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.