I would argue that racism, for example, is a feature of machine learning - it's not a bug.