and constraint optimization.