What makes a machine learning model a black box, and why can its decisions be difficult to understand? How do explainable AI techniques help developers interpret model predictions? What challenges arise when balancing model accuracy, transparency, and user trust in AI applications?