"Using diverse training data is what should be done to prevent bias from entering an AI system when training it. Diverse training data means that the data covers a wide range of features and patterns that are relevant for the AI task. Diverse training data can help prevent bias by ensuring that the AI system learns from a balanced and representative sample of the target population or domain. Diverse training data can also help improve the accuracy and generalization of the AI system by capturing more variations and scenarios in the data."