Testing artificial intelligence- and machine learning-based systems presents two key challenges. First, the same input can trigger different responses as the system learns and adapts to new conditions. Second, it tends to be difficult to determine exactly what the correct response of the system should be. Such system characteristics make test scenarios difficult to set up and reproduce and can cause us to lose confidence in test results. Yury Makedonov will explain how to test AI/ML-based systems by combining black box and white box testing techniques. His "gray box" testing approach leverages information obtained from directly accessing the AI’s internal system state. Yury will demonstrate the approach in the context of testing a simplified ML system, then discuss test data challenges for AI using pattern recognition as an example and share how data-handling techniques can be applied to testing AI.