Principle: Decisions made by an AI agent should be possible to understand, especially if those decisions have implications for public safety, or result in discriminatory practices.

AI systems should not be trained with data that is biased, inaccurate, incomplete or misleading.